Know where the claim comes from.

None. Primary regulatory record. Note the structural blind spot: confidential draft S-1 submissions do not appear until public filing, so absence of a filing is not absence of an IPO process.

The single most important source for this sector in the next 12 months: the first public S-1 from OpenAI or Anthropic will be the first audited disclosure of frontier-lab gross margin, compute obligations and customer concentration. Also carries Cerebras and CoreWeave filings.

Coverage: US registrants and foreign private issuers · back to 1994

Access (inherited): api · free

Public domain US government data. SEC requires a declared User-Agent with contact details and rate limiting to 10 requests/second.

Vendor with a direct commercial interest in AI demand appearing durable. The reported financials are audited; the CEO commentary is not evidence.

Data Center revenue is the cleanest single high-frequency proxy for foundation-model compute demand. Watch the China assumption line in guidance: Q3 FY2027 assumes zero China Data Center compute revenue.

Coverage: Global · back to 1999

Access (inherited): rss · free

Press releases are freely redistributable with attribution; NVIDIA trademarks restricted.

Management has an incentive to frame capex as demand-led. Capex, RPO and lease-commitment figures are audited; growth commentary is not.

Poll four separate IR pages plus the 10-Q lease-commitment footnotes. The footnotes matter more than the headline: Moody's estimates ~$660B of hyperscaler lease commitments sit off balance sheet.

Coverage: Global, US-listed · back to 1990s for Microsoft; 2004+ for Alphabet

Access (inherited): public_page · free

Public company disclosures, freely citable. Webcast transcripts may carry redistribution restrictions.

First-party vendor communication. Compute commitments, revenue run-rate disclosures and corporate actions are reliable; capability and safety claims are self-assessed.

The only channel that has disclosed Anthropic run-rate revenue and $1M+ customer counts. Also carries the confidential S-1 announcement and the open-weights position statement.

Coverage: Global · back to 2021

Access (inherited): public_page · free

Company content, citable with attribution; no redistribution of full text.

First-party vendor communication with strong incentives around capability framing. The August 2026 Hugging Face incident report is unusually candid and detailed, and should be treated as primary.

Mixed reliability by content type: incident reports and pricing are high-quality primary data; AGI framing is marketing. Label accordingly.

Coverage: Global · back to 2015

Access (inherited): public_page · free

Company content, citable with attribution.

None. Primary regulatory record.

Became the most consequential regulator in this sector in 2026. Track dockets EL26-67 through EL26-72: RTO/ISO responses to the June 2026 Section 206 show-cause orders determine large-load interconnection rules and cost allocation.

Coverage: United States (six organised markets) · back to the 1990s via eLibrary

Access (inherited): rss · free

US public domain government records.

None commercially. Methodological caveat: EIA has historically under-predicted data-centre load growth, and its aggregates smooth over severe regional variation.

The essential corrective to 'AI power crisis' coverage: national demand growth is 1-3% while ERCOT runs near 10%. Use the regional breakouts, not the national headline.

Coverage: United States, by region and RTO · back to 1997 for STEO

Access (inherited): api · free

US public domain; EIA open data API requires a free registered key.

None. The only free, unsponsored, high-frequency measure of US firm-level AI adoption. Self-reported by firms, and 'AI use' is respondent-defined.

Biweekly microdata release. The single best falsifier of adoption hype: 19.8% national AI use as of May 2026 with no significant change below 20 employees. Watch the sub-20-employee series for the upside scenario trigger.

Coverage: United States, by firm size, sector and state · back to 2022; AI questions added 2023

Access (inherited): bulk_download · free

US public domain government statistics.

Independent, non-vendor, philanthropically funded. The limitation is capacity rather than bias: some datasets go stale, notably the LLM inference price-trends series last updated 2025-03-12.

Benchmarking dashboard covers 390 models across 80 benchmarks and the notable-models database exceeds 3,600 entries; CSV downloads available. The sector's most valuable non-vendor capability dataset.

Coverage: Global · models database back to 1950

Access (inherited): bulk_download · free

Datasets published under open licences with attribution requirements; check per-dataset terms.

Commercial evaluator; methodology is published but the business depends on vendor and enterprise attention. Independent of any single lab.

Versioned index with effort-tiered scoring, which matters now that inference-time compute is a capability variable. Caution: the /models pricing endpoint returned HTTP 429 repeatedly during this research and could not be read — Phase 2 should budget for an API or licensed feed.

Coverage: Global · rolling, versioned indices (currently v4.3)

Access (inherited): public_page · freemium

Site terms restrict bulk scraping; attribution required for chart reuse.

Marketplace telemetry with strong selection bias: its users are price-sensitive individual developers, not enterprise procurement. Systematically over-represents cheap and open-weight models relative to the enterprise market.

The most granular public usage telemetry in the sector and the earliest indicator of Chinese open-weight penetration. Never use it as a market-share denominator; pair it with enterprise survey data.

Coverage: Global · rolling, roughly 12 months

Access (inherited): api · freemium

Rankings data published under Creative Commons licensing per the site.

Trade press with named bylines and an industry advertising base. Strong on capacity, power and earnings detail; light on financial scepticism.

The best single feed for GW-denominated capacity news, neocloud earnings detail, and project withdrawals or moratoria — the last of which has no systematic public dataset anywhere else.

Coverage: Global · back to 2010s

Access (inherited): rss · freemium

Copyright DCD; headline-and-link citation only, no full-text redistribution.

Central bank research; independent of industry. Relies in places on sell-side issuance projections, which are themselves estimates.

Source of the AI-debt duration-supply analysis: ~$300B of 2026 AI-related IG issuance producing ~$360B of 10-year equivalents, roughly one-eighth of Treasury duration supply. Pair with the FOMC minutes for the financial-stability framing.

Coverage: United States · back to the 1990s

Access (inherited): rss · free

US Federal Reserve publications, free to cite with attribution.

None. Primary policy record. Note that FSR language on AI valuations is deliberately hedged.

The FOMC has explicitly flagged high AI-firm equity valuations and leveraged infrastructure financing as financial-stability risks, and names the AI buildout as an inflation driver. The 2026-09-16 decision was unresolved at this research date.

Coverage: United States · back to 1993 for minutes

Access (inherited): rss · free

US public domain.

Official EU position; presents deferrals in the most favourable framing. Cross-check deadline changes against the Official Journal.

Authoritative on GPAI obligations and on which deadlines actually bind. The August 2026 high-risk employment-AI deferral to December 2027 is the best current example of a marketed deadline that did not hold.

Coverage: European Union · back to 2021

Access (inherited): rss · free

EU public sector information, reusable with attribution under Decision 2011/833/EU.

None. Primary legislative record. Introduction is not passage: most AI bills do not advance.

Track the Ban Artificial Superintelligence Act (Sanders/Casar, introduced 2026-09-03) and agent-regulation bills. Use alongside sponsor press offices, which publish before the text appears.

Coverage: United States · back to 1973

Access (inherited): api · free

US public domain; congress.gov API requires a free key.

Hong Kong-based, owned by Alibaba Group — a direct participant in the Chinese model market it covers. Read Alibaba/Qwen coverage with that in mind. Still the most accessible English-language source on Chinese AI labs.

Regional source for this sector. Supplies the Chinese lab picture — Qwen open weights, Moonshot valuation, Zhipu compute constraints, MiniMax token-ranking claims — that Western outlets cover thinly. Phase 2 should add a Chinese-language source; this dossier had none.

Coverage: China, Hong Kong, Asia · back to the 1990s

Access (inherited): rss · freemium

Copyright SCMP; partial paywall; headline-and-link citation only.

Commercial interest in open-model momentum. The Hub download and model-count APIs are raw platform telemetry and are reliable; the blog is editorial.

Model release cadence, download counts and open-weight ecosystem share. Also published the technical timeline of the July 2026 agent intrusion, which is primary security data.

Coverage: Global · back to 2020

Access (inherited): api · free

Hub API free with rate limits; model licences vary per repository and must be checked individually.

Subscription business model creates an incentive toward scoop-driven framing, but the reporting on private AI company financials is the best available and frequently the only source.

Valuable but PAYWALLED — flag for Phase 2 licensing budget. It is currently the only outlet reporting Anthropic gross-margin projections and cash-flow-positive timing, and this dossier could read only the headline and dek.

Coverage: US-centric, global subjects · back to 2013

Access (inherited): licensed · paid_high

Hard paywall. Full text requires a licence; headlines and deks are public. No circumvention.

VC-authored with a direct portfolio interest: Menlo is an Anthropic investor and its findings show Anthropic leading. Small samples (n=150 in the last verified edition). Treat every figure as an estimate, never a fact.

The most-cited enterprise LLM spend dataset and the most-laundered. Phase 2 must pin every citation to a specific edition and date: the widely-quoted 2026 'Anthropic 40%' figure could not be verified against the primary report in this research.

Coverage: United States, enterprise · back to 2023

Access (inherited): public_page · free

Free publication; citable with attribution.

Law firm alerts are marketing for legal services and select for issues the firm sells into, but are factually careful on docket detail. Utility Dive carries named bylines.

Fastest reliable translation of FERC dockets into what they mean for data-centre interconnection. Useful because FERC's own releases omit quantitative load and queue figures.

Coverage: United States · back to 2012

Access (inherited): rss · free

Copyright holders vary; headline-and-link citation.

Peer review is pending for preprints. Authors' financing-mix figures rely on sell-side projections (Morgan Stanley, Moody's) that are themselves estimates.

Source of the $8.2T / 200GW US buildout estimate and the off-balance-sheet lease analysis. The most rigorous public work on the sector's financing structure and the best counterweight to both bullish and bearish press narratives.

Coverage: Global, US-focused · back to the 1970s

Access (inherited): rss · freemium

NBER working papers partially paywalled for non-subscribers; SSRN and author pages usually free.

Industry-funded consortium with member vendors who sell security products; still the only body publishing structured analysis of autonomous-agent intrusions.

Published independent analysis and a CISO post-mortem on the 2026 Hugging Face agent intrusion. Currently the closest thing to a neutral clearinghouse for agentic-security incidents — a category with no regulator-run reporting regime.

Coverage: Global · back to 2009

Access (inherited): public_page · free

Free registration required for some artifacts; redistribution restricted.

The two trackers disagree by design on deal inclusion, and both are marketing for a commercial product. Neither is truth; report both.

Essential for frontier lab funding, with the standing caveat that two companies took 43% of global H1 2026 venture funding, so any AI sector total is really a two-company total. Report medians and ex-AI figures where available.

Coverage: Global · Crunchbase back to 2007; Venture Pulse back to 2012

Access (inherited): api · freemium

Crunchbase API is paid for bulk access; KPMG Venture Pulse is a free PDF with attribution requirements.

Academic and broadly neutral, but compiles heavily from vendor and commercial datasets, so vendor bias propagates through it. Annual cadence means figures are stale for much of the year.

Chapter PDFs carry the detailed figures; the web version does not render the underlying data, which made extraction unreliable in this research. Phase 2 should ingest the chapter PDFs directly rather than scraping the site.

Coverage: Global · back to 2017

Access (inherited): bulk_download · free

Published under Creative Commons BY-ND; attribution required, no derivatives.

None. Statutory disclosure. Selection bias only in that it covers US registrants; SAP, Infosys and TCS file differently or abroad.

THE backbone for this sector. Earnings exhibits (8-K EX-99.1) carry cRPO, RPO, NRR and segment revenue. Critical limitation: NRR and seat counts are NOT mandated disclosures, so the central seat-versus-consumption question cannot be answered from filings alone.

Coverage: US registrants plus foreign private issuers filing 20-F/6-K · back to 1994

Access (inherited): api · free

Public domain. SEC requires a declared User-Agent and rate limiting to 10 requests/second. No redistribution restriction.

Vendor-controlled framing. 'AI ARR' definitions are self-set, non-comparable across companies and unaudited. Treat headline AI metrics as fact-with-caveat, and vendor narrative as marketing.

Faster than EDGAR for non-US issuers (SAP, Infosys). Prioritise the cRPO / current backlog line and any separately-identified AI ARR line; track pricing-page changes alongside.

Coverage: Global (US, EU, APAC listings) · typically 10+ years of quarterly releases

Access (inherited): rss · free

Press releases generally permit quotation with attribution. Webcast recordings often restricted from redistribution.

Commercial research vendor selling to both buyers and vendors in this sector. Model methodology is not published. Revised its 2026 forecast four times in nine months (9.8% to 14.2%), which is itself a reliability signal.

The single most-quoted number in the sector, and a single modeller with no independent audited counterpart. Phase 2 should capture every revision to build a forecast-stability series. Full data requires a licence budget.

Coverage: Worldwide with regional splits · press-release series back to the 1990s; full data licensed

Access (inherited): rss · free

Press releases are quotable with attribution to Gartner. Underlying forecast documents are paywalled and licence-restricted — no redistribution.

Indeed's own marketplace share biases coverage toward roles advertised on Indeed; measures the flow of new vacancies, not the stock of employment. Seasonally adjusted to a Feb 2020 = 100 baseline.

The highest-frequency read on software labour demand. At 76.12 on 2026-09-04 it directly contradicts the BLS decade projection — Phase 2 should carry both series side by side rather than choosing one.

Coverage: United States (also available for UK, DE, FR, CA, AU) · back to February 2020 baseline; daily 7-day averages

Access (inherited): api · free

FRED API is free with registration. Indeed data redistribution subject to Hiring Lab terms; attribution required.

Bessemer is an investor in the asset class its index measures. Index constituents are rules-based but the universe definition is BVP's. Price index only — multiples come from the BVP site, not FRED.

The cleanest daily read on software sentiment. Pair with the BVP index site for the average revenue multiple (7.5x as of 2026-09-15), which FRED does not carry.

Coverage: Primarily US-listed cloud software · back to 2018-10-02

Access (inherited): api · free

FRED API free with registration; Nasdaq index data subject to Nasdaq licensing for commercial redistribution.

Projections are modelled from industry output and locked on a multi-year cycle, so they lag structural technology shifts by design. No commercial bias.

Authoritative for the employment stock and the decade projection (1,905,400 jobs 2025, +10% to 2035). Deliberately slow-moving — use as the counterweight to high-frequency postings data, never as a nowcast.

Coverage: United States · occupational series back decades

Access (inherited): api · free

Public domain US government work. BLS API free with registration.

Maintainer-led governance with no identified formal foundation; stewardship is concentrated and originated at Anthropic. Supporter quotes in announcements are vendor marketing.

The integration-standard tripwire for this sector. Each spec revision (latest 2026-07-28) changes what agents can do to enterprise applications. Complement with npm/PyPI download telemetry for independent adoption measurement.

Coverage: Global · specification revisions from November 2024

Access (inherited): rss · free

Open specification, permissively licensed. SDK download statistics are published by the project.

Self-selected respondent panel drawn from a community whose traffic is itself declining because of AI — a structural sampling bias that likely worsens each year. Stack Overflow also sells developer products.

Longest continuous public series on developer tool adoption and sentiment. The 2026 finding — 84% usage against 29% trust — is the most-cited adoption/trust divergence in the sector. Raw microdata is downloadable, which is rare and valuable.

Coverage: Global, English-dominant · annual survey back to 2011

Access (inherited): public_page · free

Results published under an open data licence with attribution; raw response data downloadable.

Produced by a cloud vendor that sells AI developer tooling, so framing favours AI investment. Methodology, sample size and instrument design are published, which is unusual and raises reliability substantially.

The most methodologically transparent public evidence on AI's effect on software delivery (2026.01: ~5,000 practitioners, 100+ hours of interviews). Its instability-tax and J-curve findings are the best available counterweight to vendor productivity claims.

Coverage: Global · State of DevOps reports back to 2014

Access (inherited): public_page · free

Reports free to download; redistribution of full PDFs restricted, quotation with attribution permitted.

Analyst firm selling to the services providers it covers. The 2026-06-30 contract-renegotiation piece disclosed no sample size, response rate or methodology — treat as informed industry commentary, not empirical research.

Best early-warning source on services pricing-model change. Use for direction and executive quotes, never for quantified market claims given the absent methodology.

Coverage: Global, services-heavy (US, India, EU) · research archive back to 2010

Access (inherited): rss · free

Public summaries free; full research licensed. No redistribution of licensed material.

Sells advisory to both enterprise buyers and the providers it ranks. PEAK Matrix placements are a commercial product. Blog analysis is more candid than the ranked research.

Strong on services pricing and on the data-platform market. The 2026-09-08 data platform analysis is the best available independent read on Snowflake/Databricks convergence. Full research requires licensing budget.

Coverage: Global, strong India and sourcing coverage · research archive back to 2010s

Access (inherited): rss · free

Blogs quotable with attribution; full research licensed and paywalled.

Openly sceptical editorial stance, which is a useful corrective to vendor marketing but can overstate negative cases. Named bylines throughout. UK-based.

Broke the 'AI deflation' framing for Indian IT (2026-04-28) with direct executive quotes and figures. One of the few outlets that reads non-US earnings calls closely. RSS is clean and free.

Coverage: Global, strong UK/EU and India coverage · back to 1994

Access (inherited): rss · free

Copyright Situation Publishing; quotation with attribution and link. No bulk redistribution.

Ecosystem-adjacent and dependent on founder and investor access; tends to amplify narrative framings such as 'SaaSpocalypse'. Named bylines and direct attributed quotes.

Valuable for dated, attributed investor quotes and for tracking narrative formation. Its March 2026 SaaSpocalypse piece is the canonical articulation of the displacement thesis. Treat quoted investors as `opinion`, never as `fact`.

Coverage: Global, US-centric · back to 2005

Access (inherited): rss · free

Copyright Yahoo/TechCrunch; quotation with attribution. No scraping of full articles.

Deal inclusion rules differ materially from KPMG's — the canonical worked example of definitional divergence in this programme (H1 2026 global VC: Crunchbase $510B vs KPMG $560.4B). Self-reported company data.

Necessary for private-company funding and exit tracking, but never quote a total without naming the tracker and its methodology. Data licence is a real Phase 2 budget line.

Coverage: Global · back to 2007

Access (inherited): api · paid_high

API requires a paid licence; strict redistribution limits. Editorial content quotable with attribution.

Big-4 firm selling services to the companies it counts. Broader deal inclusion than Crunchbase (corporate and PE rounds), producing systematically higher totals.

Pair with Crunchbase and report both. The divergence is a feature: it reveals where the consensus funding number is soft. Published only as a PDF, so Phase 2 needs a parser rather than an API.

Coverage: Global with regional splits · quarterly back to 2013

Access (inherited): public_page · free

Free PDF download; quotation with attribution permitted.

Vendor-authored by definition, but the list prices and unit definitions are commitments rather than claims. Enterprise discounting means list is not transaction price.

The single best leading indicator for the seat-to-consumption shift, and almost nobody monitors it systematically. Diffing pricing pages over time shows the packaging change months before it appears in disclosed revenue. Use Internet Archive snapshots for history.

Coverage: Global, with regional price books · archived via Internet Archive back 10+ years

Access (inherited): public_page · free

Public pages. Respect robots.txt and rate limits; no circumvention of any access control.

None on the primary text. Commission communications carry policy framing; implementation dates have slipped before (AI Act high-risk obligations deferred to 2027-12-02).

Primary text for the Data Act (switching from 2025-09-12, all switching charges banned 2027-01-12) and the AI Act. Multilingual, which partly addresses the non-English source requirement. Watch for deferrals — they are the single biggest killer of regulation-driven trend theses in this sector.

Coverage: European Union, all official languages · back to 1951

Access (inherited): api · free

Reuse permitted under Commission Decision 2011/833/EU with attribution.

Company-controlled disclosure; guidance is a management commitment with incentive to smooth. NASSCOM advocates for the industry it reports on.

The mandatory regional leg for this sector — the AI-deflation evidence originates here, not in US filings. Watch revenue per employee and headcount alongside revenue; the decoupling is the trend. These firms employ several million people between them.

Coverage: India-headquartered, global delivery · quarterly results back 15+ years

Access (inherited): public_page · free

Press releases and results quotable with attribution; also filed with SEBI/NSE/BSE and, for ADR issuers, the SEC.

Media trackers rely on announcements and on employers' self-declared rationale — 'AI' is a convenient framing for post-2021 cost correction, so attribution is systematically unreliable. WARN filings are statutory and far more reliable but US-only and threshold-limited.

Use WARN as the fact base and media trackers only for the stated rationale. Never treat an employer's AI attribution as causal evidence — record it as `marketing` or `opinion`.

Coverage: United States primarily · WARN back to 1989; media trackers back to 2022

Access (inherited): public_page · free

WARN data is public record. Media tracker content is copyrighted; facts are not.

Authors hold positions in the securities discussed and banks may have advisory relationships with the issuers. Public notes are the marketing layer above licensed research.

Valuable for the structural-versus-narrative framing of the 2026 sell-off and for named index figures (MSCI World Software & Services -20% through February 2026). Always disclose the positional conflict when citing.

Coverage: Global equity markets · varies by publisher

Access (inherited): public_page · freemium

Public insight pieces quotable with attribution. Full research is licensed and must not be redistributed.

BVP invests in the category. The index universe definition is proprietary and constituents change, so historical multiple comparisons are not strictly like-for-like.

Carries the average revenue multiple (7.5x) and average growth (20.6%) that FRED's price series does not. Note that the public SaaS multiple and NRR benchmark space is otherwise dominated by SEO content farms — this and company filings are the only defensible sources.

Coverage: US-listed cloud software · back to 2018

Access (inherited): public_page · free

Public page; Nasdaq index licensing applies to commercial redistribution.

Download counts include CI pipelines, mirrors and automated builds, systematically overstating human adoption. Free of commercial bias but easy to misread.

The only independent check on protocol and tooling adoption claims — e.g. MCP's reported half-billion monthly SDK downloads. Always deflate for CI traffic before treating as an adoption measure.

Coverage: Global · npm back to 2010; PyPI back to 2016 via BigQuery

Access (inherited): api · free

npm and PyPI data are public; GitHub API requires a token and rate-limits. Attribution appreciated, not required.

Issuer-controlled. TSMC is systematically conservative on guidance and gives capex guidance verbally on the earnings call rather than in the published results summary, so the call transcript is the primary artefact for capex, not the results page.

The single most valuable feed in this sector. The monthly revenue report (published around the 10th) is a near-realtime read on AI demand roughly six weeks ahead of anyone else's quarterly.

Coverage: Global; Taiwan-domiciled issuer · quarterly results back to 1990s; monthly revenue reports back to 2000s

Access (inherited): public_page · free

Public issuer disclosure; copyright retained by TSMC. Attribution expected; no bulk redistribution of PDFs.

Company-controlled announcements; strategic partnerships are announced when commercially convenient.

Carried the ASML 12-inch photomask initiative (8 Sep 2026) and the Sony image-sensor JV (11 Aug 2026) before most secondary coverage.

Coverage: Global · back to late 1990s

Access (inherited): public_page · free

Copyright TSMC; attribution required.

Issuer-controlled and promotional in tone on HBM leadership; financial figures are audited-standard but product claims ('industry-leading power efficiency') are marketing.

Primary source for HBM4/HBM4E timing and for long-term agreement counts. A genuine regional (Korean) primary source.

Coverage: Global; Korea-domiciled issuer · back to roughly 2015

Access (inherited): public_page · free

Copyright SK hynix; attribution required.

Issuer-controlled. Samsung reports Device Solutions at divisional level only, so memory and foundry economics cannot be separated from the public disclosure - a material limitation for foundry analysis.

Also the best public source on Samsung Foundry's 2nm progress, which it discloses only qualitatively and without naming customers.

Coverage: Global; Korea-domiciled issuer · back to roughly 2012

Access (inherited): public_page · free

Copyright Samsung Electronics; attribution required.

Issuer-controlled but subject to SEC liability. Micron's prepared remarks are unusually specific on contract pricing and bit-growth assumptions - the best public memory pricing disclosure available.

Micron's off-calendar fiscal quarters mean it reports memory pricing roughly a month before the Korean makers, making it the sector's leading indicator on DRAM and NAND price direction.

Coverage: Global; US-domiciled issuer · back to 1990s via EDGAR

Access (inherited): public_page · free

Public issuer disclosure; freely citable with attribution.

None as a source; the filings themselves are issuer-authored but carry legal liability. Covers only US registrants, so TSMC, Samsung, SK hynix and Kioxia are partially or wholly outside it.

Note: efts.sec.gov full-text search was unreachable through the research proxy on 2026-09-15; the Archives path worked. Phase 2 should verify full-text search access.

Coverage: US registrants and foreign private issuers filing 20-F/6-K · back to 1993/1994

Access (inherited): api · free

US government work, public domain. Fair-access rate limits and a declared user-agent are required.

Issuer-controlled. NVIDIA does not break out networking separately in the press release, and its CFO commentary (published separately) carries the supply and China detail that the release omits.

The guidance language ('not assuming any Data Center compute revenue from China') is the cleanest single indicator of export-control reality available anywhere.

Coverage: Global; US-domiciled issuer · back to 1999

Access (inherited): public_page · free

Public issuer disclosure; attribution required for the newsroom copy.

Broadcom aggregates custom accelerators and networking into one 'AI semiconductor revenue' line and does not name customers, so the XPU-versus-networking split must be inferred.

The best available quantitative proxy for custom-ASIC displacement of merchant GPUs.

Coverage: Global; US-domiciled issuer · back to 2009 (Avago)

Access (inherited): public_page · free

Public issuer disclosure.

ASML stopped publishing net bookings prominently in its 2026 releases, which removes what had been the sector's best forward-order indicator. Capacity plans are disclosed, order books increasingly are not.

Next Capital Markets Day is 10 June 2027, at which the 2030 model will be reset. Monitor for any restoration of bookings disclosure.

Coverage: Global; Netherlands-domiciled issuer · back to 1995

Access (inherited): public_page · free

Public issuer disclosure.

Compiled from member-company submissions, so it reflects the reporting perimeter of SEMI members. The free press releases give global and regional headlines only; the full regional table is behind the Equipment Market Data Subscription.

PAYWALL FLAG: the regional breakdown - the part that reveals China's equipment purchasing trajectory - is subscription-only. This is where licensing budget has to go for this sector.

Coverage: Global with seven-region breakdown · back to the 1990s

Access (inherited): public_page · freemium

Headline figures citable with attribution; the underlying WWSEMS dataset is licensed and may not be redistributed. Phase 2 will need a paid subscription for regional detail.

Industry-run and compiled from member submissions; definitions of product categories are set by the members whose revenue is being measured. Free access is limited to headline forecast press releases.

The reference series for 'the semiconductor market size'. Any market-size figure not traceable to WSTS or a company's own revenue should be treated as a vendor model.

Coverage: Global with four-region breakdown · back to 1986

Access (inherited): public_page · freemium

PAYWALL FLAG: detailed monthly billings by product category are member-only. Headline forecasts are citable with attribution.

Policy instrument, not analysis. Rule text is authoritative; the accompanying press framing is political. Licence approval and denial statistics are not published, so the practical effect of a rule change is unobservable from this source alone.

Pair with the Federal Register API rather than scraping BIS. Critical gap: BIS publishes no licence-approval statistics, so 'policy loosened' cannot be distinguished from 'trade happened' without company disclosure.

Coverage: US export control, global effect · Federal Register archive back to 1994

Access (inherited): public_page · free

US government work, public domain.

Purely administrative. The difficulty is analytical, not editorial: Section 232 semiconductor coverage is defined by narrow TPP and DRAM-bandwidth performance bands inside three HTS headings, so tariff exposure cannot be read off trade codes alone.

Monitor headings 9903.79.01/.02 for any Phase 2 expansion to additional semiconductors or manufacturing equipment, which the January 2026 proclamation left open.

Coverage: US imports · HTS revisions back decades

Access (inherited): public_page · free

US government work, public domain.

Taiwanese research house with a commercial subscription business; free press releases are the marketing tier for paid reports. Consistently early and well-sourced on Taiwan supply chain, but forecasts are directional and methodology is not published.

Regional source of record for foundry revenue rankings, DRAM/NAND contract pricing and CoWoS capacity. Treat quantitative estimates as estimates, never as facts.

Coverage: Global with strong Taiwan, Korea and China coverage · back to roughly 2010

Access (inherited): public_page · freemium

Free press releases citable with attribution; full reports licensed. No redistribution.

Taiwan supply-chain trade press with genuine channel access but a poor record on timing and a habit of reporting single-supplier rumours as plans. Frequently the origin of 'TSMC reportedly...' stories that circulate globally without attribution.

Valuable for provenance tracing: when a CoWoS or capex figure appears in Western coverage, it usually originated here or at a Taiwanese daily. Useful for deduplicating syndicated claims into a single source.

Coverage: Taiwan, China, Korea supply chain · back to 1998

Access (inherited): public_page · paid_high

PAYWALL FLAG: most substantive content is subscription-only, and terms prohibit redistribution. Headlines are public.

Korean national outlets with strong access to Samsung and SK hynix supply chains and a national-champion framing. Consensus-versus-actual comparisons are reported more precisely than in Western coverage.

Required regional coverage for this sector. Seoul Economic Daily carried HBM4 2027 pricing expectations (~$4-5/Gb) well ahead of English-language trade press.

Coverage: South Korea; global memory market · archives back over a decade

Access (inherited): public_page · free

Copyright retained; attribution and linking expected; no bulk reproduction.

Japanese business daily with the best access to Japanese equipment and materials makers (Tokyo Electron, Shin-Etsu, JSR, Hoya) and to METI policy. Japan-centric framing on Rapidus.

The materials and chemicals layer of this sector is almost entirely Japanese and almost entirely uncovered in English outside Nikkei. Identified as a data gap in this dossier.

Coverage: Asia, with unmatched Japan depth · archives back many years

Access (inherited): public_page · paid_low

PAYWALL FLAG: metered paywall. Subscription required for consistent access; no circumvention.

Named-byline independent research with genuine supply-chain modelling, but operates a paid consulting and data business with clear commercial interests in the topics it covers. Its figures are models, not observations, and are frequently laundered into other coverage as facts.

Best available public modelling of Chinese AI chip production constraints. Every number is an estimate and must be labelled as one.

Coverage: Global, strong China and Taiwan · back to roughly 2020

Access (inherited): public_page · freemium

PAYWALL FLAG: substantive analysis is largely subscriber-only.

Advertising-supported trade press with heavy vendor contribution; some articles are effectively sponsored. Useful for deployment reality, weak for independent verification of vendor claims.

The best routine source for rack power density, cooling and interconnection-queue constraints - the part of this sector's story that sits outside the fab.

Coverage: Primarily North America · back to 2015

Access (inherited): public_page · free

Copyright Endeavor; attribution and linking expected.

Sector-dedicated outlet with an affiliated market-intelligence business; systematically optimistic framing and a strong tendency to report vendor milestone claims without independent benchmarking. Logical-qubit counts reported here are not comparable across vendors.

Useful as a comprehensive index of who exists and what they claim. Must be paired with arXiv and peer-reviewed sources before any quantum claim is treated as established.

Coverage: Global · back to 2020

Access (inherited): public_page · freemium

Copyright retained; attribution expected.

Preprints are not peer reviewed and corporate quantum groups routinely post results timed to marketing announcements. Peer-reviewed venues lag the press release by six to eighteen months.

The only way to separate genuine quantum error-correction progress from marketing. Track logical error rate per logical qubit per cycle and code distance, not qubit counts.

Coverage: Global · arXiv back to 1994

Access (inherited): api · free

arXiv metadata and full text are available under an open API with per-paper licences varying; Nature and Science content is largely paywalled.

Vendor-authored. IBM's roadmap has historically been more specific and better-kept than competitors', but Starling's 2029 target is a commitment, not a result, and should never be cited as evidence of current capability.

The reference roadmap the whole sector is benchmarked against. Kookaburra (2026) delivering qLDPC memory with an attached logical processing unit is the next falsifiable checkpoint.

Coverage: Global · back to 2016

Access (inherited): public_page · free

Copyright IBM; attribution required.

Regulatory filing system, neutral. Chinese-language interface with partial English; monthly revenue filing is mandatory for listed Taiwanese companies, which is why Taiwan gives better high-frequency visibility than any other semiconductor jurisdiction.

Mandatory monthly revenue disclosure across the whole Taiwanese supply chain - foundry, OSAT, ODM and test - is the highest-frequency hard data available in this sector anywhere in the world. Underused outside Taiwan because the interface is Chinese-language.

Coverage: Taiwan-listed companies: TSMC, UMC, ASE, MediaTek, Quanta, KYEC · back to the early 2000s

Access (inherited): public_page · free

Public regulatory disclosure; no stated redistribution restriction, but rate limiting applies.

Official trade statistics, neutral, but reported in aggregate categories that merge memory types. Korean-language primary release with delayed English summaries.

Published within the first days of each month covering the prior month, and within the first ten days for the first twenty days of the current month. The fastest public read on global memory demand that exists.

Coverage: South Korea exports, global demand proxy · decades

Access (inherited): public_page · free

Korean government data; generally free to use with attribution.

Named-byline independent technology press with a deliberately sceptical editorial voice. Strong on consumer and enterprise price impact, which mainstream semiconductor coverage neglects; occasionally leans on single analyst estimates.

Best routine tracker of memory-price pass-through to consumers - the mechanism by which this sector reaches ordinary people and headline inflation.

Coverage: Global, UK/US-centric · back to 1994

Access (inherited): public_page · free

Copyright Situation Publishing; attribution and linking expected.

US government perspective; scope limited to what CISA and partners choose to publish. Under-reports non-US and non-critical-infrastructure activity. Publication volume is itself sensitive to CISA's own staffing, which contracted ~1/3 in 2025-26, so a fall in advisory count is not evidence of a fall in threat activity.

Anchor Tier-A source for this sector. Joint advisories (AA-numbered) carry TTP detail that vendor reports paraphrase weeks later.

Coverage: US primary; joint advisories co-sealed with Five Eyes, EU and Japanese agencies · back to 2010 as US-CERT; current alert format from 2021

Access (inherited): rss · free

US Government work, public domain; attribution expected. No scraping restrictions beyond standard rate limits.

Inclusion requires evidence of active exploitation plus a clear remediation path, so it systematically under-counts exploitation of products with no patch and of non-US-visible targets. It is a floor, not a census.

Direct JSON feed is the single cheapest high-signal telemetry in the sector. Note: fetching via generic HTTP clients through some proxies returns 403; use a browser user-agent. Pair with DBIR's finding that only 26% of KEV entries were fully remediated by surveyed orgs.

Coverage: Global vulnerabilities; US federal remediation mandate (BOD 22-01) · back to 2021-11 catalog launch; entries backfilled to older CVEs

Access (inherited): bulk_download · free

Public domain US Government work. JSON and CSV feeds explicitly published for machine consumption.

EU institutional perspective; incident corpus is built from open-source and member-state reporting, so it over-weights publicly visible events (DDoS at 76.7% of ETL 2025 incidents is an artefact of hacktivist self-publicity, not of impact). 28.5% of its own incidents lack sector attribution.

The only non-vendor annual threat landscape with a published incident-curation methodology. Essential counterweight to US-centric framing.

Coverage: EU-27 plus EFTA · Threat Landscape annual since 2013

Access (inherited): public_page · free

CC BY 4.0 for most ENISA publications; attribution required.

Covers US-listed issuers only. Captures the disclosure decision, not the incident: only 29 issuers used Item 1.05 in the first two years while 50 used voluntary Item 8.01, so filing counts measure legal strategy as much as breach frequency.

The only breach dataset in this sector that is legally compelled, timestamped and machine-readable. Watch SEC Petition 4-856 (filed 2025-05-22) which asks the Commission to rescind Item 1.05 entirely.

Coverage: US-listed issuers, including foreign private issuers via 6-K · EDGAR back to 1993; Item 1.05 exists from 2023-12-18

Access (inherited): api · free

Public domain. SEC requires a declared User-Agent with contact email and 10 requests/sec cap.

Self-reported complaints only; massive under-reporting of both consumer and enterprise loss, and losses are as-reported not as-adjudicated. Skews to US victims and to fraud rather than intrusion. Ransomware complaint counts (3,611 in 2025) are an order of magnitude below leak-site counts.

Best available non-vendor dollar-loss series for fraud and abuse. 2025 report: 1,008,597 complaints, $20.877B losses, +26% YoY.

Coverage: US primary, some international complaints · annual reports back to 2000

Access (inherited): bulk_download · free

US Government work, public domain.

Vendor-published but with an unusually transparent methodology and ~80 external contributors (law enforcement, IR firms, ISACs, cyber insurers). Contribution bias: over-weights incidents that reach an IR firm or an insurer. Critically, the 2026 edition covers 2024-11-01 to 2025-10-31, so it contains no 2026 data at all despite the '2026' label.

The one vendor study that clears the Tier-B bar on methodology. Always state the observation window when citing it; the label year is one year ahead of the data.

Coverage: Global, ~90+ countries in recent editions · annual since 2008

Access (inherited): public_page · free

Free PDF, no registration wall in recent years; redistribution of the PDF restricted, citation permitted with attribution.

US-only nonprofit, funded partly by identity-protection vendors. Counts publicly reported compromises and notice volumes, so it tracks notification law coverage as much as attack volume. ITRC itself flags that only 24% of H1 2026 notices disclosed an attack vector, the lowest share it has recorded.

H1 2026: 1,803 compromises, 471.2M victim notices, of which a single Instructure Canvas event accounted for 275M. Best source for mega-breach concentration.

Coverage: United States · breach tracking back to 2005

Access (inherited): public_page · freemium

Reports free with registration for the full PDF; attribution required; the underlying notice database is a paid product.

Vendor with a commercial interest in blockchain-analytics demand. Measures only on-chain payments it can attribute, so figures are revised upward for years afterwards (2024 was restated to $892M). Does not capture non-crypto extortion or unreported payments.

2025: $820M paid, -8% YoY; payment rate possibly an all-time-low 28%; median payment $59,556 (up 368%). The direction — fewer payers, larger individual payments — is the analytically important part.

Coverage: Global on-chain flows · back to 2019

Access (inherited): public_page · freemium

Blog summaries free; full report requires registration. Attribution required.

Broker-sourced: reflects Marsh's own placed book, which skews large-corporate and US/UK. A broker has a mild interest in showing clients that it obtains falling rates. Rate 'change' is renewal-on-renewal for comparable programmes, not price levels.

Q2 2026: cyber rates -4% globally, twelfth consecutive quarterly decline; US only -2%. The cleanest available price signal for how insurers actually rate cyber risk, as opposed to how vendors describe it.

Coverage: Global, split into US, UK, Europe, Asia, Pacific, LAC, IMEA · quarterly index back to 2012

Access (inherited): public_page · free

Free public release each quarter; attribution required.

Regulator's own description of its regime; says what is required, not what is complied with. No enforcement data published yet because obligations only began 2026-09-11.

Manufacturer obligations live from 2026-09-11: 24h early warning, 72h notification, 14-day final report on actively exploited vulnerabilities, via the CRA Single Reporting Platform to a national CSIRT and ENISA. OSS stewards from 2027-12-11. This is the highest-volume new machine-readable incident stream in the sector.

Coverage: EU-27, extraterritorial for any product placed on the EU market · CRA adopted 2024-10; reporting pages from 2025

Access (inherited): public_page · free

© European Union, reuse permitted under Decision 2011/833/EU with attribution.

Launched partly as a hedge against US CVE Program funding instability, so its existence is itself a political statement about dependence on MITRE/CISA. Enrichment is still thinner than NVD for older CVEs.

Monitor alongside cve.org and NVD. Divergence between EUVD and NVD enrichment status is a leading indicator of fragmentation in global vulnerability coordination.

Coverage: Global vulnerabilities, EU institutional view · live from 2025-05

Access (inherited): api · free

Open access; ENISA reuse terms with attribution.

Authoritative for what is proposed and finalised, silent on whether deadlines will hold. CIRCIA's final rule has already slipped from 2025 to May 2026 to an expected Sept-Autumn 2026, which is the single best illustration in this sector of regulatory-deadline slippage.

Best single monitor for the US regulatory wrapper. Track CIRCIA (72h incident / 24h ransom payment reporting, ~300,000 covered entities), CMMC interim final rule, and FAR cyber standardisation together.

Coverage: United States · back to 1994 online

Access (inherited): api · free

Public domain; documented free JSON API with no key required.

Audited revenue is fact; ARR, 'platformisation customers' and 'AI ARR' are company-defined non-GAAP metrics that change definition between quarters. Treat segment ARR as management commentary, not measurement.

The demand-side reality check on every vendor threat narrative. CrowdStrike Q2 FY27 (quarter ended 2026-07-31): $5.84B ARR, +25%. Palo Alto FY26 (ended 2026-07-31): $11.48B revenue, NGS ARR $9.10B, +63% — but that growth is substantially inorganic post-CyberArk.

Coverage: Global revenue, US-listed reporting · since each company's IPO

Access (inherited): rss · free

Public company disclosures; freely citable.

Frontier-lab vendor source. Anthropic has a reputational interest in demonstrating both that misuse is serious and that it detects misuse well. Visibility is limited to abuse of Anthropic's own models; it says nothing about attacks run on other models or on self-hosted open-weight models. No independent verification of the September 2026 findings had been published as of this research date.

Sept 2026 report (covering 2025-12 to 2026-08) is the most concrete public account of agentic AI in real intrusion campaigns: 13-agent autonomous collection fleets, 2,100+ Azure tokens across 40+ tenants in 34 hours, initial-access-to-bulk-exfiltration in 2-3 hours. Mirror sources: OpenAI and Google GTIG publish comparable reports; triangulate across all three before treating any claim as established.

Coverage: Global, actor-attributed · from 2024

Access (inherited): public_page · free

Public web pages; attribution required.

Volunteer consortium with heavy vendor participation in working groups; taxonomy work is strong, prevalence claims are weak. 'State of Agentic AI Security and Governance' tracks project telemetry (advisory counts, release cadence), not enterprise deployment.

v2.01 (2026) tracked 53 agentic projects and 42 regulatory instruments across 10 jurisdictions; 28 of 53 tracked projects are coding agents. Prompt injection maps to six of the ten Agentic Top 10 categories — the closest thing to a consensus taxonomy for this attack surface.

Coverage: Global · LLM Top 10 from 2023; agentic material from 2025

Access (inherited): public_page · free

CC BY-SA; attribution and share-alike.

Industry consortium funded by security vendors; research notes often synthesise vendor findings (e.g. its MCP note relies on OX Security scanning). Exposure estimates such as '200,000 vulnerable MCP instances' are modelled, not enumerated.

Its 2026-05-04 MCP note is the best single public inventory of the MCP attack surface: STDIO command-injection across all official SDKs, tool poisoning, rug-pull and cross-server shadowing, with named CVEs (CVE-2025-49596, CVE-2025-54136, CVE-2026-30623, CVE-2026-30615).

Coverage: Global · from 2009

Access (inherited): public_page · free

Free download with registration for most artifacts; CSA attribution terms.

Standards-setting authority with no commercial interest, but it publishes algorithms and guidance, not adoption data. Migration progress must be measured elsewhere.

Pair with EO 14412 deadlines (2030 key establishment, 2031 signatures for federal HVAs) and with Cloudflare Radar for actual deployed share. The gap between the standard, the mandate and the deployment is the whole PQC story.

Coverage: US standards with de facto global adoption · PQC competition from 2016; FIPS 203/204/205 finalised 2024-08

Access (inherited): rss · free

Public domain.

Vendor telemetry, but measurement rather than opinion. Represents only traffic traversing Cloudflare's network, which over-weights consumer web and under-weights enterprise east-west, VPN and machine-to-machine traffic. 'Post-quantum protected share' therefore flatters overall internet readiness.

Roughly two-thirds of browser-originated traffic to Cloudflare was post-quantum protected by mid-2026 — the only hard, continuously updated PQC adoption metric in existence. Note the non-commercial licence clause.

Coverage: Global, country-level breakdowns · from 2022; PQC series from 2023

Access (inherited): api · freemium

Free API with token; CC BY-NC 4.0 on Radar data, so commercial redistribution needs a licence — relevant for Phase 2 budgeting.

Healthcare only, US only, and only breaches affecting 500+ individuals. Entity-reported affected counts are frequently revised upward months later. Still the only legally compelled, per-incident, publicly enumerated breach register in the US.

The methodological model for what CIRCIA could become. Use it to sanity-check vendor claims about healthcare breach trends, which are routinely inflated.

Coverage: United States healthcare sector · back to 2009 (HITECH)

Access (inherited): bulk_download · free

Public domain; CSV export provided.

UK national perspective; 'nationally significant incident' is an NCSC-defined category that has been redefined, so year-on-year counts are not strictly comparable. Attribution statements are policy acts as well as technical ones.

Best non-US, non-EU national source. OGL v3.0 makes it unusually easy to redistribute, unlike most vendor research.

Coverage: United Kingdom, with international joint advisories · Annual Review since 2017

Access (inherited): rss · free

Open Government Licence v3.0 — permissive, attribution required.

Primary material is Japanese-language; English summaries lag and compress. Coverage reflects Japanese critical-infrastructure definitions (15 sectors), which do not map cleanly onto US 16 or EU NIS2 categories.

Required non-English regional source for this sector. The ACD law phases in through 2027 and gives government access to cross-border communications data plus offensive pre-emption powers — the most significant constitutional shift in allied cyber policy in a decade, and badly under-covered in English-language security media.

Coverage: Japan · NISC from 2015; ACD law enacted 2025-05-16

Access (inherited): public_page · free

Japanese government terms of use; reuse generally permitted with attribution.

Record volume tracks CNA participation and researcher incentives as much as underlying vulnerability creation; a rising CVE count is not straightforwardly a rising risk count. The programme's contract funding has been publicly precarious since April 2025, which is itself a monitorable risk. Site requires JavaScript; use the GitHub bulk repository instead.

Use the cvelistV5 Git repository, not the website. Cross-check against EUVD; a persistent divergence would be the clearest signal that vulnerability coordination has fragmented along geopolitical lines.

Coverage: Global · back to 1999

Access (inherited): bulk_download · free

CVE List is free to use with attribution; CVE is a registered trademark of MITRE.

Certification body with a direct commercial interest in demonstrating a skills gap that its certifications address. Self-selected respondent panel skewed to ISC2 members. The 'workforce gap' figure is a model output, not a count of vacancies, and should never be cited as fact.

The 2025 edition (published 2025-12-04) is directionally useful for one thing: budget and headcount constraint persisting while AI reshapes skill demand. Everything quantitative in it should carry an 'estimate' or 'marketing' label.

Coverage: Global, ~15 markets · annual since 2004 (as (ISC)2 Global Information Security Workforce Study)

Access (inherited): public_page · free

Free with registration; attribution required.

Counts disclosed, announced rounds only; systematically misses unannounced and non-US rounds, and double-counts extensions. Diverges materially from KPMG Venture Pulse on the same quarters for definitional reasons (see macro brief). Editorial framing favours 'funding is strong' narratives.

Use for round-level facts (company, amount, date, lead) and never for sector totals without pairing against a second tracker.

Coverage: Global, US-skewed · back to 2007

Access (inherited): public_page · paid_high

News free to read and cite; the underlying database API is enterprise-licensed. Phase 2 licensing budget will be needed here.

Named-byline trade press with real sourcing, but The Record is owned by Recorded Future (a threat-intelligence vendor) and Cybersecurity Dive runs heavy sponsored content adjacent to editorial. All three depend on vendor and agency access, which shapes what gets covered and how sceptically.

Fastest reliable route to federal cyber policy movement (CIRCIA slippage, CISA 2015 extensions, CISA staffing). Use as a discovery layer that points at Tier-A primary documents, not as the citation itself.

Coverage: US-centric with international policy coverage · CyberScoop from 2016, The Record from 2020, Cybersecurity Dive from 2019

Access (inherited): rss · free

Free RSS; standard copyright, headline-and-link use only.

US federal statistical agency; independent of policy offices by statute. Conservative on demand inflections — historically under-forecast data-centre load. No commercial interest.

The single highest-value feed for this sector. Machine-readable via the EIA Open Data API v2. Release is ~9th business day monthly; the accompanying data browser gives full time series.

Coverage: United States, with global oil/gas balances · back to 1997 (monthly archive)

Access (inherited): api · free

US Government work, public domain; citation requested, no redistribution limits.

Same agency as STEO. Today in Energy is editorially selective — it highlights what EIA analysts find notable, so absence of a topic is not evidence of absence.

Form EIA-860M gives planned capacity additions and retirements plant-by-plant — the best leading indicator of what actually gets built versus announced.

Coverage: United States, state and balancing-authority level · back to 1990 for most series

Access (inherited): api · free

Public domain; attribution requested.

Statutory ERO with a reliability mandate; structurally biased toward flagging risk, and its assessments have been credibly criticised (Grid Strategies, July 2026) for excluding likely-to-connect queue resources and non-firm imports. Read as an upper bound on risk.

LTRA (winter) and Summer/Winter Reliability Assessments are the anchor documents. Level 2/3 Alerts — e.g. the May 2026 data-centre load-loss alert — are the genuinely novel signal and are published irregularly.

Coverage: United States, Canada, part of Baja California · LTRA annual since 1995

Access (inherited): public_page · free

Copyright NERC; reproduction permitted with attribution for non-commercial use.

Primary legal record, no editorial bias. Volume is enormous and unfiltered; commissioner statements carry political colouring that the orders themselves do not.

Dockets RM26-4 (large-load interconnection ANOPR/NOPR) and EL25-49 (PJM co-location) are the live rule-making fronts. The June 18 2026 Section 206 show-cause orders to all six RTOs generate compliance filings on fixed clocks — highly predictable monitoring targets.

Coverage: United States interstate transmission and wholesale markets · back to 1981 (full text from 2000)

Access (inherited): api · free

Public record; bulk access permitted subject to rate limits in FERC's eLibrary API terms.

The RTO is a party in interest in its own market design disputes; its load forecast has been contested by state consumer advocates (Maryland OPC) and by manufacturers as inflated. Auction results themselves are audited facts.

Largest single power market on earth by served load and the epicentre of the data-centre question. The January load forecast and the annual Base Residual Auction are the two fixed-date events that move the sector.

Coverage: 13 US states plus DC (Mid-Atlantic, Ohio Valley) · load forecasts annual since 1990s; BRA results since 2007

Access (inherited): rss · free

Public; PJM data terms restrict commercial redistribution of some market datasets.

ERCOT itself has publicly warned its 474 GW large-load queue is 'higher than expected future load growth' — it publishes its own forecast with a credibility caveat attached. Texas-specific market design limits generalisability.

The only US market publishing a granular large-load interconnection queue. The August 2026 Batch Zero suspension makes the queue's composition, not just its size, the thing to watch.

Coverage: Texas (~90% of state load) · back to 2002

Access (inherited): bulk_download · free

Public data; ERCOT terms of use prohibit representing derived data as ERCOT-official.

Independent of PJM management by design and frequently adversarial to it. Strong analytical rigour; a consistent editorial view that PJM market design over-rewards incumbent capacity.

The only body that decomposes wholesale cost increases by driver. Its 2026 finding that data-centre load accounted for 9% of PJM wholesale costs ($10.48/MWh) is the closest thing to an audited attribution figure in the sector.

Coverage: PJM footprint · annual reports back to 1999

Access (inherited): public_page · free

Public reports; copyright Monitoring Analytics, attribution required.

DOE-funded national lab; methodologically transparent and peer-reviewed. Publishes annually with a ~6-month lag, so it is a structural indicator, not a fast one.

The authoritative source for queue composition and completion rates. Its 2026 edition — gas requests +86% while solar, wind and storage each fell 16-19% — is the cleanest single evidence of the sector's capital rotation.

Coverage: All US ISO/RTOs plus non-ISO utilities · back to 2000

Access (inherited): bulk_download · free

Public; CC-BY-style attribution to LBNL requested. Underlying data released as downloadable spreadsheets.

Intergovernmental body funded by member states; historically conservative on solar deployment forecasts and has revised them upward repeatedly. Headline reports are free; the detailed datasets behind them are licensed.

The only genuinely global counterweight to US-centric reporting. Note the licensing wall: Phase 2 needs budget here if it wants the underlying electricity statistics rather than the narrative reports.

Coverage: Global, with OECD depth and weaker emerging-market granularity · back to 1974

Access (inherited): public_page · freemium

Free summaries; full datasets require paid licence. Redistribution of licensed data prohibited.

UK national regulator; publishes its own methodology and the cap decomposition into wholesale, network and policy cost. No commercial interest; politically exposed on affordability.

Best available quarterly read on how a gas-exposed retail market transmits geopolitical shocks to households. The October 2026 decision explicitly attributed the rise to Middle East conflict — a clean, dated causal claim from a regulator.

Coverage: Great Britain (England, Scotland, Wales; not Northern Ireland) · cap decisions quarterly since 2019

Access (inherited): public_page · free

Open Government Licence v3.0 — free reuse with attribution.

European TSO association — members are the entities being assessed, which softens incident reporting. The Iberian blackout expert panel nonetheless produced a technically frank root-cause analysis.

Genuine realtime API with per-country generation, load and cross-border flows. The single best non-US structured dataset in this sector.

Coverage: 39 European TSOs across 36 countries · back to 2015 for most series

Access (inherited): api · free

Free registration required for API; ENTSO-E Transparency Platform terms permit reuse with attribution.

Issuer disclosure — legally constrained but promotional in framing. 'Slot reservation agreements' are a company-defined, non-GAAP pipeline metric with no standard conversion rate to firm orders; treat as an intent indicator.

The closest available proxy for the global gas-turbine order book. Siemens Energy and Mitsubishi Heavy Industries IR pages are the necessary triangulation partners — none of the three alone gives the market picture.

Coverage: Global · since 2024 spin-off from GE

Access (inherited): rss · free

Public issuer disclosure; SEC filings are public domain, press releases are copyright GE Vernova.

Independent safety regulator; the record is complete and unspun. Slow-moving and procedurally dense — the signal is in docket milestones, not press releases.

The only way to distinguish a real SMR from a press release. Construction permit issuance (TerraPower, March 2026) and application acceptance are the falsifiable milestones; everything upstream of a docketed application is marketing.

Coverage: United States · back to 1999 (paper record earlier)

Access (inherited): api · free

Public record, no restrictions.

Explicitly promotional — DOE communications advance the administration's nuclear-expansion agenda and state targets (400 GW by 2050) with no stated methodology. Loan disbursement figures are hard facts; capacity targets are policy aspiration.

Use for verified loan closings and disbursements (Palisades $1.52B, Crane $1B) and ignore the aspirational capacity targets unless a second source confirms. Section 202(c) emergency orders are also posted here.

Coverage: United States · varies by administration

Access (inherited): rss · free

US Government work, public domain.

Advertiser-supported trade press with named bylines and consistent sourcing to primary documents. Editorially sympathetic to grid modernisation; covers utility and developer perspectives more than ratepayer ones.

The fastest reliable secondary source for RTO, FERC and state PUC developments, and it consistently links the primary document. Best used as a discovery layer that routes to Tier A sources.

Coverage: United States, some Canada · back to 2012

Access (inherited): rss · free

Free with registration; copyright Industry Dive, no bulk redistribution.

Long-established generation-sector trade publication. Vendor advertising is prominent and some features are sponsor-adjacent; named technical bylines are reliable, listicles less so.

Unusually good on plant equipment, turbine and transformer supply chains, where general business press is thin. Cross-check any market-size figure it repeats.

Coverage: Global, US-weighted · back to 1882 (digital archive from ~2000)

Access (inherited): rss · free

Free registration; copyright Access Intelligence.

The investor-owned utility trade association. Its capex figures are member-reported and it lobbies for rate-base growth — a direct commercial interest in the narrative that enormous investment is required. Figures are nonetheless the only industry-wide aggregate available.

Its rolling list of large customer projects and tariffs (August 2026 edition) is the best single register of who is signing what. Treat the $1.4T capex figure as an industry plan, not a commitment — it is revised upward annually.

Coverage: United States investor-owned utilities (~72% of US customers) · capex series back to 2000s

Access (inherited): public_page · free

Public summaries free; detailed datasets are member-only. Attribution required.

Philanthropically funded advocacy-adjacent research organisation with an explicit decarbonisation mission. Its unit-level datasets are rigorous and widely cited by IEA and academics; framing of findings is directional.

The practical route to Chinese power-sector unit data without reading Chinese-language filings. Its H1 2026 finding that curtailed wind and solar (360 TWh) exceeded China's entire demand growth is the kind of number no vendor report will surface.

Coverage: Global, with exceptional China and India coverage · back to 2014

Access (inherited): bulk_download · free

CC BY 4.0 for most trackers — free reuse with attribution.

Commercial research arm of a terminal business; headline totals are published free as marketing for a paid subscription. Methodology is documented but the underlying deal-level data is not auditable without a licence.

PAYWALLED — Phase 2 must budget here or accept only the press-release headline numbers. Its data-centre demand trajectory (1,114 TWh by 2050) is materially more conservative than US utility forecasts imply, and that divergence is itself worth tracking.

Coverage: Global · investment series back to 2004

Access (inherited): licensed · paid_high

Free press summaries only; full reports require enterprise subscription. Redistribution prohibited.

Chinese state statistics and state-affiliated press. Capacity and generation figures are generally reliable and internationally reconciled; framing is nationalist and curtailment or utilisation data is published selectively or with lag.

Essential non-English regional source — China is roughly half of global demand growth. The September 2026 milestone (installed solar capacity overtaking coal) came from NEA first. Pair with Global Energy Monitor for what NEA does not publish.

Coverage: China · back to 2000s

Access (inherited): public_page · free

No explicit reuse licence; attribution to NEA/CEC standard practice.

Primary legal record of the two most consequential state-level large-load proceedings. No editorial bias; navigation is poor and there is no bulk API.

PUCT Docket 59220 (FGE Goodnight / co-located data centre curtailment, July 2026) and the PUCO AEP Ohio data-centre tariff are the two precedent-setting decisions other states are copying. State dockets now lead FERC on large-load policy.

Coverage: Texas; Ohio · back to ~2000

Access (inherited): public_page · free

Public record; no scraping restrictions stated but rate limits apply in practice.

RTO self-reporting, audited auction results. MISO's seasonal construct makes its prices non-comparable with PJM's annual construct without adjustment — a common source of misleading cross-market comparisons.

The necessary control case against PJM. MISO capacity prices fell 42% in the same year PJM's hit the cap — monitoring only PJM produces a systematically distorted view of US capacity scarcity.

Coverage: 15 US states and Manitoba · PRA results back to 2009

Access (inherited): bulk_download · free

Public; MISO data terms require attribution.

Consultancy whose clients include transmission developers and clean-energy interests; it has a commercial stake in arguing that more transmission and queue reform would resolve adequacy concerns. Its critiques of NERC are technically substantive and should be read as the opposing brief, not neutral analysis.

Valuable precisely because it is adversarial to NERC. Holding both against each other — NERC's 13-of-23 elevated-risk finding versus Grid Strategies' '5-93% above target reserves' — is more informative than either alone.

Coverage: United States · back to 2019

Access (inherited): public_page · free

Free public notes; copyright Grid Strategies.

University research centre, peer-review-adjacent and methodologically transparent. Its headline flexible-load finding is widely cited by data-centre developers to argue against new build, which gives it a constituency it did not choose.

Source of the ~100 GW curtailment-enabled headroom estimate that now underpins FERC's flexible-load service-class proposal. A rare case of an academic paper directly shaping a rule-making.

Coverage: United States · back to 2005

Access (inherited): public_page · free

Free download; academic attribution expected.

Issuer-controlled. Financing closes, PPA signatures and named lenders are checkable facts; capacity, cost and schedule claims are promotional until a regulator or lender validates them. Private companies face no disclosure obligation at all.

Treat named non-recourse lenders as the real verification signal — eight banks underwriting Fervo's $421M Cape Station facility is stronger evidence of technical bankability than any capacity claim. Announcement without a lender or a regulator attached is marketing.

Coverage: Global, US-weighted · varies by company

Access (inherited): rss · free

Copyright respective issuers; press material generally free to quote with attribution.

The regulator's own record of its own decisions. No commercial bias, but it is a scope-limited dataset: CDER novel approvals only, excluding biologics reviewed by CBER, supplemental indications, generics and devices. Counting 'the number of FDA approvals' from this page alone systematically undercounts agency output.

The canonical approval count. Pair with the Drugs@FDA and openFDA APIs for structured pulls; this page is HTML-only and must be scraped or read manually. 39 novel approvals through 2026-09-11.

Coverage: US · annual lists back to 2011; predecessor datasets to the 1990s

Access (inherited): public_page · free

US Government work, public domain. No restrictions on reuse; attribution conventional.

Primary regulator statements, but selective: FDA announces what it chooses to announce. Safety actions, first-in-class approvals and policy pilots are covered; routine approvals and most rejections are not. Complete Response Letters are generally not announced by FDA at all, so rejection data must come from sponsors.

Source of the Elevidys Boxed Warning (2025-11-14), the Lumipulse Alzheimer's blood-test clearance (2025-05-16) and the CNPV second approval (2026-02-26). This is the highest-value single feed for the sector.

Coverage: US · back to the late 1990s

Access (inherited): rss · free

US Government work, public domain.

FDA explicitly states the list is not comprehensive and that devices are identified primarily by AI-related terms appearing in summary descriptions. It therefore undercounts AI-enabled devices whose submissions did not use those terms, and it says nothing about whether a cleared device is in clinical use. It is a clearance register, not an adoption register.

Last updated 2026-09-04 with the newest entries dated 2026-06-29. Radiology dominates the panel breakdown. The page is a long HTML table that truncates on automated fetch — Phase 2 should pull the underlying device records via the openFDA 510(k) and PMA endpoints instead.

Coverage: US · authorisations back to the mid-1990s, list published from 2022

Access (inherited): public_page · free

US Government work, public domain.

The price-setter publishing its own results. Savings estimates are counterfactual models produced by the agency running the programme and should be labelled as estimates, not facts. Selection criteria and the spending figures used to justify them are CMS's own.

Selected-drug lists are on the public page but the actual maximum fair prices and the agency's explanations sit inside downloadable ZIP archives, which is where a Phase 2 pipeline needs to point. Key dates: cycle 1 effective 2026-01-01, cycle 2 (includes semaglutide) 2027-01-01, cycle 3 (first Part B drugs) 2028-01-01.

Coverage: US · programme launched 2023; three cycles announced to date

Access (inherited): public_page · free

US Government work, public domain.

Registration is mandatory for many but not all trials, and non-US trials are inconsistently registered here versus in EU CTIS or the Chinese registry. Results posting is frequently late or absent. Sponsor-written descriptions are marketing-adjacent; the structured fields are not.

The single best structured feed for converting pipeline claims into checkable facts — trial start dates, enrolment, endpoints, sites. Used here to verify rentosertib's Phase III (NCT07687459). Cross-reference the Chinese registry (chinadrugtrials.org.cn, CTR numbers) for China-only trials.

Coverage: Global, US-weighted · back to 1999

Access (inherited): api · free

Public domain; NLM asks that the data not be represented as endorsed by NIH. Bulk API access permitted under published rate limits.

Legally binding disclosure with liability attached, so it is the most reliable source for segment revenue, R&D spend, risk factors and deal terms. Covers US-listed issuers only — private biotech, European and Chinese-listed companies are absent, and non-GAAP framing in 8-K exhibits is management's.

Note: efts.sec.gov was blocked by the agent proxy during this research pass, so EDGAR figures could not be pulled directly. Phase 2 must confirm network access. Essential for verifying product-level revenue such as Elevidys, Casgevy and Foundayo rather than relying on trade-press restatements.

Coverage: US-listed issuers, including foreign private issuers on 20-F · back to 1993-94

Access (inherited): api · free

Public domain. SEC requires a declared User-Agent header and enforces rate limits; automated access without one is blocked.

Peer-reviewed and the highest evidentiary standard available for clinical claims, but subject to positive-results publication bias and to author conflicts of interest — industry-sponsored trials are common and the sponsor frequently employs the authors. Read the competing-interests statement, not just the abstract.

The publication of record for the rentosertib Phase 2a trial (2025-06-03). Under this programme's rules, a clinical efficacy claim is only a 'fact' when it appears here or in a regulator's own document — press releases announcing trial results remain 'marketing'.

Coverage: Global · back to 1995

Access (inherited): rss · freemium

Most content is subscription-only; some articles are open access under CC BY. Text and data mining requires a licence for subscription content. Abstracts are freely readable.

Peer-reviewed health informatics. Studies of commercial AI tools are frequently conducted by the deploying health system, which has sunk cost in the deployment; several publish with vendor products anonymised as 'Product A' and 'Product B', which limits independent replication. Early-adopter selection bias is endemic in this literature and is usually declared.

Source of the only randomised evidence found on ambient scribe effect size (Vol 33 Issue 5, May 2026). This is the journal that will settle whether healthcare AI claims survive contact with measurement. JMIR Medical Informatics is the natural companion feed.

Coverage: Global, US-weighted · back to 1994

Access (inherited): rss · freemium

Mixed open access and subscription. Check per-article licence before redistribution.

EU regulator's own output, with an institutional interest in demonstrating EU competitiveness in medicines regulation. Divergence from FDA on the same product is common and informative rather than an error by either.

The essential non-US regulatory feed. Also carries the EU medicines shortages catalogue, which is an early indicator of supply-chain stress — Saxenda (liraglutide) appeared on it during 2026. CHMP monthly opinion highlights are the best forward indicator of EU approvals.

Coverage: EU / EEA · back to the mid-1990s

Access (inherited): rss · free

Reuse permitted under the EMA legal notice with source acknowledgement.

Primary and legally constrained for financial figures, but promotional in framing for clinical and launch claims. Trial-result press releases are vendor claims until the data is published or the label is approved, and must be labelled 'marketing' under this programme's rules.

The single most consequential company feed in the sector given Lilly's Q2 2026 revenue of roughly $23bn. Note that some deep-linked press-release URLs 404 once archived — resolve from the index page rather than storing permalinks.

Coverage: Global · back to the late 1990s

Access (inherited): rss · free

Copyright Eli Lilly; quotation with attribution is standard practice. No bulk redistribution.

Danish-listed issuer subject to EU market-abuse disclosure rules, so financial statements are reliable; competitive and pipeline framing is promotional. Reports in Danish kroner, which introduces FX translation noise in any USD comparison against Lilly.

The key non-US primary corporate source for this sector. The landing page returns only a template on automated fetch — Phase 2 should use the feed or the dedicated company-announcement archive. Quarterly reports are the authoritative source for oral Wegovy revenue in DKK.

Coverage: Global · back to the early 2000s

Access (inherited): public_page · free

Copyright Novo Nordisk; quotation with attribution standard.

A vendor with a direct commercial interest in the AI-drug-discovery narrative. Platform-productivity metrics (candidate nominations, 12-18 month timelines) are self-reported and unaudited and must be labelled 'marketing'. Trial registration numbers and site counts within the same releases are independently checkable and are reliable.

Included because it is the primary source on the only AI-originated asset in Phase III, and because it is a China/Hong Kong-headquartered primary source in a sector whose centre of gravity has partly shifted east. Always cross-check its clinical claims against ClinicalTrials.gov and the peer-reviewed literature.

Coverage: China, Hong Kong SAR, US · back to roughly 2015

Access (inherited): public_page · free

Copyright Insilico Medicine; quotation with attribution standard.

Named bylines, consistent sourcing and correction practice. Advertising-funded by the industry it covers, and its story selection follows public-company news flow, so private and non-US companies are under-covered.

The best free daily feed for biopharma. Its quarterly IPO-tracking pieces are the most methodologically explicit of the trackers surveyed, which is why its IPO counts differ from BioSpace's.

Coverage: US-weighted, global coverage · back to 2014

Access (inherited): rss · free

Copyright Industry Dive. Free registration for some content; no bulk scraping under terms of use.

Strong investigative health and biotech reporting with named reporters. Editorially sceptical of industry claims, which is useful but is a stance. Most substantive data reporting sits behind the STAT+ paywall, so automated collection returns headline figures only.

This is where the licensing budget has to go. The M&A running totals used in this dossier ($134bn YTD to 2026-06-22) came from a STAT+ piece whose methodology sits behind the paywall — Phase 2 needs a subscription to audit it.

Coverage: US-weighted, global · back to 2015

Access (inherited): rss · paid_high

Copyright Boston Globe Media. STAT+ content is licensed per seat; redistribution prohibited. Do not circumvent the paywall.

Well-sourced biopharma trade reporting with named bylines and strong access to financing and FDA-decision news. Substantially paywalled. Cited by other trackers (Ropes & Gray, among others) as an underlying data source, which creates syndication risk — two outlets citing Endpoints are one source, not two.

Second licensing priority after STAT+. Its China-biotech coverage is the strongest in English-language trade press, which matters given the out-licensing trend.

Coverage: Global, US and EU weighted · back to 2016

Access (inherited): rss · paid_high

Copyright Endpoints News; enterprise licensing required for redistribution.

Named bylines and real reporting, but the business is partly a life-sciences job board and events operation, and coverage leans toward industry-positive framing. Its aggregate counts (IPO tallies, FDA decision tallies) use undisclosed inclusion criteria and diverge materially from BioPharma Dive's — treat as one estimate among several, never as the settled figure.

Useful for sector-sentiment pieces and its H1 FDA-decision counts. Its H1 2026 IPO count of 18 against Morse's 12 is one of the two headline contradictions recorded in this dossier.

Coverage: US-weighted · back to the mid-2000s

Access (inherited): rss · free

Copyright BioSpace; standard fair-use quotation.

Named bylines, fast on financing and regulatory news. Advertising- and events-funded by the industry covered; sponsored content is labelled but sits alongside editorial. Fierce Healthcare's digital-health coverage frequently restates vendor funding claims without independent verification.

Best free daily coverage of financing rounds and FDA actions across biotech, pharma, medtech and healthcare AI. Four separate titles worth polling: Fierce Biotech, Fierce Pharma, Fierce Healthcare, Fierce Medtech.

Coverage: US-weighted, global · back to the mid-2000s

Access (inherited): rss · free

Copyright Questex; standard fair-use quotation.

Rock Health is itself a digital-health venture investor, so it has a commercial interest in the sector appearing fundable. Its inclusion criteria are explicit — US-headquartered companies, deals above $2m, venture rounds only — which makes it comparable over time but means it excludes non-US activity and most corporate and PE rounds. Not directly comparable to Crunchbase or KPMG totals.

The standard series for US digital health venture funding. H1 2026: $7.4bn across 244 deals, 45% of capital in 20 megadeals, zero IPO completions. Its consistent methodology makes the concentration trend credible even where the absolute totals are contested.

Coverage: US only · back to 2011

Access (inherited): public_page · free

Free to read and cite with attribution; underlying deal database is proprietary.

Primary for the existence and named parties of MFN agreements, and political for everything else. Claims such as '89% of the branded drug market' and '$700m in patient savings' are self-reported by an interested party with no published methodology and no independent audit — record them as estimates, never as facts. The full agreement texts are not published.

The only public register of which manufacturers have signed MFN deals. Archive copies aggressively — content is removed at administration transitions, and these agreements' terms may become historically contested.

Coverage: US · current administration only; prior content is archived off-domain at each transition

Access (inherited): public_page · free

US Government work, public domain.

The legally operative text of rules, proclamations and notices, as distinct from the press release describing them. No bias in content; the practical limitation is volume and the gap between a proclamation's effective date and the implementing customs guidance.

The correct primary source for the 2026-04-02 Section 232 pharmaceutical proclamation and for IRA implementation guidance. Law-firm advisories such as Arnold & Porter's are faster and more readable but are secondary; use them to find the citation, then read the instrument.

Coverage: US · back to 1994 in full text; 1936 in scanned form

Access (inherited): api · free

US Government work, public domain. Well-documented free JSON API with generous rate limits.

An index, so it inherits the biases of the literature: positive-results publication bias, industry funding of the trials most relevant to this sector, and systematic under-representation of negative and non-English results. Indexing is not endorsement — predatory journals are present.

The gate that separates a 'fact' from 'marketing' for every clinical claim in this sector. Note that PMC article pages serve a reCAPTCHA to automated agents — use the E-utilities API rather than fetching article HTML.

Coverage: Global · back to 1966 (MEDLINE); selected records to the 1800s

Access (inherited): api · free

Metadata and abstracts free via E-utilities under NCBI usage policy (rate-limited, API key recommended). PMC full text varies by licence; the Open Access Subset permits text mining, the rest does not.

Named bylines and consistent use of CMS enrolment data. Provider- and payer-side coverage is US-only. Advertising-funded by the industry covered.

The most reliable free tracker of Medicare Advantage enrolment by carrier, working from monthly CMS files. Source of the 2026 MA deceleration data. MedTech Dive is the companion feed for the device sector, which this research pass could not cover in depth.

Coverage: US · back to 2014

Access (inherited): rss · free

Copyright Industry Dive; no bulk scraping under terms of use.

Wire service with strong verification standards and genuine China and Japan bureaus, which most US trade press lacks. Because it is a wire, its stories are syndicated across dozens of outlets — those republications are one source, not many, and are the single most common way a Phase 2 pipeline will double-count.

Source of the China out-licensing analysis used here (2026-02-13). Its DealForma- and analyst-sourced figures on Chinese licensing are the best English-language series available. Deduplicate syndicated copies before counting sources.

Coverage: Global · decades

Access (inherited): public_page · enterprise

Copyright Thomson Reuters. Redistribution requires a licence; the public site restricts automated access. Reuters Connect is the licensed feed product.

Produced by a firm that sells transaction advisory services to the companies doing the deals, so it has a direct commercial interest in deal activity appearing robust and in the patent cliff appearing large. Deal-value definitions (announced versus closed, inclusion of contingent value) are not fully disclosed, and its totals diverge from STAT's and CNBC's.

Useful for structural framing (the $300bn-plus LOE exposure, the shift to mid-cap bolt-ons and CVR structures) rather than for point figures. Always name PwC and the date when citing the number, per this programme's rules on modelled estimates. J.P. Morgan's quarterly biopharma deal reports are the comparable bank-side series.

Coverage: US-focused · semi-annual editions back roughly a decade

Access (inherited): public_page · free

Free to read; copyright PwC; attribution required. No redistribution of the full report.

Written by practitioners for client acquisition, so they are fast, accurate on the legal text and biased toward emphasising complexity and compliance burden (which is what generates engagements). They rarely misstate an instrument, but they select which instruments to cover based on client demand.

The fastest route to understanding a new instrument — the Section 232 tariff structure, the BIOSECURE implementation timeline and the WuXi 1260H designation were all first traced through these. Always follow through to the Federal Register or statute before treating a detail as fact.

Coverage: US, with some EU coverage · FDA Law Blog back to 2007

Access (inherited): rss · free

Free to read; copyright the respective firms; attribution expected. Not legal advice.

Regulator-produced and definitional for US banking. No commercial interest. Covers only FDIC-insured institutions, so it structurally excludes the fastest-growing part of credit intermediation (private credit, fintech balance sheets, credit unions). Aggregates can mask distribution: a handful of GSIBs dominate the totals.

NOTE FOR PHASE 2: the QBP landing pages and PDF returned HTTP 403 to automated fetches during this research (2026-09-15). Use the FFIEC CDR bulk download or the FDIC BankFind API instead of scraping the QBP web pages. The press-release pages under /news/press-releases/ fetched fine.

Coverage: United States, all FDIC-insured depository institutions · back to 1984 via BankFind/Call Report data

Access (inherited): public_page · free

US Government work, public domain. Attribution expected. Bulk Call Report data available via FFIEC CDR.

Regulator statistical release, no commercial interest. Weekly estimates are sample-based and revised; the 'loans to non-depository financial institutions' line is the single most useful series for tracking bank-to-private-credit interconnection but was only broken out relatively recently, limiting history.

Pair with FRED series for programmatic access. The NDFI lending line is the leading indicator for T-07-03.

Coverage: United States · back to 1973

Access (inherited): api · free

Public domain US Government work; FRED redistribution permitted with attribution.

Built on an anonymised 5% sample of Equifax credit files. Consequently it captures only credit-bureau-reported debt: most BNPL is invisible to it, which systematically understates young and thin-file consumer leverage. No commercial bias.

PDF at a stable path pattern (/medialibrary/interactives/householdcredit/data/pdf/HHDC_YYYYQn.pdf) which fetches cleanly; the HTML landing page does not render data.

Coverage: United States, national and state-level · back to 1999Q1

Access (inherited): bulk_download · free

Free use with attribution to FRBNY Consumer Credit Panel/Equifax. Underlying microdata is licensed and not redistributable.

Primary filings, legally attested. Bias is selection: only registrants file, so private credit funds, non-traded vehicles and private fintechs are partially or wholly invisible. Management discussion is self-serving by construction; the audited numbers are not.

IMPORTANT ACCESS NOTE: /cgi-bin/browse-edgar is disallowed by robots.txt for agents, but efts.sec.gov/LATEST/search-index and data.sec.gov both work. Best channel for BDC 10-Qs (non-accrual and PIK disclosure) and bank 10-Qs.

Coverage: US registrants and foreign private issuers listed in the US · full text back to 2001; filings back to 1993

Access (inherited): api · free

Public domain. SEC requires a declared User-Agent and rate limiting to 10 requests/second.

Primary. The institutional voice shifts materially with the Vice Chair for Supervision, so speeches are policy signals, not neutral analysis; read dissenting governor statements alongside majority releases to see the actual spread of views.

The single highest-signal feed for T-07-06. Speeches by the Vice Chair for Supervision reliably precede proposals by weeks.

Coverage: United States · back to 1996

Access (inherited): rss · free

Public domain US Government work.

The authoritative legal text. No bias, but volume is high and legal text buries the calibration numbers deep in the impact analysis; the accompanying agency fact sheets are more usable but more spun.

Best single programmatic channel for GENIUS Act, Basel and 1033 rulemaking. Comment dockets on regulations.gov reveal industry positions before rules land.

Coverage: United States · back to 1994 online, 1936 in print

Access (inherited): api · free

Public domain. Well-documented free JSON API, no key required.

Primary regulator output. The enforcement-action database is the only public, systematic read on sponsor-bank/BaaS supervisory pressure. Chartering decisions are a live political instrument, so the flow of trust-charter approvals is a policy signal as much as a prudential one.

ACCESS NOTE: occ.gov returned a robots.txt fetch failure to automated agents on 2026-09-15; use the RSS feed rather than the HTML index. Enforcement-action search is the key asset for T-07-18.

Coverage: United States, national banks and federal savings associations · enforcement actions searchable back to the 1990s

Access (inherited): public_page · free

Public domain US Government work.

Primary but institutionally unstable: headcount fell from roughly 1,700 to roughly 1,300 by December 2025 and the bureau has been rescinding rather than issuing rules. Enforcement counts have collapsed, so an absence of actions is now a measure of the agency's capacity and politics, not of market conduct. Complaint data is self-selected and unverified.

Must now be monitored alongside state AG offices and state banking departments (see S-07-22), which have absorbed much of the enforcement function.

Coverage: United States · complaints back to 2011; enforcement back to 2012

Access (inherited): api · free

Public domain; complaint database has a published API with documented field-level caveats about unverified narratives.

Consensus product of national authorities, so it lags the market and is diplomatically hedged; where members disagree the report reports ranges rather than conclusions. That hedging is itself informative. No commercial interest. Data is member-supplied and therefore uneven in coverage across jurisdictions.

The 2026-05-06 private credit vulnerabilities report is the best single cross-jurisdiction dataset on bank-to-private-credit exposure. PDFs fetch cleanly.

Coverage: Global, G20 jurisdictions · back to 2009

Access (inherited): public_page · free

Free to reproduce with attribution; PDFs are openly downloadable.

Central-bankers' bank; institutionally sceptical of private stablecoins and non-bank credit, and that prior colours framing even where the data is neutral. Statistical series are the global gold standard for cross-border banking and payments.

ACCESS NOTE: guessed quarterly-review URLs 404. Use the RSS publications feed to discover the actual issue URLs rather than constructing them.

Coverage: Global · locational banking statistics back to 1977

Access (inherited): bulk_download · free

Free use with attribution; BIS Stats Explorer offers a documented SDMX API.

UK-centric, and the FPC has an institutional incentive to demonstrate vigilance, which can overweight tail risks. Offsetting virtue: it has been consistently earlier and more numerically specific than US regulators on non-bank leverage, hedge-fund positioning and AI-related valuation risk.

Non-US primary source with genuine analytical independence from the US regulatory cycle. Fetches cleanly at /financial-stability-report/YYYY/<month>-YYYY.

Coverage: United Kingdom with global market commentary · back to 1996

Access (inherited): public_page · free

Open Government Licence v3.0 - free reuse with attribution.

The ECB is the project's sponsor and advocate, so cost and adoption framing is favourable to issuance; the countervailing bank-cost estimates come from European banking associations with the opposite interest. Read both. Timelines are contingent on EU legislation the ECB does not control.

Rulebook version numbers are the real progress metric; draft v0.91 was released July 2026. Pilot targeted 2027, first issuance 2029 conditional on 2026 legislation.

Coverage: Euro area · back to 2020

Access (inherited): public_page · free

Free reproduction with attribution under ECB terms.

UK conduct regulator now under an explicit government competitiveness mandate, which biases rhetoric toward growth and away from restriction. Useful as the leading non-US indicator for BNPL regulation, open finance and AI supervision, all of which the UK has moved on earlier than the US.

CEO Nikhil Rathi's 'Rethinking regulation for the age of AI' is the clearest statement of a major regulator's agentic-AI posture as of 2026.

Coverage: United Kingdom · back to 2013 (FSA archive earlier)

Access (inherited): rss · free

Open Government Licence. Fetch /news, not /news/press-releases, which 404s.

Audited numbers are reliable; the framing is not. Both networks have a direct commercial interest in presenting account-to-account rails and agentic commerce as additive rather than substitutive, and in defining 'new flows' expansively. Volume growth is reported in constant dollars, which flatters in a weak-dollar period.

The cleanest quarterly read on whether A2A rails are actually denting card economics: watch payments volume growth vs processed transaction growth vs cross-border.

Coverage: Global · Visa back to its 2008 IPO; Mastercard to 2006

Access (inherited): public_page · free

Company-published; fair-use quotation with attribution. Q4CDN-hosted PDFs fetch cleanly at stable paths.

Commercial database selling data access; coverage is best in the US and thinnest in Asia and Africa. Its 'fintech' boundary excludes some AI-infrastructure companies that sell primarily to banks, so sector totals are not comparable to KPMG's. Announced rounds are counted, which inflates against closed-round measures.

Always pair with S-07-16. The Crunchbase/KPMG divergence documented in the macro brief recurs in the fintech-specific totals.

Coverage: Global, US-weighted · back to 2007

Access (inherited): api · freemium

Editorial content free to cite with attribution; underlying database requires a paid licence for bulk use and prohibits redistribution.

Produced by a Big-4 firm that sells advisory services to the companies it counts, and drawn from PitchBook data under licence. Includes PE and corporate rounds that Crunchbase often excludes, producing systematically higher totals. Treat the direction as robust and the level as definition-dependent.

Free PDF but no feed - Phase 2 needs a scheduled check of the landing page each quarter rather than a poll.

Coverage: Global with regional cuts · back to 2014

Access (inherited): public_page · free

Free PDF download; redistribution restricted, attribution required. Underlying PitchBook data is licensed.

Named-byline trade press with a genuinely deep cat-bond deal database, but it is advertising-funded by the ILS industry it covers and its editorial tone runs structurally bullish on ILS capacity. Its Deal Directory is nonetheless the most complete public record of cat-bond issuance anywhere.

Fastest public route to broker loss estimates (Gallagher Re, Aon, Munich Re) which are otherwise gated PDFs.

Coverage: Global reinsurance and ILS markets · cat bond deal directory back to 1996

Access (inherited): rss · free

Free to read and cite with attribution; the deal directory data is not licensed for bulk redistribution.

Produced by the world's second-largest reinsurer, which has a direct commercial interest in establishing that loss trends are rising and therefore that rates must not fall. Methodologically transparent and long-running, which makes it the reference series despite that interest. Its loss-inclusion threshold differs from Gallagher Re's, which reliably produces a several-billion-dollar gap.

Always report alongside the Gallagher Re and Aon estimates for the same period; the spread between them is a real finding, not noise.

Coverage: Global · sigma catastrophe series back to 1970

Access (inherited): public_page · free

Free download; reproduction permitted with attribution to Swiss Re Institute sigma.

A commercial data vendor selling CRE analytics; headline delinquency rates are published free as marketing while the loan-level data sits behind an enterprise licence. Covers securitised CRE only, so it misses the bank-held CRE book that matters most for regional-bank credit risk.

Cross-check bank-held CRE against FDIC Call Report noncurrent CRE, which moves differently from securitised CMBS.

Coverage: United States CMBS universe · back to the late 1990s

Access (inherited): public_page · enterprise

PAYWALL WARNING FOR PHASE 2: headline monthly rates are free and citable; loan-level and historical series require an enterprise licence. Budget for this if CRE credit is a tracked theme.

The only mandatory, auditable public window into private credit quality. Bias is scope: it covers registered BDCs and interval funds, not private funds, so it is a biased-clean sample of a mostly opaque market. Fair-value marks are manager-determined and quarterly, so deterioration shows up late.

The three metrics to extract each quarter: non-accruals as a percentage of fair value, PIK as a percentage of total investment income, and tender-offer fill rates versus the 5% NAV cap. Fill rates below 100% are the gating signal.

Coverage: United States · back to the mid-1990s

Access (inherited): api · free

Public domain; SEC rate limits apply.

Commercial database; the free research notes are marketing for the licensed product. Coverage of evergreen and semi-liquid vehicles is the best available but is still self-reported by managers and incomplete outside North America and Western Europe. It also underpins KPMG's venture figures, so citing both is not independent triangulation.

Source of the $607bn / 567-fund evergreen universe figure as of 2026-03-31, usually surfaced first through trade press rather than directly.

Coverage: Global, North America and Europe weighted · back to the mid-2000s

Access (inherited): licensed · enterprise

PAYWALL WARNING FOR PHASE 2: free summary reports are citable; the database requires an enterprise licence and forbids redistribution. Note the shared-lineage problem when triangulating against KPMG.

Fragmented across fifty-plus offices with no common schema, and actions are politically timed and press-released for maximum effect. Now indispensable because it is where consumer-finance enforcement actually happens following CFPB retrenchment; the fragmentation is the cost of the story, not a reason to skip it.

Priority states as of 2026: New York (FAIR Act, NYDFS), California (DFPI), Massachusetts, Pennsylvania, Colorado, New Jersey, Connecticut. Several have hired former CFPB enforcement staff directly.

Coverage: United States, state by state · varies by state, typically 10-20 years

Access (inherited): rss · free

Public records; free to cite. Each state has its own terms.

NPCI both operates UPI and reports its statistics, so there is an institutional interest in headline growth; volumes are nonetheless reconciled against RBI settlement data. Counts include very low-value person-to-person transfers, so transaction counts overstate commercial payment displacement relative to value.

Essential regional counterweight to US-centric payments coverage. Pair with RBI Payment System Indicators for a fetchable alternative.

Coverage: India · back to UPI launch in 2016

Access (inherited): public_page · free

Publicly published statistics. ACCESS BLOCKER: npci.org.in is disallowed by robots.txt for automated agents - Phase 2 should use RBI's monthly payment system indicators or an official data partner instead.

Central-bank-operated scheme reporting on itself; the BCB has publicly championed Pix as industrial policy against card incumbents, which colours framing but not the settlement counts. Data is authoritative because the central bank is the settlement agent.

Brazil is the furthest-advanced large market on account-to-account displacement of cards; it is the natural leading indicator for T-07-14.

Coverage: Brazil · back to Pix launch in November 2020

Access (inherited): bulk_download · free

Open public data; BCB provides an open-data API (dadosabertos.bcb.gov.br). ACCESS NOTE: the HTML statistics page is JavaScript-rendered and returns no data to a plain fetch - use the open-data API endpoints.

Explicit advocacy on behalf of large banks - every figure is produced to support a policy position, and deposit-flight and lending-contraction estimates in particular are advocacy numbers, not forecasts. Valuable precisely because they state the industry's position on the record, early, and in dated documents, and because comment letters reveal what banks are actually worried about before it reaches the press.

The Clearing House is simultaneously a trade body and the operator of CHIPS and RTP and the shared tokenised-deposit network, so its output is both advocacy and operator disclosure. Separate the two when citing.

Coverage: United States · BPI back to 2018 (predecessor bodies earlier)

Access (inherited): rss · free

Free to read and cite with attribution. Label all quantitative claims as advocacy estimates, never as facts.

Official announcements only; reports obligated award value at announcement, not appropriation or delivery. Ceiling values on IDIQ awards are stated without noting that nothing is obligated — the single largest source of overstatement in this sector if read naively.

The single most important feed for this sector. Parse for awardee, ceiling vs obligated amount, and contract type — the ceiling/obligation distinction must be modelled explicitly.

Coverage: US DoD worldwide · archive back to the 1990s

Access (inherited): public_page · free

US Government work, generally public domain; no redistribution restriction

Pre-solicitation and solicitation notices show intent, not funding. Classified and quick-turn OTA awards are absent, which systematically under-represents the defence-tech cohort.

Pair with USAspending.gov to get obligations rather than announcements. This pairing is how to separate announced from obligated.

Coverage: US federal · back to 2000 via FPDS linkage

Access (inherited): api · free

Free API key required; rate-limited; US Government data

Authoritative on obligations and outlays but lags announcements by weeks; classified programmes are aggregated or absent.

The only public source that lets you answer 'was this appropriated, obligated, or merely announced'. Essential for this sector given the ceiling-value problem.

Coverage: US federal · back to FY2008

Access (inherited): api · free

Public domain US Government data

Company-authored and legally constrained; management discussion is promotional at the margin but numbers are audited. Covers only US-listed issuers, so Airbus, Rheinmetall, Thales, Hanwha and KAI are outside it.

Newly critical for this sector: SpaceX (SPCX) began filing in 2026, converting the sector's largest private information asymmetry into quarterly Tier A data.

Coverage: US-listed issuers · back to 1994

Access (inherited): api · free

Public domain; fair-access rate limits and declared user-agent required

Company-authored. Segment definitions (Space / Connectivity / AI) are management's own and were set at IPO; they can be redefined, which would break time series. Non-GAAP adjusted EBITDA is presented prominently alongside GAAP losses.

Highest-value new source in this sector. Provides the first audited Starlink subscriber, ARPU, launch-cadence and Starshield revenue figures ever published. Feed URL is inferred from the standard Q4 IR platform pattern and should be confirmed at ingest.

Coverage: Global · from Q2 2026 (first quarter as a public company)

Access (inherited): public_page · free

Standard corporate copyright; quotation with attribution

Company-authored. Production rate statements describe authorised or momentarily achieved line rates and diverge persistently from delivered aircraft — the two must be tracked as separate series.

Pair with the Airbus O&D page. Track announced rate, rollouts and deliveries as three distinct series.

Coverage: Global · orders and deliveries back to 1955

Access (inherited): public_page · free

Corporate copyright; data tables downloadable for analysis

Company-authored; heavy December delivery seasonality makes any mid-year run-rate extrapolation misleading in Airbus's favour or against it depending on month.

Monthly XLS download is the cleanest structured commercial-aerospace dataset available free. Compare YTD against stated annual guidance to detect guidance risk early.

Coverage: Global · back to 1974

Access (inherited): bulk_download · free

Corporate copyright; downloadable XLS for analysis

Self-reported by member states to a common NATO definition, which now spans core defence (3.5%) and a broad defence-related category (1.5%) that includes infrastructure and resilience spending. The 1.5% category is elastic and comparisons across states are weaker than they look.

Published as PDF plus Excel tables. Essential for the appropriation side of the appropriation-to-contract question, but says nothing about obligations or deliveries.

Coverage: 32 NATO members · back to 1949 in the long series; consistent definitions from 2014

Access (inherited): bulk_download · free

NATO public information; attribution requested

Institutional and policy-promotional in tone; publishes approvals and allocations but not, on public pages, actual disbursed amounts or contracted volumes — the gap where the appropriation-to-contract question actually lives.

Track SAFE loan agreement signatures and the 35% non-EU content rule's application. Actual disbursement figures are the key missing datum and may require Commission implementing acts or Court of Auditors reports.

Coverage: EU-27 plus Ukraine, EEA/EFTA participants · SAFE from May 2025

Access (inherited): rss · free

EU public documents, reuse permitted with attribution

Selection bias by design — the IG audits where problems are suspected, so the corpus over-represents failure. That is exactly what makes it the best corrective to programme-office optimism.

The highest-signal source for the gap between funded targets and delivered hardware — e.g. the July 2026 finding of ~36,000 155mm rounds/month against a 100,000 target.

Coverage: US DoD · back to 2000s

Access (inherited): rss · free

US Government work; public domain

Non-partisan and methodologically explicit, but reports lag events by months to years; the annual Weapon Systems Annual Assessment is a snapshot, not a leading indicator.

Also the authoritative source for bid-protest decisions, which are the leading indicator for whether mega-IDIQ and enterprise-agreement structures survive legal challenge.

Coverage: US federal · back to 1990s online

Access (inherited): rss · free

US Government work; public domain

Non-partisan and unusually careful about distinguishing requested, authorised, appropriated and obligated amounts — rare and valuable in this sector.

ACCESS CAVEAT: congress.gov disallows automated retrieval via robots.txt. Phase 2 must ingest via the CRS PDF endpoints or a licensed mirror rather than scraping; budget for this.

Coverage: US federal · back to 2018 for the public site; older reports via third-party archives

Access (inherited): public_page · free

US Government work; public domain

Regulatory record only — confirms licences, authorisations and mishap investigations but gives no commercial or schedule information. Mishap closure notices are terse and often lag the operator's own statement.

The authoritative check on return-to-flight claims — e.g. Blue Origin's post-May-2026 New Glenn schedule. Also carries eVTOL powered-lift certification progress.

Coverage: US launch and reentry · licensing records back to 1989

Access (inherited): public_page · free

US Government work; public domain

Applications state intent and maximum requested scope, not committed deployment — a 5,105-satellite filing is a request, not a constellation. Milestone deadlines in grants are the enforceable part.

The leading indicator for constellation and direct-to-device competition, including spectrum transactions such as the SpaceX-EchoStar deal and Amazon's D2D application.

Coverage: US market access, global constellations · back to the 1990s

Access (inherited): api · free

US Government work; public domain

Increasing share of the relevant architecture (MILNET, Space Data Network) is classified, so the public record systematically understates both spend and vendor concentration.

Watch for tranche restructuring. The FY2027 elimination of Tranche 3 transport in favour of the SDN is the key structural signal for satellite manufacturers.

Coverage: US · SDA from 2019

Access (inherited): public_page · free

US Government work; public domain

Company-authored. Reports a 'nomination backlog' that includes framework-agreement volumes not yet firm orders — a broader definition than a classic order backlog, and it should not be compared like-for-like with US primes' backlog figures.

The best single public proxy for whether European defence budgets are converting into signed European industrial orders. Read the backlog definition carefully.

Coverage: Europe and export markets · back to the 1990s

Access (inherited): public_page · free

Corporate copyright; quotation with attribution

Represents airlines against OEMs and lessors. Its MRO and parts-access work is methodologically solid but has an explicit commercial objective — loosening OEM aftermarket control. Read the asks as advocacy, the data as reliable.

Headline studies are free; the underlying economics datasets behind them are paid. Budget for IATA data licensing if Phase 2 needs the granular fleet and MRO series.

Coverage: Global, ~300 member airlines · back to 2000s online

Access (inherited): rss · free

Attribution required; some underlying datasets are licensed separately and paid

Named-byline trade journalism with strong sourcing into US civil and military space; advertising relationships with the companies it covers; lighter on Asian and commercial-Chinese space.

Best single trade source for satellite operator strategy and launch-market structure. NOTE: automated fetching of some article paths is disallowed; use the RSS feed rather than scraping.

Coverage: Global, US-weighted · back to 1989

Access (inherited): rss · freemium

Some content paywalled; robots.txt restricts automated retrieval on some paths — check before ingest

Named-byline defence trade press with good Pentagon access; heavily sponsored by primes and event revenue, and reliant on programme-office briefings, which skews toward official framing of schedule and progress.

Fast on budget-document analysis. Corroborate programme figures against the budget justification books before treating them as fact.

Coverage: US, Europe, Indo-Pacific · back to 2011

Access (inherited): rss · free

Standard copyright; attribution and linking

Counts announced rounds, so it systematically leads closed capital and can double-count extensions. Valuations are frequently reported inconsistently (pre- vs post-money) — the Anduril Series H is a live example with USD 30.5bn and USD 61bn both in circulation.

Use for funding velocity only. Never use funding totals as evidence of programme capture — cross-check every defence-tech company against DoD contract announcements and USAspending.

Coverage: Global, US-weighted · back to 2007

Access (inherited): rss · freemium

Editorial content free with attribution; the underlying database API is paid and redistribution is restricted

Independent and methodologically transparent, but publishes with a long lag (typically 12-16 months) and uses constant-price conversions that differ from national budget documents. Arms-transfer values are trend-indicator units, not sale prices — a very common misquotation.

The only genuinely global, independent series on arms production and transfers. Essential for non-US and non-European coverage, including Korea, Japan, Türkiye, India and China.

Coverage: Global · military expenditure back to 1949; arms transfers back to 1950

Access (inherited): bulk_download · free

Free for non-commercial use with attribution; commercial redistribution requires permission

Korean-origin national business daily; coverage of K-defence exports carries a national-champion framing and reports negotiations and MoUs alongside firm contracts without always distinguishing them.

Required non-English-origin regional source for this sector. Korea is now a top-tier arms exporter and Western trade press covers it with a lag. Cross-check export totals against DAPA and SIPRI — secondary aggregators circulate an unsourced USD 37bn 2026 figure that does not reconcile with official data.

Coverage: South Korea, Korean export markets · English edition from the 2010s

Access (inherited): rss · free

Standard copyright; translated content, attribute to original

Strong on budget-justification document analysis; funded by government-technology advertising and events, so coverage tilts toward programmes with active vendor marketing.

Best fast read of the annual budget request's space and autonomy lines, including the Space Data Network and MILNET disclosures.

Coverage: US · back to 2022

Access (inherited): rss · free

Standard copyright; attribution and linking

Independent academic institute; the Ukraine tracker distinguishes committed, allocated and delivered aid, which is unusually rigorous and directly transferable to the appropriation-to-delivery problem across this whole sector.

Adopt its committed/allocated/delivered taxonomy as the default schema for every defence-spending claim in this sector.

Coverage: Europe and Ukraine support · tracker from 2022

Access (inherited): bulk_download · free

Open access with attribution; CC licensing on most datasets

None material. Nominal dollars, not deflated in the headline, so real declines are understated in periods of construction inflation (3.3% y/y in March 2026). Subject to substantial revision for two months. Private manufacturing series excludes publicly funded plant.

The single most important discipline source for this sector. This is where reshoring announcements meet concrete poured. July 2026 private manufacturing $167.8bn SAAR, -21.7% y/y. Poll the subsector detail (computer/electronic/electrical versus chemical, food, transportation equipment), not just the headline — the headline decline is almost entirely an electronics phenomenon.

Coverage: United States, with experimental subnational private-manufacturing detail · back to 1993 for the current series definitions

Access (inherited): bulk_download · free

US Government work, public domain. Attribution to US Census Bureau expected. No redistribution limits.

Voluntary panel, not a probability sample — Census states statistical significance is not measurable. Aircraft and defence orders create large lumpy swings, which is why core capital goods (nondefense ex-aircraft) is the series to watch.

Core capital goods orders are the real-time proxy for equipment capex — the second leg of the announcement-versus-activity test after construction. July 2026: core capital goods $85.9bn, +0.2% m/m; machinery orders $44.8bn, +1.2%.

Coverage: United States · back to 1992 on the current NAICS basis

Access (inherited): bulk_download · free

US Government work, public domain.

None material. Annual benchmark revisions can shift the level meaningfully. Measures physical output, so it will diverge from nominal sales series during tariff-driven price inflation — that divergence is a feature, not a flaw.

The arbiter when ISM sentiment and physical output disagree. July 2026: IP +0.2% m/m, +1.1% y/y; manufacturing capacity utilisation 76.0%, 2.2pp below its 1972-2025 average. Available via FRED API as well.

Coverage: United States · back to 1919

Access (inherited): bulk_download · free

US Government work, public domain.

Country attribution follows last country of substantial transformation, so it cannot distinguish genuine relocation from transshipment — the central measurement problem in this sector. Non-monetary gold flows distorted 2025 and therefore distort 2026 year-over-year comparisons.

Exhibits 14-18 carry the country and commodity detail that actually matters. July 2026: Vietnam deficit $23.3bn and Taiwan $18.1bn both exceeded China at $15.2bn. Pair with Chinese customs data to measure the export-record-versus-import-record gap.

Coverage: United States, with full country and HS-code detail · detailed HS-level data back to 1989

Access (inherited): api · free

US Government work, public domain.

Critically, import price indexes EXCLUDE duties. They therefore measure exporter pricing behaviour, not landed cost, and systematically understate tariff-inclusive inflation. Anyone using them to argue tariffs are not being passed through is misreading the definition.

The country-of-origin cut is the useful part: July 2026 nonfuel import prices +4.5% y/y overall, China-origin +2.7% y/y. Pair with the exclusion caveat above and with CBP duty collections to reconstruct true landed cost.

Coverage: United States, with locality-of-origin detail (China, EU, Japan, Mexico, Canada) · back to 1982 for aggregate series

Access (inherited): api · free

US Government work, public domain.

Nominal dollars, so the inventories-to-sales ratio falls mechanically when prices rise faster than volumes — a real risk in a tariff-inflation period. Treat a falling I/S ratio as ambiguous between genuine destocking and price effects.

The test for whether tariff stockpiling is actually happening. June 2026: total business I/S ratio 1.30, down from 1.39 a year earlier; manufacturers 1.48 from 1.57. The data does not support a general stockpiling narrative.

Coverage: United States · back to 1992

Access (inherited): bulk_download · free

US Government work, public domain.

A diffusion index of direction, not magnitude, from a purchasing-manager panel — it can print 54 on broad but shallow improvement while physical output grows ~1%. Panel composition favours larger firms. The free ISM site blocks automated fetching; the PR Newswire distribution of the same release is machine-readable.

The verbatim panellist comments are the highest-value part and the most underused: August 2026 had tariffs in 29% of negative comments, supply chain and lead times in 46%, and the Iran conflict in 30%. Prices index 71.1% and Customers' Inventories 42.8% are the two subindices that matter most right now.

Coverage: United States · back to 1948

Access (inherited): public_page · freemium

Headline figures are freely quotable with attribution; full historical series and detailed industry data are licensed. Robots.txt restricts automated access to ismworld.org — use the press-release distribution instead.

Assessed spot rates on eight major East-West routes; does not reflect long-term contract rates, which is where most volume actually moves. Drewry sells consulting into the same market it assesses. Robots.txt blocks automated fetching of the main page.

Composite $4,476/40ft on 2026-09-10 versus $2,712 on 2026-05-21 — the volatility is the story. Cross-check against Freightos FBX and Xeneta; divergence between spot indices and contract benchmarks is itself a signal. PAYWALL WARNING: the full dataset is a licensing line item.

Coverage: Global East-West container trades · back to 2011

Access (inherited): public_page · freemium

Headline composite freely quotable with attribution; full route history and the Container Forecaster are paid. Automated scraping is restricted by robots.txt — Phase 2 should budget for a licence or rely on secondary republication.

Derived from one company's freight-payment transaction base, so it reflects Cass's customer mix — skewed toward large shippers and North American domestic freight. The expenditures index includes fuel, which inflated 2026 growth substantially (diesel $5.13/gallon, fuel surcharges +52% y/y).

The most important single freight datapoint of 2026: August shipments +2.1% y/y, the first annual gain since January 2023, ending a 42-month downturn. Use the truckload linehaul index (153.9, +11.3% y/y) rather than expenditures to strip out fuel.

Coverage: United States, primarily domestic freight · back to 1990

Access (inherited): public_page · free

Free to read and cite with attribution. Published with ACT Research commentary.

Reflects the spot market only, which is roughly 10-20% of truckload volume, so it overstates volatility relative to the contract market DAT also publishes. DAT operates the load board it measures, a genuine commercial interest.

The fastest read on the trucking capacity shock: week of 26 Jul-1 Aug 2026 dry van $2.96/mile (+66c y/y), reefer $3.35, flatbed $3.60, with equipment posts down 11-12%. Load-to-truck ratios (dry van 10.9, reefer 19.4) lead rates by 2-4 weeks.

Coverage: United States, with lane and regional detail · back to 2010 for national averages

Access (inherited): public_page · freemium

Trendlines headline data is free with attribution; lane-level and historical API access is paid (enterprise tier).

Covers transpacific containerised trade only; East Coast and Gulf ports must be added for a national picture. Ports have a mild promotional interest in volume narratives. Published with roughly a three-week lag.

August 2026: loaded imports 500,301.50 TEU (-0.84% y/y), loaded exports 115,561.25 TEU (-9.28%), CY2026 to date +1.51%. The export weakness is the underreported half. Pair with the Port Optimizer signal for a two-to-three-week forward view.

Coverage: San Pedro Bay complex; comparable data available from NY/NJ, Savannah, Houston · back to 1995

Access (inherited): public_page · free

Public agency data, free to use with attribution.

Non-partisan university modelling. Distinguishes observed effective rates from pre-substitution rates, which is exactly the distinction most commentary elides. Its 'effective' definition differs from Tax Foundation's 'applied' rate — do not mix the two.

The most reliable single number for 'what is the US tariff rate actually': 6.7% effective in July 2026 versus 2.3% in January 2025. Also tracks refunds, which matter enormously post-IEEPA: ~$107bn certified by CBP as of 2026-08-21 against ~$298.5bn gross collections.

Coverage: United States · tariff series from 2024

Access (inherited): public_page · free

Free to cite with attribution; academic use terms.

None — this is the primary legal text. The difficulty is volume and the fact that the operative detail lives in annexes of HTS codes rather than in the preamble.

The only authoritative source for what is actually tariffed. 2026 alone produced: IEEPA revocation, Section 122 imposition (24 Feb) and expiry (24 Jul), Section 232 metals modification to full customs value (6 Apr), Section 301 forced-labour tariffs on 60 economies (24 Jul), pharmaceuticals at 100% (31 Jul), Section 338 on Canada at 50% (15 Sep). Law-firm alerts are faster to read but always verify the annex.

Coverage: United States · back to 1994 online, 1936 in scanned form

Access (inherited): api · free

US Government work, public domain. Full-text API with no rate-limit concerns for reasonable use.

Primary filings, but the applicant evidence within them is advocacy — the $3.5bn shipper savings and 2.1m truckload figures are applicant projections. Shipper, labour and state filings on the same docket are the counterweight.

The evidentiary record for the first transcontinental Class I merger, accepted 2026-05-28 with a decision expected around mid-2027. Support-versus-opposition filing counts from major shippers are the best leading indicator of the outcome. Also carries STB's weekly rail service performance data.

Coverage: United States rail network · back to 1996 (ICC records earlier)

Access (inherited): public_page · free

US Government work, public domain.

Industry association with an advocacy interest in rail's competitive position. The underlying carload and intermodal counts are carrier-reported and reliable; the framing is not neutral. The data centre page renders through interactive charts, so the weekly PDF press release is the fetchable artefact.

Weekly carloads by commodity are the highest-frequency read on physical industrial activity in North America — chemicals, metals, aggregates and intermodal each tell a different story. Phase 2 should target the weekly PDF, not the JavaScript dashboard.

Coverage: United States, Canada and Mexico Class I railroads · back to 1988 for weekly series

Access (inherited): public_page · free

Free to cite with attribution; Rail Time Indicators is a paid publication.

Conducted by an advocacy organisation among its own membership, which skews to larger US manufacturers and to firms with a policy interest. The quarter-on-quarter movement in the challenge rankings is more informative than the levels.

The 25.6-point single-quarter jump in raw-material costs as a top challenge (57.5% to 83.1%, Q1 to Q2 2026) is the kind of movement that would be invisible in a price index. Also tracks expected sales, production, capex and employment growth, which can be compared against subsequent Census outturns.

Coverage: United States · back to 1997

Access (inherited): public_page · free

Free to cite with attribution; full write-ups published as PDFs.

Trade association with an advocacy interest in German industrial policy, but the order-intake data is reported in real (inflation-adjusted) terms, which is methodologically stronger than most US equivalents. Domestic-versus-foreign split is the analytically useful cut.

The best non-US leading indicator of global factory capex, because German machinery builders supply the world's plants. July 2026 orders +2% y/y real with weak domestic and stronger international demand; 2026 production forecast cut to zero growth. REGIONAL SOURCE — satisfies the non-US coverage requirement for this sector alongside Eurostat and the Bank of Korea.

Coverage: Germany, with regional state detail and export-market breakdown · long-running monthly series, decades

Access (inherited): public_page · free

Free to read and cite with attribution; some member content restricted.

None material. Member-state data quality varies and Ireland's figures are heavily distorted by multinational contract manufacturing, which can move the euro-area aggregate by itself.

May 2026: euro area -0.2% m/m and -1.2% y/y, Germany -0.8% y/y, France +3.0%, Italy +1.1%; capital goods +2.6% y/y but non-durable consumer goods -10.7%. The SDMX API is well-documented and stable. REGIONAL SOURCE.

Coverage: European Union and euro area, with member-state detail · back to 1991 for most member states

Access (inherited): api · free

Free reuse under the Commission's reuse policy with attribution.

National statistical authority with a clear state interest in the narrative around export resilience and rare-earth controls. Systematically diverges from partner-country import records — Rhodium documented an 18% decline in US records against 9% growth in Chinese records for the same flows. Use as one side of a two-sided comparison, never alone.

Essential for rare-earth and magnet export volumes by destination, which is where Chinese licensing discretion becomes visible — Japan at zero in July 2026 while South Korea exceeded pre-control levels. Silverado Policy Accelerator and DIGITIMES republish derived series in English, which is often more usable than the primary pages.

Coverage: China, with partner-country and HS-code detail · long-running, with periodic definitional changes

Access (inherited): public_page · free

Official statistics, free to cite. English pages lag the Chinese originals.

Airline trade association; its financial forecasts are produced for an industry audience and have historically been revised substantially within a year. The traffic statistics themselves are compiled from carrier reporting and are reliable.

The 2026 forecast history is itself a useful cautionary record: 2.6% growth in March, cut to 0.2% in June after the Hormuz closure. Volume and revenue decoupled — tonnage flat, yields +6.5%, revenue $162bn. Track the pressroom RSS rather than the publications store.

Coverage: Global, with six regional carrier groupings and major trade lanes · back to 1990 for CTK series

Access (inherited): public_page · freemium

Press releases and headline figures free with attribution; detailed monthly analysis and the CargoIS dataset are paid. PAYWALL: full lane-level data is a licensing line item.

Self-reported and 'using AI' is loosely defined, so it may understate AI embedded in purchased equipment and overstate casual use. Biweekly cadence produces noisy short-run movement. Nonetheless it is the only probability-sample measure of AI adoption by sector that exists.

The antidote to industrial-AI marketing: roughly 12% of manufacturing firms reported using AI in April 2026 against nearly 40% in information services, and 96% of AI-using firms reported no employment change. Whenever a vendor cites a 50-70% manufacturing AI adoption figure, this is the series that refutes it.

Coverage: United States, by sector, firm size and state · back to 2022; the AI module from late 2023

Access (inherited): api · free

US Government work, public domain.

Management framing is promotional and segment definitions are chosen by the issuer — Siemens' 'industrial AI' growth is substantially data-centre electrical and software demand, which the segment reporting does not separate. Guidance is an opinion; reported figures are auditable fact. Non-US issuers (Siemens, Maersk, Hapag-Lloyd, Lynas) do not file with EDGAR and require their own IR pages.

Book-to-bill and backlog are the sector's best forward indicators — Siemens hit 1.22 and a record €124bn backlog in Q2 FY2026. Kinaxis's disclosure that ~10% of its installed base is on paid or trial AI modules is the rare quantified adoption figure. Earnings-call transcripts are where quantified tariff cost impacts appear before they appear anywhere else.

Coverage: Global issuers; EDGAR covers US registrants only · EDGAR full text back to 2001, filings back to 1993

Access (inherited): api · free

EDGAR is public domain with a fair-access rate limit (10 requests/second) and a required user-agent header. Individual IR pages carry their own terms; several (Rockwell, Caterpillar) redirect or block automated fetching.

Derived from AIS vessel tracking with modelled trade-volume estimation, so port-call counts are observational while trade values are estimates. The web front end is an ArcGIS application that does not render to plain text, so the datasets must be pulled directly rather than scraped from the page.

The authoritative independent measure of chokepoint transits, and the right answer to the weakest evidence in this dossier — Hormuz transit volumes were only obtainable this cycle from a Tier-C tracker with no named author. Phase 2 should wire the chokepoint datasets directly and never rely on secondary reporting for transit counts again.

Coverage: Global — roughly 1,400 ports and major chokepoints including Suez, Bab el-Mandeb, Hormuz and Panama · daily from 2019

Access (inherited): bulk_download · free

Open data under IMF terms with attribution; no redistribution restrictions for analysis.

Self-selected sample of a trade publication's own readership (166 respondents in 2026), conducted in partnership with a vendor-funded association (MHI) and a consultancy. Systematically biased toward operators already interested in automation. Directionally useful year over year because the methodology is consistent; the absolute levels are not population estimates.

The 2026 edition: 52% using at least one robot type (up from 48%), only 3% with no plans (down from 9%), 29% funding via robotics-as-a-service, and labour cost ranked first by 67%. The RaaS share is the number to track — it is how automation gets financed at a 3.50-3.75% policy rate. This covers demand-side adoption; robot vendors themselves belong to sector 13.

Coverage: Primarily United States · fifth annual edition as of 2026

Access (inherited): public_page · free

Free to read and cite with attribution; registration may be required for the full report.

A think tank with a stated preference for lower and simpler taxes, which colours framing though not the underlying compilation. Its 'applied' rate (11.8% for 2026) is trade-weighted and statutory and is NOT comparable with Penn Wharton's 'effective' rate (6.7% in July 2026) — the two are routinely conflated in press coverage.

The best single running inventory of which authority covers what: Section 232 steel/aluminium/copper at 50%, autos and heavy trucks 25%, pharmaceuticals 100% from 2026-07-31, polysilicon 15% from 2026-12-04; Section 301 forced-labour 10-12.5% on 60 economies; Section 338 Canada 50% from 2026-09-15. Estimates new tariffs cover 54% of 2026 US goods imports.

Coverage: United States, by tariff authority and trading partner · from 2018

Access (inherited): public_page · free

Creative Commons attribution; free to reuse with credit.

Published by the European OEM lobby. Compiled from national registration authorities so the underlying counts are reliable, but framing and commentary consistently argue the industry's political case on CO2 targets, tariffs and regulatory burden. Note also that it reports by manufacturer group, not by ownership nationality, so 'Chinese brand share' cannot be read directly off it.

Publishes around the 20th-24th of the following month and takes a SUMMER BREAK - there is no July release, and August 2026 data was scheduled for 2026-09-24. Any pipeline must model this gap explicitly or it will silently report stale European data through late September. PDF press releases are also posted at acea.auto/files/ with a predictable naming pattern.

Coverage: EU, EFTA and UK, with country-level detail · monthly series back to the 1990s; current online archive several years deep

Access (inherited): public_page · free

Free to view and quote with attribution to ACEA. Data compiled from national authorities; redistribution of full datasets is restricted. No public API, so any automated collection must respect the site's terms and rate limits.

Industry association closely aligned with Chinese industrial policy. Headline framing emphasises growth and NEV leadership. Export definitions have changed historically (treatment of used-vehicle exports, CKD kits and parallel exports), so long time series require methodology checks. Primary site is Chinese-language.

THE single most important series in this sector - China is where the marginal vehicle is now produced and exported. The official site served a stale index on 2026-09-15 (most recent item June 2025), so a Phase 2 pipeline should treat Gasgoo, CnEVPost and Xinhua as reliable same-week intermediaries and reconcile back to CAAM when the site updates. Release is usually around the 10th-12th of the following month.

Coverage: China national, with NEV and export splits · monthly series back to the early 2000s

Access (inherited): public_page · free

Chinese-language public releases; reuse generally permitted with attribution. Detailed datasets are sold to members. Check the site's terms before automated collection.

CPCA measures domestic retail, which diverges sharply from CAAM's wholesale/shipment series because the latter includes exports. In 2026 that difference is the whole story: CAAM shows a roughly flat market while CPCA shows NEV retail down 10.1% YoY in August with eight consecutive monthly declines. CnEVPost is an independent English-language intermediary with no obvious commercial stake, but is still secondary to CPCA itself.

Pair with CAAM rather than substituting for it - the wholesale/retail gap is the key analytic signal for China in 2026. CnEVPost also republishes SNE Research battery share tables monthly, making it a two-for-one feed.

Coverage: China domestic retail, with brand-level detail · CnEVPost archive from roughly 2020; CPCA series much longer

Access (inherited): rss · free

CnEVPost content is copyrighted; quote with attribution and link. CPCA's own detailed data is sold. Do not bulk-republish.

Commercial data vendor whose customers are dealers, lenders and OEMs. Framing tends to be constructive about market conditions and about dealer economics. Its own full-year forecast (15.8m for 2026) sat in tension with its own August SAAR print (16.8m) as of 2026-09-15, which is a reminder to read the series rather than the headline.

The most reliable free read on US vehicle affordability. Publishes ATP around the 10th of the following month. Includes EV-specific ATP and incentive share, which is the cleanest public view of EV price competitiveness after the credit expiry.

Coverage: United States · ATP series back well over a decade; Manheim index since 1995

Access (inherited): public_page · free

Free press releases, free to quote with attribution to Kelley Blue Book or Cox Automotive. The underlying transaction datasets are licensed and expensive. Do not scrape the paid dashboards.

US federal statistical agency with no commercial interest. Vehicle powertrain-share analyses rely on licensed third-party data (Omdia, Wards), so EIA is authoritative for the framing but the underlying counts are a vendor's. Coverage is US-only.

The single best Tier-A free source for US powertrain mix. Its July 2026 piece established that 2026 is the first year of declining US BEV sales and share - a load-bearing fact for the whole sector. Full API plus RSS makes it trivially pollable.

Coverage: United States · Today in Energy archive back to 2009; underlying energy series to the 1940s

Access (inherited): api · free

US government work, generally public domain, free to reuse with citation. The EIA Open Data API has documented terms and rate limits; third-party licensed data embedded in articles may carry its own restrictions.

Regulator-collected operator filings, so the incentive to overstate is limited by legal exposure. However, public versions are REDACTED after a confidentiality review, and operators litigate over what is commercially sensitive - so absence of a figure is not absence of the underlying activity.

THE highest-value robotaxi source in existence: quarterly vehicle miles travelled during passenger service, EMPTY MILES, idle/dwell time and occupancy, per operator. This is the only public dataset that lets you compute deadhead share and therefore assess robotaxi unit economics. Q2 2026 filings: Waymo under deployment; Nuro, Tensor, Waymo, WeRide and Zoox under pilot. Requires ZIP extraction and parsing, so budget engineering time.

Coverage: California only, but California is the only jurisdiction publishing robotaxi passenger miles at all · from 2019 for pilot programmes; commercial deployment from 2023

Access (inherited): bulk_download · free

California public records, free to download and analyse. Filings are distributed as ZIP archives of CSV/PDF. Attribution to CPUC expected; check redaction notices before republishing operator-level detail.

Self-reported by permit holders using inconsistent internal definitions of 'disengagement'. The DMV itself states the reports are NOT designed for comparative analysis across companies - a warning routinely ignored in press coverage. Coverage also shrank as operators moved testing to Texas and Arizona, which have no equivalent requirement.

IMPORTANT CAVEAT for Phase 2: on 2026-09-15 the public disengagement-reports page served 2021-vintage CSVs, while the DMV's February 2026 press release cited aggregate 2024-25 mileage of over 9 million miles. The per-company files for recent periods may require a request to AVarchive@dmv.ca.gov. Do not assume the linked CSVs are current.

Coverage: California only · annual CSVs from 2015

Access (inherited): bulk_download · free

California public records, free. CSV downloads. Attribution to California DMV; the DMV's comparative-analysis caveat should be reproduced whenever company figures are ranked.

Operator-published and therefore self-interested: Waymo chooses the benchmark construction, the severity definitions and the comparison geography. The figures are grounded in NHTSA Standing General Order filings, which is a real constraint, but IIHS has publicly stated that data-collection regime is inadequate for monitoring scaled deployment. Treat every percentage as a claim requiring a stated definition.

Provides cumulative rider-only miles broken out by metro - 220.6m through March 2026, of which Phoenix 80.6m and Atlanta only 5.4m. The per-metro split is what exposes how skewed 'we operate in N cities' claims are. Updates track NHTSA SGO reporting timelines.

Coverage: Phoenix, SF Bay Area, Los Angeles, Austin, Atlanta and expanding · from September 2020

Access (inherited): public_page · free

Free public page with downloadable release notes. Waymo's terms govern reuse of its charts and data; cite as company-reported, never as independent.

Federal regulator. Policy direction shifts with administrations, so the same agency can be a deregulatory actor and a safety enforcer within one cycle - read releases as policy artefacts as well as data. The SGO crash dataset depends on manufacturer self-reporting with varying completeness.

Covers three separately monitorable streams: (a) AV exemptions and rulemaking - the Zoox grant of 2026-07-30 is the first commercial FMVSS exemption ever; (b) the SGO crash dataset, the only national AV incident series; (c) defect investigations, which are the leading indicator of any robotaxi reversal scenario. Set alerts on all three.

Coverage: United States · SGO ADS/ADAS incident data from 2021; recalls and investigations far longer

Access (inherited): api · free

US government work, generally public domain. NHTSA provides open datasets and APIs for recalls, complaints and investigations; SGO incident data is published as periodic CSV releases with a documented redaction policy.

Neutral publication of record; the rules themselves are of course political documents. No commercial bias in the source, though agency rationale sections are advocacy.

The primary-source anchor for every US vehicle regulation claim. The 2026-02-18 EPA rescission of the GHG endangerment finding and all vehicle GHG standards (91 FR 7686, docket EPA-HQ-OAR-2025-0194, effective 2026-04-20) is the highest-consequence automotive rulemaking of the year. Filter by agency plus CFR title 40 part 86 and title 49 part 531/533.

Coverage: United States · full text back to 1994; scanned back to 1936

Access (inherited): api · free

US government work, public domain. The federalregister.gov API is free, documented and explicitly intended for programmatic use with no key required. Ideal for automated monitoring.

US federal data with no commercial interest. Counts network-reported stations and ports, so it can lag decommissioning and may double-count during network transitions. Reflects installed capacity, not utilisation - which is the number that actually determines charging economics.

Gives US totals and Level 2 / DC fast splits continuously: 81,681 stations and 256,263 ports (75,589 DC fast) as of 2026-09-14. Full API with an API key makes this one of the easiest high-quality feeds to automate. Its main blind spot is utilisation, which no public source covers.

Coverage: United States, Canada and Mexico station data, with US state-level detail · station records back to the 1990s with continuous updates

Access (inherited): api · free

US government work. The NREL Developer Network API is free with a registered API key and documented rate limits. Bulk downloads permitted. Attribution to AFDC/NREL requested.

Proprietary methodology that is not published, so the figures are estimates rather than facts and should be labelled as such. Korean firm; some observers note its coverage of Chinese domestic installations depends on assumptions. It is nonetheless the de facto industry benchmark, which means it moves markets regardless of methodological opacity.

PAYWALL NOTE for licensing budget: this is the clearest example in the sector of a high-citation-value source that exists primarily as a paid report. The free republished headline (CATL 39.9%, BYD 14.7%, seven Chinese firms at 72.8% of the top ten for Jan-Jul 2026) is usually sufficient for trend detection; the paid detail is needed for supply-chain analysis.

Coverage: Global, with manufacturer-level GWh and share · monthly rankings back roughly a decade

Access (inherited): licensed · paid_high

Full reports are PAID and expensive. Headline rankings are widely republished by CnEVPost, electrive and trade press and are free to cite with attribution to SNE Research. Phase 2 should budget for a licence if manufacturer-level GWh is load-bearing, or rely on the free secondary republication if share percentages suffice.

Commercial research house with a clean-energy client base; historically its battery price series has been directionally accurate and widely validated. Annual cadence means the figure is stale for most of the year - as of 2026-09-15 the most recent survey was December 2025 and did not reflect China's new lithium-battery consumption tax effective 2026-09-01.

The canonical battery cost reference: 2025 survey gave $108/kWh global average packs, LFP $81/kWh vs NMC $128/kWh, China $84 vs North America $121 vs Europe $131. Publishes in early-to-mid December, so schedule a check then. The regional spread is the single most useful number for assessing Chinese vehicle price competitiveness.

Coverage: Global, with China, North America and Europe splits and LFP/NMC chemistry splits · annual survey since 2010

Access (inherited): public_page · freemium

Headline figures are published free in a summary blog post and are free to cite with attribution to BloombergNEF. The full survey is enterprise-licensed. Do not reproduce charts without permission.

Chinese automotive trade publisher with a commercial research arm selling to suppliers and OEMs. Reports CAAM and company data promptly in English but with an industry-promotional framing. Chinese-industry perspective; treat competitive commentary as interested.

The best English-language same-week window into CAAM data and, uniquely, into Chinese automaker exports BY DESTINATION REGION - which is the data that shows Latin America and the CIS mattering more than Europe for BYD, Chery and Geely. Required for any non-US-centric view of this sector.

Coverage: China, plus Chinese OEM activity worldwide · archive back over a decade

Access (inherited): rss · freemium

Free articles with attribution; research institute reports are paid. Respect the site's terms for automated collection.

Company-published and legally constrained by securities law, so hard numbers are reliable - but DISCLOSURE IS SELECTIVE AND CHANGES. Baidu's Q2 2026 release reported Apollo Go city counts and cumulative kilometres while omitting the quarterly ride counts it had previously disclosed. Reduced granularity as claims grow is itself a signal worth tracking.

Track disclosure CHANGES, not just disclosed values - a metric that disappears from a release is often more informative than one that is reported. Key entities: Tesla (quarterly deliveries, first business days of the quarter), Aurora (driverless mileage), Baidu (Apollo Go), Alphabet (any Waymo segment breakout, which would be a major signal), Stellantis, VW, Nissan.

Coverage: Global, company-specific · IR archives typically 10+ years

Access (inherited): rss · free

Free public filings and press releases. Free to quote with attribution. Most IR sites offer RSS and email alerts; SEC EDGAR full-text search covers the US-listed subset programmatically.

Funded by US auto insurers, so it has an institutional interest in credible crash-risk quantification and, arguably, in technologies that reduce claims. Methodologically rigorous and willing to publish inconvenient findings - its July 2026 Waymo study reported a 4% HIGHER crash rate in Austin alongside large reductions elsewhere, and criticised the national AV data regime its own analysis depends on.

The only substantial INDEPENDENT check on operator-published AV safety claims. Critically, its study and Waymo's own figures both rest on NHTSA Standing General Order data, so they are not fully independent of each other - and their headline percentages use different severity definitions and are not interchangeable.

Coverage: United States · research archive back decades

Access (inherited): rss · free

Free public releases and downloadable research papers. Free to cite with attribution to IIHS. Bulk reuse of ratings data is restricted.

Official EU source. Proposals are political documents and routinely diverge from prior commissioner signalling - the December 2025 package's 90% target and its flexibilities were not fully predictable from earlier statements. Crucially, a Commission PROPOSAL is not law: Parliament and Council negotiation can change or kill it.

Track the 2035 CO2 revision through trilogue, not just at proposal stage. The macro brief's AI Act lesson applies exactly here: the most-cited deadline in vendor and OEM communications may not bind on schedule. Pair with EUR-Lex for the legal text and with Parliament committee trackers for the negotiating position.

Coverage: European Union · EUR-Lex full legislative history back decades

Access (inherited): rss · free

EU documents are generally reusable under the Commission's reuse decision with attribution. EUR-Lex offers bulk download and a SPARQL endpoint for legislative texts.

Central-bank research with no commercial interest. Note the Philadelphia Fed's own argument that headline delinquency rates overstate new borrower distress because they reflect accumulating unresolved delinquencies - a methodological caveat that most press coverage of 'record auto delinquency' omits.

The affordability and credit-stress anchor for US vehicle demand: 60+ day auto delinquency at 1.68% (highest since 2008), subprime around 6% (highest in 20+ years), 29.3% of new-vehicle trade-ins in negative equity. The downside scenario for this whole sector runs through these series, so they are worth monitoring even though they are not 'automotive' sources.

Coverage: United States · NY Fed household debt series from 1999; Philadelphia Fed credit reports several years

Access (inherited): public_page · free

US government work, free to download and cite. Underlying credit-bureau panel data is licensed and not redistributable.

Independent research provider, US-based, with a broadly sceptical analytical posture toward Chinese industrial policy claims. Clients include governments and investors. Quantifies what Chinese official sources leave implicit, such as subsidy magnitude, which is valuable but is estimate not fact.

The best available quantification of Chinese auto subsidy scale - RMB 342bn of 2025 central subsidies, roughly 3% of central fiscal revenue, against a 10% YoY fall in average car prices. This is the analytical bridge between 'China exports a lot of cars' and 'China exports deflation', which is the part that matters for European and emerging-market incumbents.

Coverage: China, with global spillover analysis · research archive back over a decade

Access (inherited): rss · freemium

Public notes are free to read and cite with attribution; client research is paid. Standard copyright applies.

Operator-published mileage and route counts with no regulator publishing an equivalent independent series for driverless trucking - there is no CPUC equivalent for freight. Aurora is US-listed so material claims carry securities-law exposure, but the key operating metrics are voluntary disclosures.

MONITORING GAP TO FLAG: there is no regulator-published driverless-truck mileage series anywhere. Everything rests on operator disclosure - Aurora's roughly 440,000 cumulative commercial driverless miles through end-June 2026 across ten routes. Cross-check against FMCSA inspection and crash records for the relevant carriers as a partial independent check.

Coverage: United States, concentrated in Sun Belt states · Aurora IR archive from its 2021 listing

Access (inherited): rss · free

Free public press releases, free to quote with attribution. FMCSA carrier and inspection data is public and downloadable in bulk.

Independent English-language China EV news site with no obvious vendor sponsorship, run by a small named team. Reports CPCA, CAAM, SNE and company data quickly and with consistent methodology notes. China-based, so subject to the same information environment as other Chinese media, but it does flag data caveats more often than most.

The highest-value single English-language feed for Chinese market data, carrying CPCA retail, CAAM wholesale, SNE battery shares and weekly insurance-registration data in one place. For a sector whose centre of gravity is China, this is a tier-1 requirement rather than a nice-to-have.

Coverage: China, plus global battery and Chinese OEM coverage · archive from roughly 2020

Access (inherited): rss · free

Copyrighted; free to read, quote with attribution and link. No bulk republication.

Chinese state news agency. Factually accurate on official statistics - it reports CAAM and ministry figures verbatim - but framing is explicitly promotional of Chinese industrial policy and it will not report negative structural interpretations. Use for official figures and policy announcements; never for competitive assessment.

Useful as the fastest English-language carrier of Chinese official policy statements - for example the 30-to-60-day supplier payment rules, the overseas fair-competition guidance and the 30%-NEV-fleet-by-2030 target, alongside the August 2026 export figures. Always triangulate its numbers against CAAM or CPCA and label its framing as interested.

Coverage: China national · extensive archive

Access (inherited): rss · free

Free to read; Xinhua asserts copyright over its content. Quote briefly with attribution; do not republish in full.

Co-published by the US venture industry's trade body, which has a direct advocacy interest in the sector appearing healthy and in favourable policy treatment. PitchBook's deal universe is US-skewed and depends on disclosed rounds. Deal counts are revised upward for several quarters after initial publication.

The single highest-value free release in this sector. Q2 2026 edition carried $412.7B US H1 deployment, 87.5% megadeal share, 86% AI share, 48.1% top-three fundraising concentration and first-time fund formation at a decade low. Watch for upward revision of deal counts in subsequent quarters - the 'collapsing deal count' story is partly a reporting-lag artefact and must be re-checked.

Coverage: US primarily, with global comparison sections · Quarterly back to 2006

Access (inherited): public_page · free

Free PDF with registration; attribution to PitchBook-NVCA required; redistribution of the PDF restricted; underlying platform data is licensed separately.

Commercial data vendor. Analysts are named and the underlying data is proprietary and unauditable by readers. Vendor benefits from the sector appearing large, active and in need of data subscriptions.

Source of the 0.05x 2021-vintage DPI figure, the -$202B net LP cash flow figure, the 93.5% exit-value concentration figure and the VC-backed IPO aftermarket analysis. The most decision-relevant free commentary in the sector.

Coverage: Global, US-weighted · News archive back to ~2013; underlying data back to 1990s

Access (inherited): rss · freemium

Free articles; no bulk redistribution; charts are licensed. Platform data requires an enterprise subscription.

Primary data from Carta's own cap-table and fund-admin customer base. Sample is self-selected: US-weighted, skewed toward smaller and earlier-stage companies and sub-$100M funds. Carta has a commercial interest in early-stage activity appearing measurable and in being cited as the reference dataset. Sample composition changes as the customer base grows, which can create spurious trends.

The only free public source for down-round rates, bridge rounds, round counts by stage, AI vs non-AI valuation splits and fund-level DPI by vintage. Q1 2026 down-round rate 11.4%; AI Series A median ~$300M vs ~$55M non-AI; 2019-20 vintage median DPI barely above zero. Carta did not publish a Q2 2026 State of Private Markets at the time of research - a gap Phase 2 should monitor.

Coverage: US primarily · Quarterly series back to ~2018

Access (inherited): public_page · free

Free public reports; attribution required; no redistribution of underlying datasets.

Produced by an institutional investment adviser whose clients are the LPs being measured. Sample is limited to funds reporting to Cambridge, which introduces survivorship and selection bias. Publication lags roughly six months behind the data period.

The most independent available check on the DPI story: calls exceeded distributions by 1.6x since the start of 2022 versus distributions exceeding calls by 1.3x in 2012-21. Slow but high-credibility. Essential for triangulating any vendor DPI claim.

Coverage: US and global PE/VC · Index history back to 1981 for US VC

Access (inherited): public_page · free

Public PDFs; attribution required; index data itself is licensed.

No commercial interest. The genuine limitation is relevance rather than bias: BFS counts all EIN applications, the vast majority of which are not venture-relevant, so it is a proxy for entrepreneurial activity rather than a startup-formation measure. Seasonal adjustment revisions are routine.

The only Tier-A, free, high-frequency, API-accessible formation series in this sector. August 2026: 531,728 seasonally adjusted applications, -7.8% m/m; projected formations 28,501, -4.6%. Use as the independent counterweight to vendor funding data, never as a venture metric on its own.

Coverage: United States, national, state, county and NAICS · Back to 2004 (monthly), weekly series back to 2006

Access (inherited): api · free

US Government work, public domain; no restrictions; citation appreciated.

No commercial interest. Form D is a notice filing: it reports offering size and amount sold as self-reported by the issuer, is not audited, and many venture rounds are filed late, amended, or exempted. Not a substitute for deal databases, but the only legally-required primary record.

The highest-value under-used source in this sector. Form D gives an unfiltered, legally-required view of private offering volume that is independent of every vendor in this registry, and ADV filings give adviser-level AUM and fund counts. Phase 2 should build a Form D pipeline before buying any commercial dataset.

Coverage: United States · Form D electronic filings back to 2008; EDGAR back to 1993

Access (inherited): api · free

Public domain. SEC fair-access policy requires a declared user-agent and rate limiting to 10 requests/second.

Commercial database with a freemium funnel. Coverage depends on self-reporting and press releases, so it under-counts undisclosed and non-US rounds. Its global totals differ materially from KPMG's on the same period - a definitional, not erroneous, divergence.

Best free source for regional quarterly cuts (Asia, Europe) and for seed graduation-rate analysis. Its H1 2026 global total of $510B against KPMG's $560.4B is the programme's canonical worked example of methodological divergence - always report both.

Coverage: Global · News back to 2015; database back to 2007

Access (inherited): rss · freemium

Free articles; API and bulk data require a paid licence; attribution required; no scraping of the database.

Big-4 firm with an advisory interest in the venture and private-capital client base. Broader deal-inclusion methodology than Crunchbase, producing systematically higher totals and deal counts.

The necessary second reading against Crunchbase. H1 2026: $560.4B global, 8,440 Q2 deals (lowest count since Q3 2017), Asia $50.8B, Europe $25.6B. Never quote one tracker without the other.

Coverage: Global with regional cuts (Americas, Europe, Asia) · Quarterly back to 2014

Access (inherited): public_page · free

Free PDF; attribution to KPMG Venture Pulse required; no redistribution.

Both firms earn advisory fees on the volume they report, so there is a direct commercial interest in the market appearing large and growing. Volume is estimated from transactions each adviser observes; the two firms' totals differ. Pricing data is the most commercially sensitive part and is the least disclosed.

The de facto scoreboard for secondaries: H1 2026 above $120B, +20%, GP-led at 53.7%, single-asset CVs at $34B, dry powder down 10% with the overhang multiple near 1.0x. Weight the volume figures; discount the accompanying commentary for conflict.

Coverage: Global · Semi-annual series back to ~2010

Access (inherited): public_page · free

Free summary PDFs, often behind a registration wall; full reports are distributed to clients. No redistribution.

Consultancy with a large private-capital client base and an interest in the sector's continued growth. Methodology relies on Preqin/PitchBook underlying data plus a proprietary LP survey, so it is partly a resampling of other sources in this registry rather than a fully independent one.

Best single annual view of PE-side structure: distributions at ~6% of AUM vs a ~14% historical average, five-year rolling DPI-to-AUM at a record low ~10%, continuation vehicles at $115B in 2025, 16,000 companies held four-plus years, US semi-liquid PE at $204B. Published each February.

Coverage: Global · Annual editions back to 2015

Access (inherited): public_page · free

Free PDF; attribution required; no redistribution or derivative commercial use.

Index and data provider that sells private-capital benchmarks; commercial interest in benchmark adoption. Built on the acquired Burgiss cash-flow dataset, which is genuinely independent of GPs in a way most vendor data is not.

One of very few non-GP-sourced views of fund cash flows, and the best public source on evergreen vehicles (~$500B AUM as of Q1 2025, +30% over twelve months, wealth investors at one-fifth). Licensing budget likely required for the underlying data.

Coverage: Global · Burgiss-derived fund cash-flow data back to the 1970s

Access (inherited): rss · freemium

Free blog and research summaries; index and dataset access is licensed; no redistribution.

No commercial interest. The genuine limitation is that published agenda dates are explicitly non-binding - outside counsel described the October 2026 target for 'Enhancing Retail Exposure to Private Markets' as ambition rather than schedule. Staff positions can change without formal rulemaking, as the 2025 withdrawal of the 15% closed-end fund limit showed.

The single most outcome-determining monitoring target for this sector in 2026-27. Track both formal proposals and informal staff statements, since the most consequential 2025 change was a staff position, not a rule.

Coverage: United States · Back to 1990s

Access (inherited): rss · free

Public domain; fair-access rate limits apply to automated retrieval.

No commercial interest. Highly politically sensitive: the 2025 Executive Order direction can be reversed by a subsequent administration, and fiduciary safe harbours are litigated rather than settled.

Governs access to the ~$14T DC pool covering 90M+ participants - the largest potential new LP base in the sector's history. Cross-check every date against the Federal Register directly; secondary law-firm alerts carried inconsistent dates during this research.

Coverage: United States · Back to 1990s

Access (inherited): rss · free

Public domain.

Issuer-produced and selectively framed, but subject to securities-law liability and auditor review - the only audited, legally-accountable financial disclosure in the entire sector. AUM definitions differ between managers and are not comparable without adjustment.

The highest-reliability monitoring target in this registry. Perpetual-capital and wealth-channel AUM, realisations, redemption levels and fee-related earnings are all disclosed here quarterly. Note: Blackstone's IR press-release URL structure changed and returned 404s during this research; verify feed endpoints before relying on them.

Coverage: Global · Back to each firm's IPO (Blackstone 2007)

Access (inherited): rss · free

Public company disclosure; freely quotable with attribution.

No vendor interest; these are required filings. But disclosure formats are not standardised across funds, so redemption rates must be reconstructed fund by fund, and repurchase-offer results are sometimes disclosed only in tender-offer documents rather than periodic reports.

The early-warning system for the retail-access thesis. Q1 2026 already showed Blue Owl OBDC II closing redemptions, Cliffwater receiving requests for 14% against a 7% cap and BCRED lifting its limit to 7.9%. Reconstructing a standardised redemption-rate series across the largest twenty vehicles would be one of the highest-value Phase 2 datasets in this sector.

Coverage: United States · Back to each fund's inception

Access (inherited): api · free

Public domain; SEC fair-access rate limits apply.

European-headquartered trade publication with a stake in European tech appearing significant. Its deal universe differs from Dealroom's and from US trackers, producing different country shares for the same period.

Regional counterweight to US-centric data. H1 2026: EUR 44.1B across 1,740 European deals, UK EUR 18.7B, Germany EUR 6.3B, France EUR 6.0B. Its UK share (42.4% of H1) differs from Dealroom's 48.4% YTD - use both.

Coverage: Europe · Back to 2013

Access (inherited): rss · freemium

Free articles with a metered paywall; attribution required; data reports are paid.

Commercial European database, part-funded by public-sector ecosystem contracts, which creates an incentive for the ecosystems it reports on to look active. Many public pages are undated, which is a real citation problem.

The best free European ecosystem dashboards. Source of the UK's record 48.4% share of European VC in 2026 YTD. Flag: the cited page carries no visible publication date and must be recorded as undated.

Coverage: Europe primarily, global coverage · Back to ~2013

Access (inherited): api · freemium

Free public dashboards; API and bulk data licensed; attribution required.

Listed Indian data company; commercial interest in Indian ecosystem visibility. Uses the Indian fiscal year (April-March), which is not comparable to calendar-year Western data without adjustment.

The standard regional source for India. FY26: $11.7B, down 18%; late-stage down 38%; 47 IPOs against 31 the prior year - the clearest example anywhere of a domestic listing route decoupling from the funding cycle.

Coverage: India primarily, global coverage · Back to ~2013

Access (inherited): public_page · freemium

Free report summaries; platform access is paid; attribution required.

LP-side trade body, so systematically biased toward LP interests on fees, transparency and GP-led conflicts. That bias is useful: it is the only organised counterweight to GP-controlled reporting.

The body most likely to produce the standardised DPI and continuation-vehicle disclosure that would close this sector's single biggest data gap. Monitor for any move toward standardised reporting of CV-sourced distributions.

Coverage: Global, US-weighted membership · Reporting template guidance back to 2016

Access (inherited): public_page · free

Free templates and guidance; attribution required; membership needed for some material.

Specialist trade publication owned by a private-markets media and data group with conference and data revenue from the same market it covers. Most substantive articles are paywalled.

The most complete continuous coverage of secondaries and continuation vehicles, but the body text is paywalled - during this research only the headline and one quote were retrievable. Phase 2 needs a licensing budget here or must rely on the free Evercore/Jefferies summaries and syndicated coverage.

Coverage: Global · Back to 2014

Access (inherited): rss · paid_high

Subscription required for full articles; no redistribution; headlines and metadata are public.

Co-published by an operating venture lender with a direct commercial interest in venture debt appearing large, legitimate and growing. The underlying data is PitchBook's, so this is not an independent second source on PitchBook figures.

The only regular public series on venture debt volume: a record $68.8B in 2025 across ~1,000 transactions with flat deal count. Treat the accompanying narrative as vendor commentary.

Coverage: United States primarily · Annual editions since 2023

Access (inherited): public_page · free

Free PDF with registration; attribution required.

No commercial interest. Deliberately conservative language means financial-stability concerns are understated rather than overstated - when the FOMC names a sector, it is a strong signal.

Directly relevant: the July 2026 minutes flagged financial-stability risk from high AI-firm equity valuations and increased leveraged financing of the AI infrastructure buildout, and identified the AI buildout as an inflation driver. The 2026-09-16 decision falls one day after this research date and is unresolved.

Coverage: United States, global spillovers · Back to 1936

Access (inherited): rss · free

Public domain.

NACUBO's annual study is self-reported by participating institutions with a lag; public pension board materials are timely and legally required but vary enormously in format and detail by state. Endowments disclose selectively and often only after a transaction is complete.

The LP-side ground truth on allocation changes and secondary sales: Yale up to $6B, Harvard ~$1B, NYC pensions $5B. Public pension board agendas are the most timely free signal of LP behaviour and are badly under-monitored.

Coverage: United States · NACUBO study back to 1974

Access (inherited): public_page · free

Summary results are free; the full dataset is paid. Public pension board materials are public records.

None commercial. Survey-based with sampling error explicitly published (+/-0.9% on the e-commerce growth rate). Definition excludes food services and most services, so it is not comparable to private 'digital commerce' totals — the most common misuse of this series.

The single most important source for this sector. The e-commerce PDF at census.gov/retail/mrts/www/data/pdf/ec_current.pdf is directly machine-readable. Q2 2026 release 2026-08-18: 17.1% share, +12.2% YoY. Also available via the Census Time Series API.

Coverage: United States · back to Q4 1999

Access (inherited): bulk_download · free

US federal government work, public domain. Citation of the Census Bureau as source is requested.

Advance estimates are revised, sometimes materially, in the following month's MRTS. Nominal, not inflation-adjusted, so with headline CPI at 3.4% roughly two-thirds of reported growth is price.

Category-level detail (nonstore, food and beverage, general merchandise, clothing) is the fastest read on channel and category shift. July 2026: total $763.6bn, -0.6% MoM, +5.0% YoY; nonstore +7.7%.

Coverage: United States · back to 1992

Access (inherited): api · free

Public domain US federal government work.

None commercial. Category weights lag consumption shifts; hedonic and seasonal adjustment choices are methodologically contested but transparently documented.

The only reliable way to measure tariff and energy pass-through at category level. August 2026: headline +3.4%, core +2.4%, energy +16.3%, core goods +0.7%. The core-goods series is the direct test of tariff pass-through claims.

Coverage: United States · back to 1913

Access (inherited): api · free

Public domain. BLS asks that data not be presented as endorsed by BLS.

None commercial. The internet sales ratio is defined differently from the US Census e-commerce share, which makes naive UK-US comparisons wrong by roughly ten percentage points.

The best non-US benchmark for online penetration, and structurally ahead of the US: 28.3% in July 2026. Essential for any claim about where US online share is heading.

Coverage: Great Britain · back to 1988

Access (inherited): bulk_download · free

Open Government Licence v3.0 — free reuse with attribution.

State statistical agency with a documented national interest in presenting consumption stability; revisions and base-effect handling are less transparent than US or EU equivalents. Treat levels with more caution than directional changes.

The required non-English-origin primary source for this sector. Jan-Aug 2026: total retail RMB 32.76tn +1.1%, August alone +0.4%, online retail RMB 13.48tn +4.6%. Xinhua (english.news.cn) carries the same figures faster than the NBS English site.

Coverage: China · back to the 1990s; online retail series from 2014

Access (inherited): public_page · free

Public government release; English translations published alongside Chinese originals.

Official EU position; the framing emphasises fair competition for EU sellers. Parcel-volume figures are cited across releases with inconsistent units ('packages' versus 'items') and reference years.

Primary source for the EUR 150 exemption abolition (Council final approval 2026-02-11, effective 2026-07-01) and the temporary EUR 3 per-item duty to 2028. Council press releases at consilium.europa.eu carry the political milestones; DG TAXUD carries the legal text and guidance.

Coverage: European Union · back to the 1990s

Access (inherited): rss · free

European Commission reuse policy — free reuse with attribution (Decision 2011/833/EU).

Company self-reporting, audited. Segment definitions are chosen by Amazon and have changed over time; 'advertising services' is not fully comparable to third-party retail media estimates.

The IR press-release URL pattern is unstable; the reliable route is the EX-99.1 exhibit under sec.gov/Archives/edgar/data/1018724/, which is directly fetchable while the EDGAR cgi-bin browse endpoint is robots-disallowed.

Coverage: Global · back to 1997

Access (inherited): public_page · free

Public company disclosure; freely citable. SEC EDGAR full-text access requires a declared User-Agent.

Company self-reporting. Walmart discloses e-commerce and advertising as growth rates rather than dollar figures, which prevents independent share calculation — a deliberate disclosure choice.

Direct PDF links to the earnings release, presentation and management-call transcript are published on the quarterly results page and are reliably fetchable. Q2 FY2027 (2026-08-20) is the key source for the tariff-refund margin effect.

Coverage: Global, US-weighted · back to the 1990s

Access (inherited): public_page · free

Public company disclosure; freely citable with attribution.

Company self-reporting across four issuers with different fiscal calendars and disclosure conventions; PDD in particular does not break out Temu, which is the single largest disclosure gap in this sector.

Grouped because the monitoring pattern is identical. Target's release is the best public quantification of tariff refunds ($994mn in Q2 FY2026); Instacart's is the cleanest read on retail media take rate as a percentage of GTV.

Coverage: US, Canada, China · varies by issuer; PDD from 2018

Access (inherited): public_page · free

Public company disclosure; freely citable.

Enforcement agency perspective. The de minimis statistics CBP publishes are collection-driven and undercount volumes that enter through other channels.

Essential for tracking de minimis administration and IEEPA refund processing ahead of the statutory repeal in July 2027. The 403 on direct page fetch is a known access constraint for Phase 2.

Coverage: United States · rolling, with archives

Access (inherited): rss · free

Public domain US federal government work. Note: cbp.gov returned 403 to automated fetching during this research — use the newsroom RSS feed rather than direct page fetches.

None. Primary judicial record.

The 2026-08-13 de minimis decision shows why this must be monitored directly: the Supreme Court struck IEEPA tariffs in February yet the de minimis suspension survived on a separate 'privilege revocation' theory. Secondary coverage consistently conflated the two.

Coverage: United States · back to 1980

Access (inherited): public_page · free

Public domain US federal court records.

A 501(c)(3) with a documented preference for lower and simpler taxation; its tariff-incidence modelling assumes full pass-through to consumers, which measured core-goods CPI of +0.7% does not support.

One of very few continuously updated public estimates of the US average effective tariff rate (7.2% for 2026 as of 2026-09-10). Cross-check its incidence claims against BLS CPI rather than citing them as consumer-price outcomes.

Coverage: United States · tracker maintained since 2018

Access (inherited): public_page · free

Creative Commons attribution licence on most Tax Foundation content.

Non-partisan university modelling centre. Estimates are scenario-dependent and clearly labelled as such.

Source for the up-to-$175bn IEEPA refund estimate and the finding that IEEPA duties were about 51.9% of total customs duties in January 2026 — the figure that sizes the 2026 retail margin windfall.

Coverage: United States · back to 2016

Access (inherited): public_page · free

Academic publication; cite with attribution.

Subscription research business selling to advertisers and agencies; its forecasts are the industry's reference point, which creates an incentive toward continuity with prior forecasts. Network share estimates are modelled, not audited.

The only public source found that splits US retail media by network (Amazon 78%, Walmart 7.5%, all others 14.5%) and that publishes an ex-Amazon growth rate. That ex-Amazon figure is the most decision-relevant number in the category and is available only through this paywall or trade-press summaries of it.

Coverage: Global · ad-spend series back decades

Access (inherited): licensed · paid_high

Paywalled and licensed. Redistribution of figures is restricted; headline numbers circulate via trade press. PHASE 2 NOTE: this is a licensing-budget line item.

Subscription forecaster whose figures dominate trade-press citation. Methodology is not fully public and forecasts are revised frequently; the free tier exposes conclusions without the underlying numbers.

Useful for direction and framing ('growth is slowing', 'the long tail slipping further behind') but the specific figures require a licence. Do not treat EMARKETER and WARC as independent confirmation of each other — they observe the same advertiser panel universe.

Coverage: Global, US-weighted · back to the early 2000s

Access (inherited): licensed · paid_high

PRO+ subscription required for data tables. Free article summaries are citable with attribution. PHASE 2 NOTE: paywalled.

Adobe sells the AI-optimisation tooling that its findings imply retailers need — a direct commercial interest in the conclusion. The panel is large (over a trillion visits) but covers Adobe Analytics customers, skewing to large enterprise retailers.

The only quantified public measure of AI-referred retail traffic, and therefore load-bearing for the entire agentic-commerce narrative. CRITICAL CAVEAT: Adobe has never published AI referrals as a share of total retail traffic. Phase 2 should track whether a denominator is ever disclosed — that single number would resolve the sector's biggest open question.

Coverage: United States · AI-referral series from about October 2024

Access (inherited): public_page · free

Free to cite with attribution to Adobe Analytics.

Subscription research business. Counts are of announced openings and closures, not completed ones, so a single large announcement distorts a week and reversals are not always netted out.

The de facto standard store-closure dataset and the only weekly cadence in this sector. The public preview of the midyear 2026 report indicates a full-year outlook of roughly 7,900 closures and 5,500 openings; the line-item data is behind the paywall.

Coverage: United States and United Kingdom · back to about 2017

Access (inherited): licensed · paid_high

Paywalled databank; weekly headline summaries are partly public. PHASE 2 NOTE: licensing budget required for the underlying databank.

Advocacy organisation for US retailers with an interest in framing returns fraud and retail crime as large problems requiring policy response. NRF loss-prevention statistics have previously required public correction, so treat its crime and shrink figures with particular care.

Best available public source on US returns volume ($849.9bn projected 2025; 19.3% online return rate), but the most recent Returns Landscape found was October 2025 with no 2026 update, and the co-author Happy Returns is a UPS commercial business.

Coverage: United States · returns series since about 2015

Access (inherited): public_page · free

Free to cite with attribution. Underlying microdata not released.

Consultancy research that also functions as business development for its retail practice. Consumer figures are survey-based and self-reported intention, which systematically overstates behaviour change.

The most detailed free public dataset on grocery trade-down behaviour (47% traded into private label; volume -1.0% against price +2.2% in 2025). Pair the survey findings with retailer transaction-versus-ticket splits, which measure behaviour rather than intention.

Coverage: North America and Europe · annual series since about 2020

Access (inherited): public_page · free

Free to download and cite with attribution.

Vendor primary sources with an interest in protocol adoption. Their conspicuous silence on transaction volume is itself the most informative signal they emit.

Monitor the GitHub repositories and changelogs rather than the marketing pages — commit activity and merchant-facing endpoint changes are the only observable adoption proxy while no participant discloses volume. A first published transaction or merchant count from any of these would be a category-defining event.

Coverage: Global · from September 2025

Access (inherited): public_page · free

Public developer documentation; ACP and AP2 are published as open standards with public repositories.

Trade publication with named bylines that also sells e-commerce market research; its own rankings and estimates are proprietary and methodologically opaque, though its reporting on third-party data is straightforward.

Useful for fast coverage of Adobe, comScore and retailer data releases. Distinguish its reporting on others' data (reliable) from its own Top 1000 estimates (proprietary, unaudited).

Coverage: United States · back to about 2000 (as Internet Retailer)

Access (inherited): rss · freemium

Partial paywall; articles citable with attribution.

Trade press funded by advertising from retail-technology vendors. Reporting is bylined and straightforward; the bankruptcy tracker covers major filings only and undercounts small and private failures.

The running bankruptcy list is the most accessible public distress tracker (six major 2026 filings through 2026-07-10 versus nine in all of 2025) and is the fastest counter-evidence to retail-apocalypse claims.

Coverage: United States · back to 2014

Access (inherited): rss · free

Free with registration; citable with attribution.

None. Primary regulatory record. Coverage is limited to SEC registrants, so private retailers (Shein, Aldi, Trader Joe's) and non-US issuers are invisible.

The most reliable route to primary retailer disclosure when investor-relations sites render dynamically or redirect — which happened with Kroger, Target and Shopify during this research. Full-text search on risk-factor language ('de minimis', 'tariff refund', 'agentic') is a strong early-warning method for Phase 2.

Coverage: US-listed issuers globally · back to 1994

Access (inherited): api · free

Public domain. A declared User-Agent header is required; the cgi-bin browse endpoints are robots-disallowed but the Archives and full-text-search endpoints are accessible.

State news agency of the People's Republic of China. Statistical figures match NBS releases, but selection and framing serve a national narrative; it reports outcomes without discussing weak-demand drivers.

Publishes NBS monthly retail figures in English faster than the NBS English site itself, which lags. Use for timeliness, then verify levels against stats.gov.cn.

Coverage: China · rolling archive

Access (inherited): rss · free

Free to read; redistribution of full text restricted.

Membership trade association funded by robot manufacturers, so framing favours adoption growth. The statistics themselves are the sector standard and IFR is notably willing to contradict member marketing — it publicly stated in May 2026 that humanoid mass adoption will not occur in the near or medium term.

Roughly six releases a year, each carrying a distinct dataset (global installs, density, national markets, China policy). This is the single highest-value free source for the sector.

Coverage: Global, with national breakdowns for ~40 countries · Annual series back to the 1990s; press releases back several years

Access (inherited): public_page · free

Press releases are free to cite with attribution. The underlying World Robotics datasets are a separate licensed product and may not be redistributed.

Compiled from national robot association returns and manufacturer surveys; vendor-nationality attribution in the domestic-share series can be distorted by joint ventures and local assembly.

PHASE 2 LICENSING REQUIREMENT. The free releases give headline installs, stock and density but withhold the end-use, application and vendor-level breakdowns, the full national time series, and the service-robot data entirely. The 2026 edition, covering 2025 data, publishes around late September 2026 — days after this research date. Budget for this licence; there is no substitute.

Coverage: Global · Annual back to 1993

Access (inherited): licensed · paid_high

PAYWALLED. Sold as a licensed report; redistribution prohibited. Only the press-release summaries are freely citable.

None on the data itself. Selection bias toward US-listed and US-listing-seeking companies; most robotics firms are private or listed in Asia and never appear.

This produced the two hardest facts in this dossier: Symbotic's Q3 FY2026 revenue from the XBRL companyconcept endpoint (CIK 0001837240) and the existence and date of the Agility Robotics / Churchill Capital Corp XI Form S-4 (2026-09-04). Note that the `forms` filter on the full-text endpoint did not work reliably; a date-range query did. www.sec.gov/cgi-bin/browse-edgar is blocked by robots.txt, so use the efts and data.sec.gov endpoints.

Coverage: US-listed issuers and their filing counterparties · Full text back to 2001; XBRL facts back to roughly 2009

Access (inherited): api · free

Public domain US government data. A descriptive User-Agent is required by SEC access policy; rate limits apply.

State-affiliated exchange; commentary carries an official Chinese framing and promotional tone toward listed domestic champions. The prospectus figures it reproduces are audited and are the best available data on Chinese robotics firms.

NON-ENGLISH-ORIGIN SOURCE for this sector. The single most valuable humanoid datapoint in this dossier — Unitree's 5,632 cumulative humanoid shipments, RMB 1.70bn 2025 revenue and RMB 278.21m statutory net profit — came from here. As China hosts a reported 97% of global humanoid shipments, mainland exchange disclosure is now the primary window into real humanoid volume. The Chinese-language site carries substantially more detail than the English edition.

Coverage: China (mainland-listed issuers) · Rolling news archive; full filings via the Chinese-language site

Access (inherited): public_page · free

Public disclosure. Attribution expected. Some material is only published in Chinese.

Advertising- and event-funded trade press. Carries a high proportion of vendor-bylined and sponsored technical content alongside genuine reporting — the agriculture piece used in this dossier was authored by the vendor's own CEO. Named-byline news reporting is reliable; guest technical articles must be read as supplier marketing.

The fastest general robotics news source and the only one that reported Agility Robotics' $1.8m revenue and $140m operating loss from the S-4, and Unitree's sub-10% industrial revenue share. Its RSS feed is the reliable access route; direct article-URL construction fails frequently and the on-site search page does not render results to fetchers.

Coverage: Global, with a US and European emphasis · Archive back to 2016

Access (inherited): rss · free

Standard copyright; cite and link. No scraping restrictions encountered.

Issuer-controlled. Procedure-growth framing emphasises volume over pricing and does not disclose per-procedure revenue or discounting, which is where competitive pressure will first appear.

The sector's best single revenue and unit-economics disclosure. Quarterly procedure growth is the leading indicator for the whole surgical robotics segment. Caution: the IR landing page and news-release index rendered only as link stubs to fetchers during this research and quarterly figures had to be sourced from trade press; pull the PDF financial data tables directly in Phase 2.

Coverage: Global, with US/OUS procedure splits · Back to the 2000 IPO

Access (inherited): public_page · free

Public company disclosure; freely citable.

Issuer-controlled, and heavily exposed to a single customer, so backlog presentation should be read with customer-concentration disclosure alongside it.

The only large audited pure-play warehouse-robotics revenue line. The IR site's own pages return 404s on predictable paths; go through the SEC XBRL API (CIK 0001837240) instead, which worked cleanly.

Coverage: US primarily, expanding internationally · Back to the 2022 de-SPAC

Access (inherited): public_page · free

Public company disclosure; freely citable.

Issuer-controlled; Japanese disclosure convention reports robot orders with limited regional granularity.

FANUC's robot order intake is the best high-frequency proxy for global industrial robot demand, leading IFR's annual data by many months. NOT RETRIEVED IN THIS RESEARCH: the IR library and finance pages returned 404 to fetchers. Phase 2 should locate the stable PDF paths or use TDnet/EDINET.

Coverage: Global with Japan/China/Americas/Europe splits · Long-run archive

Access (inherited): public_page · free

Public company disclosure; freely citable.

Issuer-controlled. Publishes a quarterly order-trend series that is unusually useful and unusually under-cited outside Japan.

NON-ENGLISH-ORIGIN. Yaskawa's quarterly order trends by region are one of the few near-real-time reads on Chinese robot demand. Only the IR landing page was retrievable in this research; an IR notice dated 2026-08-18 announced the FY26 Q2 results date but no figures were accessible. The annual report is offered as a 14MB ZIP.

Coverage: Global with China detail · Long-run archive

Access (inherited): public_page · free

Public company disclosure; freely citable.

Issuer-controlled. As a near-monopoly supplier of strain wave gearing, its bookings are a genuine leading indicator but it has an interest in signalling robotics demand strength.

THE component-level leading indicator for both industrial robots and humanoids: nearly every robot joint and every credible humanoid design uses strain wave or cycloidal reduction. A sustained step-change in its consolidated bookings would be the earliest hard evidence that humanoid volume is real. Confirmed its move from TSE Standard to Prime Market on 2026-02-27; detailed results were not retrievable from the English IR pages.

Coverage: Japan, China, Europe, Americas · Long-run bookings and sales series

Access (inherited): public_page · free

Public company disclosure; freely citable.

Strongly interested party. Partner lists conflate simulation-toolchain integration with autonomous-behaviour deployment, and benchmark claims are self-reported. Useful as a census of who is working with whom; not usable as evidence of deployment.

Every claim sourced from here in this dossier is labelled `marketing`. Its real analytic value is the partner roster, which is the most complete public map of which robot OEMs and robot-AI startups are working together.

Coverage: Global · Archive back many years

Access (inherited): rss · free

Press material, freely citable with attribution.

Named-byline trade journalism with no vendor sponsorship of editorial. US-centric regulatory focus.

The most reliable free tracker of surgical robotics clearances and competitive dynamics. Tracks the FDA clearance calendar that actually sets the Intuitive-versus-challengers timetable.

Coverage: US primarily, some EU and global · Archive back to 2018

Access (inherited): rss · free

Standard copyright; cite and link.

None. Purely administrative record. Lags the company announcement by days to weeks and does not indicate commercial launch.

Primary confirmation for every surgical robotics competitive claim — Hugo, Ottava, Versius Plus and each new da Vinci indication. Should be polled directly rather than relying on trade-press reporting of clearances.

Coverage: US · Back to 1976

Access (inherited): api · free

Public domain US government data. openFDA API terms apply.

Survey response bias and small subsector cells, but it is a genuine population-based instrument covering roughly 1.2 million businesses in six rotating panels — far better than any vendor survey.

IMPORTANT GAP: BTOS publishes an AI-use series but no robotics-use series was located, so there is no population-level US measure of firm-level robot adoption to cross-check IFR's unit data against. Phase 2 should check whether the robotics supplement exists in the microdata or by special tabulation; if it does not, this is a structural blind spot for the whole sector.

Coverage: US, by NAICS sector, state and the 25 largest MSAs · Back to 2022

Access (inherited): bulk_download · free

Public domain US government data.

North American vendor trade association; publishes quarterly North American robot order and shipment statistics from member returns, so coverage depends on membership and framing favours growth.

NOT INDEPENDENTLY VERIFIED IN THIS RESEARCH — included on the basis that A3 is the recognised North American counterpart to IFR and reports two to three quarters ahead of IFR's annual data. Phase 2 must confirm the current feed URL, publication cadence and what is free versus member-only before relying on it.

Coverage: North America (US, Canada, Mexico) · Quarterly series back to the 1990s

Access (inherited): public_page · freemium

Headline statistics are published free; detailed member reports are restricted to members.

Japanese manufacturer trade association; reports production and shipment value for Japanese makers, which is a supply-side rather than demand-side measure and is sensitive to yen movements.

NOT RETRIEVED IN THIS RESEARCH — recorded as a known gap. Japan is the world's largest robot-producing country and JARA's production and export statistics are the only direct read on Japanese vendor output, which matters acutely given the rare-earth allocation risk. Phase 2 needs a Japanese-reading analyst or translation capability here.

Coverage: Japan production, with export destinations · Long-run quarterly series

Access (inherited): public_page · freemium

Summary statistics published free; detailed series are a paid member product. Much of the material is Japanese-language only.

State-published and state-framed. Output and installation figures serve industrial-policy objectives; domestic-share and localisation figures in particular should be treated as directional. Chinese-language only for most substantive material.

NOT RETRIEVED IN THIS RESEARCH — the Chinese-language primary sources were not accessible within the search budget, and IFR's summary of the 15th Five-Year Plan was used instead. This is the largest single non-English gap for this sector: China is 54% of global installations and reportedly 97% of humanoid shipments, and it is being read here through a German trade body's summary.

Coverage: China · Multi-year series

Access (inherited): public_page · free

Government publication; attribution expected. No English API.

A US policy organisation republishing Chinese customs data; the underlying data is official, the selection and framing are advocacy-adjacent on supply-chain security.

The binding physical constraint on robot production nobody watches as a robotics indicator. NdFeB magnets are roughly 80% Chinese and go into every servo motor. Japan received zero covered rare-earth exports in July 2026 — a direct threat to FANUC, Yaskawa, Harmonic Drive and Nabtesco. Carried over from sector 09's registry; monitor destination-level allocation to Japan and Korea specifically.

Coverage: China exports by destination country · Multi-year monthly series

Access (inherited): public_page · free

Dashboard republication of official customs data; cite both the dashboard and China customs.

Self-selected trade-publication readership panel of about 166 respondents, run with vendor-association partners. Respondents are automation-interested by construction, so absolute adoption levels are overstated; year-over-year changes within the same panel are the usable signal.

The ONLY quantified operator-side adoption instrument located for warehouse robotics, and the source of the RaaS-versus-capex split (29% versus 36%). It is not a population estimate and must never be reported as one. Its weakness is itself a finding: warehouse automation adoption has no authoritative public measure.

Coverage: US primarily · Several annual waves

Access (inherited): public_page · free

Free to read and cite with attribution.

Consensus committee process dominated by incumbent manufacturers; standards lag technology deliberately, which is itself the signal.

The falsifiable gate on humanoid and learned-behaviour deployment. Current safety certification (ISO 10218-1/-2, ISO/TS 15066) presumes deterministic, analysable robot behaviour and fixed or fenced workspaces; no harmonised standard yet certifies a free-roaming biped or a non-deterministic learned policy. Watch the work programme: a new work item covering mobile manipulators or learned behaviour would be the earliest credible signal that humanoid deployment is becoming legally possible at scale, well ahead of any shipment figure.

Coverage: Global · ISO 10218 back to 1992; ISO/TS 15066 from 2016

Access (inherited): public_page · paid_low

Standard texts are sold per document; committee scope, work programme and stage codes are free to view.

Issuer-controlled. Robotics is a small segment inside a semiconductor-test company, so disclosure granularity is limited and the segment narrative can be shaped by the parent's story.

The only audited public read on collaborative-robot demand, a segment IFR does not break out separately in its free releases. Reported five consecutive quarters of robotics growth as of September 2026. Its new IFR presidency also gives the cobot segment unusual influence over sector statistics framing.

Coverage: Global · Back to the 2015 UR acquisition

Access (inherited): rss · free

Public company disclosure; freely citable.

Enthusiast-origin publication with a generally pro-EV editorial stance but a consistent record of tracking Tesla's stated targets against outcomes, which is precisely what makes it useful here. Named bylines.

Tesla does not disclose Optimus production or deployment in its financial filings, so there is no Tier A source for the most-discussed humanoid programme in the world. Tracking Musk's statements against outcomes through named-byline trade press is the only available method, and it should be labelled as such. The January 2026 admission of zero deployed units against a 10,000-unit promise came through this channel.

Coverage: US primarily · Archive back to 2013

Access (inherited): rss · free

Standard copyright; cite and link.

Nielsen is a commercial measurement vendor selling the underlying currency; it has a direct interest in the Gauge remaining the reference series. Panel-plus-big-data methodology is proprietary and not independently auditable.

THE single most important source for this sector. Publishes as a press release plus an image-embedded chart, so the numbers must be parsed from release text rather than the chart. CRITICAL: Nielsen has confirmed methodological updates to both the Gauge and the MDG for autumn 2026 using ARF DASH universe estimates; the historical series will break. Phase 2 must capture both the old and new series and never splice them.

Coverage: United States · Monthly series since May 2021

Access (inherited): public_page · free

Public summary pages free to cite with attribution; underlying ratings data is licensed and may not be redistributed. No scraping of the licensed product.

Same commercial interest as the monthly Gauge; the ad-supported cut is produced partly to serve advertiser clients and frames the data for that buyer.

Gives the ad-supported split that the monthly Gauge does not, plus a demographic cut. The adults-18+ breakdown (streaming 44.4% vs total-population 48.2% in Q2 2026) is the cleanest public evidence of the age gradient in streaming adoption.

Coverage: United States · Quarterly since 2024

Access (inherited): public_page · free

Attribution required; underlying data licensed

Company self-disclosure with selective metric choice. Netflix discontinued quarterly subscriber reporting and now chooses which engagement figures to publish; the 'view hours' series is self-defined and unaudited.

WARNING for Phase 2: the IR site renders financials client-side and returned no figures to automated fetch during this research. Use the RSS press-release feed or SEC EDGAR filings rather than the IR landing pages. The engagement-vs-revenue divergence (H1 2026: revenue +13%, view hours +2%) is the single most important number this source publishes.

Coverage: Global, reported by four regions · Back to 2002 IPO

Access (inherited): public_page · free

Public company disclosure; freely citable with attribution

Self-disclosure; segment definitions have been restated repeatedly, and Disney stopped headlining Disney+/Hulu subscriber counts in FY2026, which removes a comparison the company no longer benefits from.

Earnings PDFs are hosted on a q4cdn path (s206.q4cdn.com/979796730/...) and ARE directly fetchable even when the IR landing page is not — fetch the PDF, not the page. The Sports segment P&L is the sector's best public read on sports-rights cost inflation.

Coverage: Global · Decades; segment definitions restated 2023 and again for FY2026

Access (inherited): public_page · free

Public company disclosure; freely citable

Self-disclosure. Heavy use of pro forma presentation to exclude Versant and Sky Germany, which flatters the growth rate; read the reported figure alongside the pro forma one.

Individual release detail pages at /news-releases/news-release-details/<slug> are fetchable even when the index is not. Best public source for US pay-TV video subscriber losses and for the Peacock profitability inflection.

Coverage: US, UK, EU · Back to the 1990s

Access (inherited): public_page · free

Public company disclosure; freely citable

Self-disclosure by the acquirer in a contested, litigated merger; the WBD commentary is advocacy and should be labelled opinion, not fact. Efficiency targets are management guidance.

Shareholder letters are served from ir.paramount.com/static-files/<uuid> and are directly fetchable. This is currently the best single disclosure of the linear decline curve: it separates affiliate revenue (-6%) from advertising (-14%) where most peers report a blended line. Tier1 priority because of the live merger.

Coverage: Global · Reset at the Skydance combination; prior-period figures presented as 'predecessor'

Access (inherited): public_page · free

Public company disclosure; freely citable

Self-disclosure by a company under an agreed acquisition; disclosure is constrained by merger-agreement terms.

KNOWN BLOCKER: the financial-news index renders through a client-side year selector and returned no releases to automated fetch during this research; WBD's Q2 2026 segment detail could not be obtained. Phase 2 must source WBD through SEC EDGAR or the email-alert feed instead. Its Global Linear Networks line is the best single measure of the cable decline curve and is currently missing from this sector's coverage.

Coverage: Global · Back to the 2022 WarnerMedia/Discovery combination

Access (inherited): public_page · free

Public company disclosure; freely citable

None as to authenticity — these are filed under legal liability. Segment definitions are chosen by the registrant, so cross-company comparability is limited.

ACCESS NOTE: during this research the browse-edgar path was blocked by robots.txt for the fetch tool and direct curl returned proxy 403. Phase 2 should use the data.sec.gov JSON submissions API (https://data.sec.gov/submissions/CIK##########.json) with a proper User-Agent rather than the legacy browse-edgar CGI. This is the authoritative fallback whenever an IR site renders client-side, which most of them do.

Coverage: US-listed registrants · Back to 1994

Access (inherited): api · free

Public domain. SEC fair-access policy requires a declared User-Agent with contact details and rate limiting to 10 requests/second.

UK-only. Ofcom has a public-service-broadcasting remit, which shapes which comparisons are foregrounded (broadcaster share is presented prominently), but the underlying BARB and panel data are presented transparently.

The single best non-US source in this sector and the only one giving a clean age-cohort breakdown. The 2026 edition's finding that UK 16-24 weekly broadcaster reach fell from 43% (2022) to 26% (2025) is the most consequential number in this dossier for long-range modelling. Published as a PDF; the landing page carries almost no figures, so fetch the PDF directly.

Coverage: United Kingdom, with nation-level breakdowns · Annual since 2018, with comparable series back further

Access (inherited): bulk_download · free

Open Government Licence v3.0 — free reuse with attribution

Owned by Amazon, which is a competitor in this sector. Reporting depends on distributor-supplied figures and is revised. International coverage is uneven by territory.

RELIABILITY WARNING from this research: year-table pages rendered truncated and mis-dated results to automated fetch — a '2026' request returned a table topped by a 2025 release. Do not trust a single automated pull for an annual total; cross-check against The Numbers and Comscore. Reports revenue only, never admissions.

Coverage: US domestic plus most international territories · Back to the 1970s for domestic

Access (inherited): public_page · free

IMDb terms prohibit systematic scraping and redistribution; a data licence is required for bulk use

Independent of the studios, which is an advantage over distributor-affiliated sources. Small operation; ticket counts are derived from average ticket price and are therefore estimates, not counts.

The ONLY readily accessible public source giving annual ADMISSIONS alongside revenue, which is what separates real theatrical recovery from ticket-price inflation. Source of the 2019 (1,225.6m tickets) versus 2025 (754.1m tickets) comparison that governs this sector's theatrical analysis. Note the ticket counts are modelled from average ticket price.

Coverage: US domestic plus international · Back to 1995 for the annual market series

Access (inherited): public_page · freemium

Free for non-commercial reference; commercial and bulk use requires a licence from Nash Information Services

Commercial measurement vendor; the free public insights layer is marketing for the paid product, so the most useful granularity sits behind licensing.

PAYWALL FLAG for Phase 2 budgeting: Comscore is the industry's actual box-office currency but the public page carries no usable figures — it returned none to automated fetch. If this sector needs authoritative theatrical data, this is where the licensing spend has to go.

Coverage: Global theatrical · Multi-decade via Rentrak heritage

Access (inherited): licensed · enterprise

Licensed data; redistribution prohibited without agreement

A union presenting its own bargaining outcome; gains are foregrounded and concessions are not. The 3%/year wage figure is factual; the characterisation of AI protections as 'stringent' is advocacy.

In the absence of US statute this IS the AI rulebook for scripted production. The 2026 contract runs 2026-07-01 to 2030-06-30, so the terms are fixed for four years — any forecast of AI labour displacement inside that window must clear the synthetic-performer 'significant additional value' test and the notice-and-bargain requirement. Monitor for synthetic-performer bargaining notices, which would be the first real adoption signal.

Coverage: United States · Contract cycles back decades; memoranda of agreement published per cycle

Access (inherited): public_page · free

Publicly posted; contract documents freely readable and citable

Union self-presentation, same caveat as SAG-AFTRA.

GAP FLAG: a 2026 WGA Theatrical and Television Basic Agreement exists (a Memorandum of Agreement, summary and contract-changes FAQ are referenced on this page) but the summary PDF path 404'd during this research, so the writers' AI, residual and wage terms are UNKNOWN in this dossier while the equivalent SAG-AFTRA terms are fully documented. Closing this asymmetry should be the first task of any re-run.

Coverage: United States · Contract cycles back decades

Access (inherited): public_page · free

Publicly posted; freely citable

Enforcement-agency statements reflect the enforcement posture of the administration in office; the reasoning is advocacy for a decision already taken. Directionally authoritative on outcome, contestable on analysis.

Directly and reliably fetchable, unlike most company IR. The 2026-06-12 closing statement is unusually analytical — it names A24 and Blumhouse as theatrical competitors and rejects the Disney/Fox precedent as inapplicable post-pandemic — and is therefore a substantive source on market structure, not just on the merger outcome.

Coverage: United States · Multi-decade archive

Access (inherited): public_page · free

US government work, public domain

Litigants' filings state a case, not a finding; allegations are not facts. The political incentives of elected state AGs are a live factor in the framing.

NEW AND STRUCTURALLY IMPORTANT. State AGs succeeded in halting a transaction the DOJ had cleared — a governance pattern that generalises well beyond media and should be monitored across sectors, not just here. Court docket monitoring (PACER) would be the higher-fidelity version of this source if the budget allows.

Coverage: United States, state level · n/a — action commenced 2026

Access (inherited): public_page · free

Public records; court filings freely citable

Procedural record rather than analysis; politically responsive on ownership and transfer policy.

Broadcast licence transfers are a required approval for the Paramount/WBD transaction, so the FCC docket is an independent read on deal timing that does not depend on either party's characterisation. Also the authoritative source on retransmission-consent policy, which governs the affiliate fees now in outright decline.

Coverage: United States · Decades; EDOCS and ECFS searchable

Access (inherited): api · free

US government work, public domain

Policy advocacy alongside factual reporting; European-works quota reporting is framed to support the policy.

The European-works quota and investment obligations are the main regulatory floor under streamer commissioning in Europe and therefore a counterweight to the sector-wide content-spend reduction. Also carries AI Act implementation timing, which per the macro brief slipped for high-risk employment obligations to 2027-12-02.

Coverage: European Union · AVMS Directive from 2010, revised 2018

Access (inherited): public_page · free

EU reuse policy — free reuse with attribution

A promotion agency for a national export industry; figures are presented to demonstrate policy success. Most detailed material is Korean-language only.

UNTAPPED IN THIS DOSSIER — listed as a Phase 2 requirement, not as a source used. Non-English content is over a third of Netflix viewing and no Korean primary source could be retrieved within this session's capacity. K-content production-cost and export data is the missing evidence base for the international-content trend.

Coverage: South Korea, with global export data · Annual content-industry statistics for over a decade

Access (inherited): public_page · free

Attribution required; check per-report terms

Advertising-industry trade press with named bylines; audience is media buyers, so framing favours the advertiser's view. Analyst expectations are reported as such rather than verified.

Useful precisely because it covers measurement-industry mechanics that no primary source announces — it was the only source that flagged the Nielsen DASH recalibration and the analyst expectation that streaming's reported share will fall. Tier B: corroborate its numbers against the Nielsen release before use.

Coverage: United States · Archive back to the early 2000s

Access (inherited): public_page · freemium

Standard copyright; some articles metered

Trade press dependent on studio access and studio advertising; single-sourced 'sources say' reporting is common and deal figures often precede confirmation. Penske owns both titles, so Variety and Deadline are ONE source organisation, not two, for source-diversity scoring.

ACCESS WARNING: both titles returned HTTP 402 to automated fetch during this research. Fastest source on deal and litigation developments, and the only practical way to track merger litigation between court filings — but per this contract's rules it can never be the sole support for a numeric claim, and the two titles count as one organisation.

Coverage: US-centric with global coverage · Variety archive back to 1905

Access (inherited): public_page · freemium

Standard copyright; hard paywall on some articles

Alphabet does not break out total YouTube revenue — only YouTube advertising — which understates YouTube's economic scale in this sector by excluding subscriptions and YouTube TV.

GAP FLAG: IR pages render client-side and returned no figures to automated fetch during this research, so YouTube's commercial scale is asserted in this dossier only in Nielsen viewing-share terms. Phase 2 should pull the 10-Q from EDGAR instead. Even then, the disclosure gap is structural — the largest TV distributor in the US publishes less segment detail than the smallest studio.

Coverage: Global · Back to the 2004 IPO

Access (inherited): public_page · free

Public company disclosure; freely citable

Studio trade body; the THEME report is produced to demonstrate industry economic contribution and consistently frames the sector at its most favourable. Methodology is disclosed but the selection of metrics is advocacy.

Use for global theatrical and home-entertainment framing, always labelled 'estimate' with the MPA named as modeller per the contract's §0.3 rule. Never use as the sole support for a market-size claim.

Coverage: Global, US-weighted · Annual since 2007 in its current form

Access (inherited): bulk_download · free

Free download with attribution; figures are MPA estimates and must be labelled as such

Issuer-controlled framing and timing; releases appear when they favour the issuer. Authoritative on what was announced, not on its significance.

Practical Phase 2 pattern established by this research: corporate press-release DETAIL pages and PDFs are reliably fetchable while IR landing pages and index pages are not, because the latter render client-side. Poll the RSS feeds and fetch the resulting detail URLs directly. This is how the Disney/OpenAI Sora agreement and the Disney CEO succession were captured here.

Coverage: Global · Varies by issuer, typically 5-10 years

Access (inherited): rss · free

Freely citable with attribution

Apple's own framing of changes it was compelled to make; describes court-ordered and regulator-driven concessions as developer benefits. The numbers are authoritative, the narrative is not.

The single highest-value feed in this sector. Every commission change in the EU, Japan, China and Brazil was announced here first, with exact percentages and effective dates. Pair each news item with its /support/ fee-schedule page, which carries the worked examples.

Coverage: Global, with per-storefront pages for EU, Japan, Brazil, China · Archive back to 2010s

Access (inherited): rss · free

Copyrighted; quote with attribution. No bulk redistribution of developer documentation.

Google's own statement of its fee structure; framed to minimise the appearance of court-compelled change. Numbers authoritative.

No RSS and no version history, so Phase 2 must diff the page itself. This is where the post-Epic-settlement US/UK/EEA fee split (service fee plus separate 5% billing fee) is defined.

Coverage: Global, with US/UK/EEA, Korea and India carve-outs · Current version only; no published change log — snapshot it on every poll

Access (inherited): public_page · free

Copyrighted help documentation. Attribute; do not republish wholesale.

Issuer-controlled. Roblox leads with revenue growth when bookings decelerate and with bookings when revenue lags. Always take both from the supplemental tables.

The IR site is JavaScript-rendered and does not yield content to a plain fetch; go to SEC EDGAR for the 8-K exhibit instead. Essential for DevEx payouts, DAU, hours engaged and bookings — the four numbers this sector most often conflates.

Coverage: Global · Back to the 2021 direct listing

Access (inherited): public_page · free

Public issuer disclosure; freely citable with attribution.

None. Primary regulatory disclosure. Coverage bias is structural: it only sees US-listed issuers, so EA's take-private, Epic and Valve are now invisible here.

Covers RBLX, TTWO, U, MSFT, META, NTES, SONY ADR. Note the shrinking coverage: EA left the index on 2026-08-04, so this source is getting less complete over time, not more.

Coverage: US-listed issuers · Back to 1994

Access (inherited): api · free

Public domain. Fair-access rate limits apply; declare a User-Agent per SEC policy.

Issuer-controlled but unusually clean: Nintendo reports sell-in units with no marketing gloss. Shipments, not sell-through.

A plain static HTML table that fetches reliably — rare in this sector. The authoritative source for Switch 2 hardware and software units.

Coverage: Global, with regional splits · Back to the Famicom era

Access (inherited): public_page · free

Public issuer disclosure; citable with attribution.

Chinese-language primary regulatory publication. No commercial bias; the selection itself is the policy instrument, so absence from a batch is meaningful data.

The primary source behind every China-approvals figure. Phase 2 should parse these directly rather than relying on secondary counts; Niko Partners and PocketGamer.biz are the English-language fallbacks.

Coverage: China · Back to 2016, with gaps during the 2018 and 2021-22 freezes

Access (inherited): public_page · free

Chinese government publication. Check terms before redistribution; translation is the practical barrier, not licensing.

Commercial research firm; the free Substack is a marketing surface for paid reports, so headline figures are published but methodology is not. Forecasts are revised frequently.

The best English-language coverage of NPPA approvals and Asian market structure. Its full market models are paywalled — this is one of the places Phase 2 will need a licensing budget.

Coverage: China, Southeast Asia, India, MENA · Back to roughly 2020 on the public feed

Access (inherited): rss · freemium

Substack terms; attribution required. Paid reports are licensed and must not be redistributed.

Volunteer-maintained and openly acknowledged to be incomplete. Systematically undercounts: it captures announced, English-language, press-covered events and misses contractors, QA vendors, attrition, hiring freezes and most non-English markets.

The de facto sector reference despite being Tier C. Use it for direction and named events, never as a precise total, and never alongside third-party full-year forecasts without saying which is which — that conflation is how ~4,600 actual becomes a ~14,000 headline.

Coverage: Global in principle, heavily US/UK/EU in practice · Back to 2022

Access (inherited): public_page · free

Community dataset; an AirTable view is offered for free-account access. Credit the maintainers; confirm redistribution terms before bulk reuse.

Self-reported by developers and unaudited, so it is a floor on AI use, not a measure of it. The January 2026 exemption of code and game logic breaks comparability across that date.

The only hard denominator for generative-AI adoption anywhere in the sector. SteamDB and GameDiscoverCo derive from the same underlying data and are useful cross-checks.

Coverage: Global PC market · Disclosure field exists from January 2024

Access (inherited): public_page · free

Steam Web API terms; no aggressive scraping of the storefront. Use the official Web API and respect rate limits.

Independent and methodologically transparent, which is rare here. Steam-centric by construction, so console and mobile are out of scope. Paid tier funds the analysis.

The best public work on Steam revenue distribution, price deflation and player sentiment toward AI. Its 2021-versus-2025 hit-cohort analysis is the cleanest evidence available that premium PC prices are falling.

Coverage: Global PC and console · Back to 2020

Access (inherited): rss · freemium

Substack terms; attribution required for figures. Paid data tools are licensed per seat.

Point-of-sale plus panel methodology; mobile comes from Sensor Tower, so the total is a blend of measurement approaches. Figures are revised after initial publication (the 2025 total moved from $60.7bn to $60.8bn).

The authoritative US spending series. Only headlines are free — the granular series is a licensing cost Phase 2 must budget for. ESA's press releases are the free channel.

Coverage: US, with some European coverage · Back to the 1990s under NPD

Access (inherited): public_page · paid_high

Full data is licensed and expensive; headline figures are released publicly via ESA and Mat Piscatella's public commentary.

Industry trade body. Advocacy-oriented: emphasises player numbers, family engagement and economic contribution, and does not re-base its 250,000-jobs figure for four years of layoffs.

Useful for demographics and the headline US spending total. Treat employment and economic-impact figures as advocacy estimates with a stale base year — the 2024 economic impact report is still the current citation in 2026.

Coverage: US · Annual since 2003

Access (inherited): public_page · free

Free to cite with attribution. Underlying survey microdata is not released.

Issuer-controlled. Tencent reports domestic and international games separately, which is genuinely useful; both report in RMB, so constant-currency framing matters when comparing international lines.

Tencent's interim report PDF fetches reliably and carries the evergreen-title DAU commentary that is the best available evidence on live-service concentration in the largest market.

Coverage: China and international · Back to the mid-2000s

Access (inherited): public_page · free

Public issuer disclosure under HKEX and SEC rules; citable with attribution.

Issuer-controlled. Reality Labs bundles Quest, Horizon Worlds and Ray-Ban smart glasses into one line, so segment growth can mask a declining VR games business inside it.

The only hard financial number in consumer XR, since no manufacturer publishes unit shipments. Always separate the smart-glasses contribution before citing it as VR evidence.

Coverage: Global · Segment reported separately since Q4 2020

Access (inherited): public_page · free

Public issuer disclosure; citable with attribution.

Enthusiast press with a structural optimism bias toward the category they cover, though both have reported the 2026 studio closures honestly. Named bylines.

The fastest reporting on XR hardware and studio closures. Neither has access to shipment data, so treat all unit figures they carry as third-party estimates.

Coverage: Global · Back to 2014

Access (inherited): rss · free

Standard editorial copyright; quote with attribution and link.

Trade press with a conference and events business attached, so coverage tilts toward topics with commercial partners. Named bylines and generally reliable sourcing.

The broadest daily coverage of mobile, UGC and app-store topics, and the fastest route to third-party market reports that are otherwise gated.

Coverage: Global, strongest on mobile and Asia · Back to 2005

Access (inherited): rss · free

Editorial copyright; quote with attribution.

Trade press owned by the GDC organiser; conference-adjacent agenda-setting. Strong on labour and production topics. Named bylines.

The best coverage of unionisation, studio closures and production practice, including the credited generative-AI uses in shipped titles.

Coverage: Global · Back to 1994 as Gamasutra

Access (inherited): rss · free

Editorial copyright; quote with attribution.

Aggregates third-party market reports rather than producing primary data, so it inherits the biases of the firms it summarises — including vendor-sponsored research. Its value is coverage breadth, not independence.

The most efficient way to see which market reports have been published. Always chase the underlying report before citing a figure; several sources it summarises are vendor-sponsored.

Coverage: Global · Back to 2021

Access (inherited): rss · freemium

Substack terms; the underlying reports carry their own licences — check each before citing.

Heavy betting-operator advertising and sponsorship presence, which is material given the sector's pivot toward betting and prediction markets. Disclose this when citing.

The only consistent business-side esports coverage, but the commercial entanglement with betting operators is a real bias on precisely the topic that currently matters most in esports.

Coverage: Global · Back to 2016

Access (inherited): rss · free

Editorial copyright; quote with attribution.

Regulator's own record. Institutional perspective favours finding non-compliance, but decisions, specification proceedings and designations are primary documents.

The upstream cause of Apple's October 2026 EU fee schedule. Monitor alongside Apple's own developer pages to see the gap between what was ordered and what was implemented.

Coverage: EU · Since DMA designation in 2023

Access (inherited): public_page · free

EU documents are generally reusable under Decision 2011/833/EU with attribution.

Primary court record, no bias. CourtListener's RECAP coverage is incomplete for documents nobody has purchased.

The determination of what Apple may charge on US external-link purchases is live on this docket as of 2026-09-15. This is the single highest-value legal feed for the sector.

Coverage: US · Full docket history

Access (inherited): api · freemium

PACER charges per page; CourtListener RECAP is free where documents have been contributed. Court records are public.

Regulator's own record; English pages lag the Japanese originals, so the English feed is a delayed and partial view.

Primary source for the Japanese regime that produced the 21%-plus-5% structure. Read alongside Apple's Japan distribution support page, which carries the actual rates.

Coverage: Japan · MSCA materials since 2024

Access (inherited): public_page · free

Japanese government publication; generally reusable with attribution.

Korean regulatory filing system; primary and complete for Korean issuers. Korean-language first, with partial English.

Korea is a top-four games market and the centre of gravity for PUBG and MMO live-service economics. Krafton's record 2026 quarters are visible here before they reach English trade press.

Coverage: South Korea · Back to the early 2000s

Access (inherited): api · free

Korean regulatory disclosure; public and citable. Check DART API terms for bulk use.

Trade bodies advocating for their national industries; the GSD panel excludes some digital storefronts, so digital PC in particular is undercounted.

The only systematic non-US sales measurement. Essential for correcting the US-centric bias of ESA and Circana figures, and for German and French market sizing.

Coverage: Europe, with national breakdowns · Back to the 2010s

Access (inherited): public_page · freemium

Aggregate figures are published free; the detailed GSD panel is licensed.

Third-party estimates modelled from sampled data, not platform-reported truth. Systematically miss third-party Android stores in China and now miss alternative EU, Japanese and Brazilian distribution entirely.

Important caveat for Phase 2: as alternative distribution grows under the DMA and MSCA, these estimates will diverge from true mobile spend in a way that is invisible unless flagged. Their blind spot is precisely the trend this dossier's top-ranked trend describes.

Coverage: Global mobile · Back to roughly 2012

Access (inherited): public_page · freemium

Free insights posts are citable with attribution; the underlying data is licensed.

Issuer-produced. Presents 'core operations' and adjusted EBITA measures that exclude repositioning and integration costs, which flatters post-merger comparability. Revenue is not reported on the same ex-pass-through basis as WPP or Publicis.

Backbone source for agency-sector economics. Pair with the quarterly transcript PDF, which carries the headcount and synergy detail the release omits.

Coverage: Global, US-weighted (59% of Q2 2026 core revenue) · Quarterly releases back to the 1990s; 10-K/10-Q on EDGAR from 1994

Access (inherited): public_page · free

Public issuer disclosure; freely quotable with attribution. No redistribution of full PDFs.

Issuer-produced. Reports 'revenue less pass-through costs' and 'headline' operating margin, both non-IFRS measures chosen by management. Average rather than period-end headcount is used in some disclosures, which is not comparable with peers.

The single most informative source on agency-sector stress because WPP is the one large holdco in decline and must explain it. Elevate28 savings tracking lives here.

Coverage: Global, UK-headquartered · Annual reports back to the 1990s

Access (inherited): public_page · free

Public issuer disclosure under UK Listing Rules; quotable with attribution.

Issuer-produced and the sector's strongest performer, so management framing is consistently positive. Reports 'net revenue' and 'headline' margin. Does not disclose headcount in the H1 release itself.

Non-US primary source and the control case for the 'AI deflates agency revenue' thesis: Publicis is growing revenue and headcount simultaneously.

Coverage: Global; North America 61% of H1 2026 net revenue · Results archive back to the early 2000s

Access (inherited): public_page · free

Public issuer disclosure under AMF rules; quotable with attribution. French and English versions published.

Issuer-produced. Japanese-domestic and international segments have materially different economics and are sometimes blended in commentary. English pages lag the Japanese originals.

The sector's principal non-Western primary source. IMPORTANT: several English IR sub-pages returned 404 or redirect loops during this research (2026-09-15) and 2026 results could not be retrieved. Phase 2 should poll the Japanese-language IR library and TDnet instead.

Coverage: Japan and international · Results archive back to the 2001 listing

Access (inherited): public_page · free

Public issuer disclosure under TSE rules; quotable with attribution.

Issuer-produced. Newly separately listed (December 2024), so comparative history is short. Dutch-domiciled, French-operated.

Smallest of the majors and the least covered, so it is where sector-wide stress shows up first. The IR landing pages resisted automated retrieval during this research.

Coverage: Global, Europe-weighted · Since the December 2024 separation from Vivendi

Access (inherited): public_page · free

Public issuer disclosure under Euronext Amsterdam rules.

Issuer-produced. Ad impressions and average-price-per-ad are Meta-defined metrics with no external audit and changing surface mix (Reels, Threads) inside them.

The price-per-ad line is the single best public read on whether advertising unit prices are inflating or deflating at scale. +12% in Q2 2026.

Coverage: Global with four reported regions · Back to the 2012 IPO

Access (inherited): public_page · free

Public issuer disclosure; quotable with attribution.

Issuer-produced. Google Network revenue (the ad tech business at issue in the antitrust case) is disclosed but not decomposed by product.

FLAG FOR PHASE 2: the abc.xyz IR pages are JavaScript-rendered and every automated retrieval attempt failed during this research (2026-09-15). Alphabet Q2 2026 advertising segment revenue is consequently a recorded data gap. Use SEC EDGAR 10-Q XBRL (CIK 0001652044) instead of the IR site.

Coverage: Global · Back to 2004

Access (inherited): public_page · free

Public issuer disclosure; quotable with attribution.

None as a source; the filings themselves carry issuer framing. Coverage is US registrants only, so WPP, Publicis, Dentsu and Havas are absent.

The authoritative fallback when issuer IR sites are unscrapeable, which in this sector is often. XBRL company-concept API gives clean revenue series without parsing HTML.

Coverage: US registrants · Back to 1994

Access (inherited): api · free

Public domain US government data. Fair-access rules require a declared User-Agent and rate limiting; the browse-edgar CGI is disallowed to some crawlers by robots.txt.

IAB is funded by sellers of digital advertising. Its forecasts are survey-based (n>200 brand and agency decision-makers for the Outlook Study) and consistently optimistic relative to outturn. The IAB/PwC revenue report is a census of member-reported revenue, which is stronger than the survey work but excludes non-members.

PAYWALL NOTE FOR PHASE 2: every substantive IAB report body sits behind a registration wall. Only headline numbers were retrievable during this research. Budget for account access, not licensing.

Coverage: United States · IAB/PwC internet ad revenue series back to 1996 (30 editions)

Access (inherited): public_page · freemium

Headline figures are public; full reports require free account registration and are not redistributable. Attribution to IAB/PwC required.

Industry-funded but structurally independent with congressional-oversight origins. Publishes accreditation status, not performance opinions. Does not disclose audit findings.

Underrated as a leading indicator. The Digital Advertising Auction Transparency Standards (2026-01-29) predate the FTC's Amazon auction complaint by seven months and define the benchmark both antitrust cases implicitly reference.

Coverage: United States · Back to the 1960s; web news archive shorter

Access (inherited): public_page · free

Standards documents are freely published; accreditation marks are controlled.

Advertising-funded trade publication with an events business dependent on the vendors it covers. Named bylines and unusually strong technical and legal literacy; the most reliable secondary source in this sector for ad tech mechanics.

First to report the Brinkema remedies ruling and the FTC/Amazon auction complaint detail. Its court coverage is the practical substitute for PACER access.

Coverage: US-centric with EU coverage · Back to 2008

Access (inherited): rss · freemium

Standard copyright; quote with attribution and link.

Advertising- and events-funded. Heavy use of anonymous buyer sourcing, which makes claims hard to verify but surfaces pricing and fee information no filing contains. Named bylines throughout.

The only source found during this research carrying actual ChatGPT ad pricing ($60 to $25 CPM in ten weeks; $3-$5 CPC) and the OpenAI $1bn run-rate disclosure, with an explicit correction that booked revenue is roughly $330m. That distinction is the kind of thing Phase 2 must preserve.

Coverage: US and UK · Back to 2008

Access (inherited): rss · freemium

Standard copyright; some content behind a Digiday+ subscription.

Agency-industry trade press with awards and events revenue from the companies covered. Strongest on UK and EMEA agency personnel, account moves and pitch outcomes.

Best single source for account wins and losses, which is the leading indicator for holding-company organic growth two to four quarters out.

Coverage: UK, US, Asia-Pacific editions · Back to the 1960s in print

Access (inherited): rss · freemium

Largely paywalled; standard copyright.

MATERIAL CONFLICT: Ebiquity sells media auditing and effectiveness consulting, so it has a direct commercial interest in the finding that media is opaque and measurement is broken. Its numbers are nonetheless the best advertiser-side series available and it co-publishes with the WFA.

Source of the principal-media adoption figures (96% aware / 58% use / 90% uncertain) and the MMM adoption gap (~80% run it, 15.4% let it drive budgets). Always label the commercial interest.

Coverage: Global, UK/US-weighted · Blog archive several years; audit dataset proprietary and much deeper

Access (inherited): public_page · free

Blog content freely quotable; underlying audit data is licensed.

Advertiser-side trade body, so systematically sympathetic to buyer positions on transparency and remuneration. Membership skews to very large multinational advertisers, so findings are not representative of mid-market.

Co-publisher with Ebiquity of the Media Budgets Survey and the Paid Media Effectiveness Handbook. The research index page did not render usable content during this research; Phase 2 should poll the Ebiquity co-publications instead.

Coverage: Global · Research archive several years

Access (inherited): public_page · freemium

Much of the research is member-only.

US advertiser-side trade body. Its 2016 media transparency study set the sector's contracting agenda and it retains a policy interest in that framing. Most research is member-gated.

Also a useful free relay of IAB and other trade forecasts, and tracks US state privacy rulemaking affecting marketing data.

Coverage: United States · Research archive back many years

Access (inherited): public_page · freemium

Member-only for full reports; news summaries public.

Subscription research house. Methodology for global ad spend forecasts is proprietary and its figures differ from IAB's on scope and geography, which is routinely lost in secondary citation.

PHASE 2 LICENSING BUDGET GOES HERE. WARC is the only source found that surfaced independent academic evidence on AI creative performance (the Columbia 300,000-ad study) rather than vendor lift claims.

Coverage: Global · Advertising effectiveness case database back decades

Access (inherited): licensed · enterprise

PAYWALLED. Enterprise licence required for the data; headlines are public. Redistribution prohibited.

Independent single-analyst forecasting practice, which is a strength (no vendor sponsorship) and a weakness (one modeller, no published methodology, no peer review). Its numbers were the most widely syndicated concentration estimates in 2026, so apparent multi-source agreement often traces back to this one shop.

Source of the Big Three concentration series (56% of US ad revenue 2025) and the AI-directed-spend series (2% in 2023, 12% in 2026, 27% by 2030). Phase 2 must deduplicate: Digiday, Marketing Brew and others all relay these as if independent.

Coverage: Global and US · Several years

Access (inherited): email · freemium

Newsletter distribution; quote with attribution.

Owned by Semrush, a vendor selling search and AI-visibility tooling. Much of its quantitative content is agency-contributed analysis of private account data with no disclosed sample size or methodology, which makes those posts Tier C on this contract's test even where the direction is probably right.

Practically the only public window onto search-advertising mix shifts (query-length and Shopping-ad CTR changes post-AI Mode). Treat every figure as an unaudited agency panel and always record the causation disclaimer the authors themselves give.

Coverage: US-centric · Back to 2006

Access (inherited): rss · free

Standard copyright.

Issuer-produced. 'Customer retention above 95%' is a company-defined, externally unaudited metric that has been repeated for a decade and conceals spend migration inside retained accounts.

The sector's cleanest single indicator of whether independent ad tech can grow. Q2 2026 revenue +3% against an IAB-forecast +12.3% market is the number to watch quarterly. Note the investors.thetradedesk.com paths 404 to automated retrieval; use the thetradedesk.com press room.

Coverage: Global · Back to the 2016 IPO

Access (inherited): public_page · free

Public issuer disclosure.

Court-reporting wire with reporters physically present at hearings. No advertising-industry commercial interest, which makes it a valuable non-endemic check on trade-press characterisation of rulings.

Use alongside PACER/CourtListener for US v. Google (EDVA) and FTC v. Amazon. The sealed-memorandum and 30-day-final-judgment detail came from here, not from the endemic trade press.

Coverage: US federal and state courts · Back to the 1990s

Access (inherited): rss · free

Standard copyright.

None. Coverage depends on crowd-sourced RECAP uploads, so some documents lag or are missing, and sealed filings are absent by definition.

HIGHEST-PRIORITY ADDITION FOR PHASE 2. The Google remedies memorandum was sealed until on or about 2026-09-16 and the proposed final judgment was due within 30 days of 2026-09-02. Both land on this docket, not in any press release.

Coverage: US federal courts · Decades, with uneven RECAP coverage before 2010

Access (inherited): api · free

Free Law Project terms; bulk use requires API token and rate limiting. Court opinions are public domain.

Enforcement-agency framing: complaints are allegations presented adversarially. Figures such as the $20bn Amazon estimate are litigation positions, not audited findings.

Primary source for FTC v. Amazon (filed 2026-08-26 with 22 state attorneys general over Sponsored Products auction mechanics). Pair with the state AG press offices, which sometimes publish complaint exhibits the FTC does not.

Coverage: United States · Back to the 1990s online

Access (inherited): rss · free

US government public domain.

Pay-to-present environments. Almost every quantitative claim made on stage is vendor-sponsored and unaudited. Useful as a leading indicator of narrative, explicitly not of adoption.

Include only as a velocity input under the contract's rule that attention is not adoption. The gap between Cannes narrative and the following quarter's holdco organic growth is itself a useful signal.

Coverage: Global · Decades

Access (inherited): public_page · paid_high

Session content often restricted to attendees.

A regulator reporting on the effectiveness of its own regime. Ofcom has an institutional interest in the Online Safety Act appearing to work, and its 2026 report frames traffic declines at compliant sites favourably. It is also unusually candid — it publishes its own non-representativeness caveat and criticises named platforms. Treat conclusions as interested, figures as sound.

The single highest-value source for this sector. The Use of Age Assurance Report 2026 (2026-07-15) is the only large-scale public dataset on age assurance in operation. Two further statutory outputs are scheduled and should be diarised: the rapid assessment to Parliament on under-16 age checks (October 2026) and the statutory report on app store protections (January 2027). Enforcement bulletins carry fine amounts and investigation counts.

Coverage: United Kingdom · Online Safety Act duties from July 2025; Ofcom communications data back to 2003

Access (inherited): public_page · free

Open Government Licence v3.0 — reuse permitted with attribution

None on the data itself — these are filings made under legal liability. Selection bias applies: only US-listed companies file, so ByteDance, Discord, Telegram and every private consumer app are absent, which is a large share of the sector.

IMPORTANT ACCESS NOTE for Phase 2: the legacy /cgi-bin/browse-edgar endpoints are disallowed by robots.txt for automated agents. Use data.sec.gov/submissions/CIK{10-digit}.json to enumerate filings, then fetch documents under www.sec.gov/Archives/. This route worked for Pinterest (CIK 0001506293) in this research cycle. Relevant CIKs: META 0001326801, SNAP 0001564408, PINS 0001506293, RDDT 0001713445, RBLX 0001315098, GOOGL 0001652044, AAPL 0000320193.

Coverage: US-listed issuers, global operations · back to 1994

Access (inherited): api · free

Public domain. SEC requires a declared User-Agent header and rate limiting to 10 requests per second.

Apple's own framing of changes it is making under legal compulsion; commission changes are presented as programme updates rather than as concessions. The factual content — rates, dates, API names, deadlines — is authoritative because developers must act on it.

The most under-monitored high-value source in this sector. Commission rate changes for the EU, China, Japan and Brazil, the Declared Age Range API, the Significant Change API, and the Social Media content descriptor and Time Allowance all appeared here first and only here. Phase 2 should parse every item for rate changes, age/consent APIs and jurisdiction-specific compliance deadlines.

Coverage: Global, with jurisdiction-specific notices · rolling archive, several years

Access (inherited): public_page · free

Apple copyright; quotation with attribution. No bulk redistribution.

Company self-reporting. Financial figures are audited and legally constrained; narrative framing in the newsroom is promotional. Meta's compliance posts (e.g. Australia) present its own enforcement favourably and omit business impact.

Watch the legal-matters paragraph in each earnings release, not just the metrics. The Q2 2026 release contains the sector's most consequential disclosure — youth-related US trials that 'may ultimately result in a material loss'. The newsroom separately carries teen-safety and attorneys-general announcements.

Coverage: Global · back to 2012 IPO

Access (inherited): public_page · free

Company copyright; factual figures freely citable with attribution

A regulator publishing research on its own regime that found the regime largely ineffective — a strong indicator of institutional honesty, but its framing of remedies is naturally interventionist.

ACCESS NOTE: esafety.gov.au returned a robots.txt fetch timeout for automated retrieval in this research cycle; the findings had to be obtained via secondary reporting. Phase 2 should budget for a manual or alternative retrieval path. This is the only jurisdiction with post-implementation outcome data on a minimum-age regime, which makes it disproportionately valuable.

Coverage: Australia · Social media minimum age regime from December 2025; eSafety research back to 2015

Access (inherited): public_page · free

Creative Commons BY 4.0 for most Australian Government material; check per document

The enforcing authority's own account of its rules. Guidance documents describe recommended behaviour rather than binding obligation, and the Commission has an interest in presenting voluntary guidelines as consequential.

ACCESS NOTE: several sub-paths (including /en/policies/age-verification) are robots-disallowed; the /en/library/ document pages are fetchable. Watch for Article 28 enforcement proceedings specifically — none was verifiable in this research cycle, and the first one will be a significant sector event. Also track the EU age-verification blueprint and Digital Identity Wallet timeline.

Coverage: European Union · DSA from 2022; enforcement steps published from August 2023

Access (inherited): public_page · free

Commission Decision 2011/833/EU reuse policy — free reuse with attribution

Enforcement agency communications carry political framing that changes with administration; the September 2025 AI chatbot release is explicitly branded to the sitting administration. Factual content — which companies received orders, what was requested, the Commission vote — is authoritative.

The Section 6(b) AI companion chatbot study (orders issued 2025-09-11 to seven companies) has not yet produced a public report. When it does it will be the single most authoritative document on companion-app engagement monetisation and minors, and should be treated as a scheduled high-priority event.

Coverage: United States · decades

Access (inherited): public_page · free

US Government work, public domain

Non-partisan and methodologically transparent — publishes field dates, sample sizes and margins of error. Limitations are inherent to survey research: self-reported behaviour, and US-only samples in most of the teen work.

The only source that can adjudicate the outcome question underlying minors-safety regulation. Its April 2026 finding that most teen users report social platforms neither help nor hurt their mental health sits awkwardly with the legislative premise and should be tracked across waves. Microdata is downloadable, which is rare and valuable.

Coverage: Primarily United States; some international series · teen and social media series back to roughly 2005

Access (inherited): public_page · free

Free to cite and reproduce with attribution; datasets available under a data use agreement

Company self-reporting. Snap discloses regional DAU in detail, which is unusually transparent, but withholds Snapchat+ subscriber counts and churn, which is the metric its subscription narrative rests on.

The sector's best early-warning indicator for mature-market attention shift, because Snap breaks out North America, Europe and Rest of World DAU separately. Q2 2026 North America -7% is the datapoint to track forward. Watch for any first disclosure of Snapchat+ subscriber numbers.

Coverage: Global with regional breakdown · back to 2017 IPO

Access (inherited): public_page · free

Company copyright; figures freely citable with attribution

Promotional. Payout figures are cumulative, rounded and chosen for effect, and cannot be differenced into an annual series. Economic-impact figures ($55bn US GDP, 490,000 jobs) are commissioned estimates whose modeller is not named on the page.

The origin point for the sector's most-laundered statistics. Phase 2 must record the publication DATE alongside any payout figure, because the 2024 '$70bn over three years' figure still outranks the current one in search. Also the authoritative source for monetisation policy changes such as the 2026-07-20 inauthentic content clarification.

Coverage: Global · back to 2005

Access (inherited): public_page · free

Google copyright; quotation with attribution

Third-party modelled estimates, not company disclosures. Covers Apple App Store and Google Play consumer spend only — it excludes web subscriptions, which understates true consumer spend for apps like ChatGPT by an unknown and probably large factor. Commercial interest in newsworthy headline numbers.

The only practical source for consumer AI app and AI companion app revenue in the absence of company disclosure. Always label its figures as `estimate` with the app-store-only caveat attached. The AI companion category figure available at this research date is from August 2025 and is stale — re-check for a 2026 update as a priority.

Coverage: Global app stores · app store data back to roughly 2012

Access (inherited): public_page · freemium

Attribution required; API access under commercial terms

Vendor research with a commercial interest in large headline numbers and in demonstrating measurement coverage. Methodology for panel-based estimates is not fully public. Independent of Appfigures, so useful for triangulation.

LICENSING NOTE for Phase 2: the headline State of Mobile report is free but the underlying category and time-spent data that would actually answer the attention-shift question is behind a commercial licence. If the sector needs defensible time-spent and category-share data, this is where budget has to go. Blog headlines alone are not sufficient.

Coverage: Global · roughly a decade of app market data

Access (inherited): public_page · freemium

Attribution required; full reports under commercial licence

An interested party. NetChoice is the plaintiff in most of the cases it lists and frames every statute as unconstitutional censorship. Useful as a comprehensive index of what is being challenged; not a neutral account of case merits or status.

Use as a discovery index, then verify each case's actual status through CourtListener or PACER. In this research cycle the listing's date semantics (ruling versus filing versus argument) could not be disambiguated, so statuses were not asserted. Phase 2 should pair this with a court-docket source.

Coverage: United States, state-level · active cases plus recent archive

Access (inherited): public_page · free

Organisation copyright; case names and public filings are public record

None editorially. Coverage of federal dockets is good but incomplete where documents have not been purchased into RECAP; state court coverage is patchier, which matters because much minors-safety litigation is state-level.

The correct primary source for the litigation layer that S-17-13 only indexes. Essential for tracking the Texas SB2420 injunction sequence, NetChoice challenges, and Meta's youth-related trials — the last of which Meta itself has flagged as potentially producing a material loss.

Coverage: United States federal and selected state courts · decades of opinions; RECAP dockets from 2009

Access (inherited): api · free

Free Law Project terms; most content public domain; API requires a token and reasonable rate limits

Company self-reporting with a strong interest in being seen as the safety leader among platforms serving minors. Creator earnings figures are company-stated and unaudited as a standalone metric; economic-impact figures are commissioned estimates.

ACCESS NOTE: corp.roblox.com redirects to about.roblox.com and individual article slugs were not reliably resolvable in this research cycle; ir.roblox.com is JavaScript-rendered and returned no data. Use SEC EDGAR (CIK 0001315098) for financials. Uniquely valuable because Roblox discloses annual creator earnings on a consistent basis — $923m in 2024, over $1.5bn in 2025 — which is the only clean creator-payout growth series in the sector.

Coverage: Global · back to 2021 listing

Access (inherited): public_page · free

Company copyright; figures citable with attribution

Company self-reporting; audited financials. Regional ARPU disclosure is unusually granular and genuinely informative about the geography of monetisation value.

ACCESS NOTE: the IR press release index page is JavaScript-rendered and returned no content; retrieve via SEC EDGAR (CIK 0001506293) instead, which worked. The US/Canada versus Europe versus Rest of World ARPU spread ($8.30 / $1.35 / $0.23 in Q2 2026) is the sector's best public quantification of geographic monetisation asymmetry.

Coverage: Global with US/Canada, Europe and Rest of World breakdown · back to 2019 IPO

Access (inherited): public_page · free

Company copyright; figures citable with attribution

A company under active FTC study and litigation exposure publishing about its own safety measures. Strong incentive to present a comprehensive safety posture; it discloses no usage or revenue impact from the measures described, which is the number that would make the disclosure assessable.

SAFETY-SENSITIVE SOURCE. The category bellwether: this company removed open-ended chat for under-18 users and built in-house age estimation. Content here concerns minors and crisis support and should be routed for editorial review before any external publication. Track for the first disclosure of engagement or revenue impact from the under-18 restriction.

Coverage: Global · back to roughly 2022

Access (inherited): public_page · free

Company copyright; quotation with attribution

Industry-supported non-profit. Its analysis is technically careful and it does not advocate simplistically, but its funding base means it is not a disinterested party on platform regulation. Disclose the funding model when citing.

The clearest public comparison of the companion-chatbot statutory wave — California SB 243, New York S-3008C, Maine LD 1727, Utah SB 452 — with obligations and effective dates set side by side. For a sector where the regulatory layer moves faster than the product layer, a good legislative tracker is worth more than most trade press.

Coverage: United States, with EU coverage · back to 2008

Access (inherited): public_page · free

Organisation copyright; quotation with attribution

Trade press serving the biometrics and identity industry. Named bylines and accurate reporting of regulator documents, but an evident commercial orientation toward the growth of the age-assurance market it covers. Reads regulator findings favourably for vendors.

The most consistent daily coverage of age-assurance deployments, vendor contract wins and regulator statements anywhere. Useful for discovery of vendor activity that no primary source publishes, but every number should be traced back to the underlying regulator or company document before use.

Coverage: Global · back to roughly 2012

Access (inherited): public_page · free

Publisher copyright; quotation with attribution

Company self-reporting; audited financials. Reddit's data-licensing revenue line is strategically important and reported only in aggregate as 'other revenue', which limits analysis of the AI-licensing business.

ACCESS NOTE: investor.redditinc.com returned a site-blocked error to automated retrieval in this research cycle; figures had to be obtained from a press-release republisher. Use SEC EDGAR (CIK 0001713445) as the reliable route. Reddit is the cleanest demonstration in the sector that margin sits in audience aggregation rather than creator payment.

Coverage: Global · back to 2024 IPO

Access (inherited): public_page · free

Company copyright; figures citable with attribution

Primary legal text, no editorial bias. The practical difficulty is fragmentation: fifty separate systems with inconsistent formats and no unified API, which is why secondary trackers dominate discovery.

The authoritative text behind SB 243, SB2420 and the rest of the 2025-26 wave. Phase 2 should use a commercial or non-profit multi-state tracker for discovery and fall back to the individual legislature site for authoritative text. Budget note: comprehensive multi-state legislative tracking is a paid category.

Coverage: United States, state-level · varies by state, typically back to the 1990s

Access (inherited): public_page · free

Statutory text is public domain

None. Primary legal text.

Free Speech Coalition v. Paxton (No. 23-1122, 2025-06-27) is the controlling decision for the sexual-content age-verification layer. The live question for this sector is whether the Court takes a case extending the analysis to general-purpose social media; a grant of certiorari in any NetChoice case is a named early indicator in the dossier's regulatory scenario.

Coverage: United States · complete

Access (inherited): public_page · free

Public domain

Promotional corporate communications from a private company with no filing obligation. TikTok discloses no creator payout figures and no audited user numbers, so almost everything material about the second-largest short-form video platform is unverifiable from primary sources.

COVERAGE GAP FLAG. The US newsroom returned only one 2026 item in this research cycle and nothing on US ownership structure, creator payouts, age assurance or AI labelling. TikTok's US corporate status as of 2026-09-15 could not be verified from any source and is the highest-priority unknown in this sector. Phase 2 should add the EU DSA transparency database and TikTok's DSA transparency reports, which are mandatory disclosures and therefore more informative than the newsroom.

Coverage: Global with regional newsrooms · back to roughly 2019

Access (inherited): public_page · free

Company copyright; quotation with attribution

Mandatory statements of reasons submitted by platforms themselves, so categories reflect each platform's own classification choices and volumes are not directly comparable across services. But it is compelled disclosure rather than voluntary communication, which makes it far more useful than any platform newsroom.

The single best structured, machine-readable dataset in this sector and the only one covering TikTok, Meta, Snap, Pinterest and X on a common schema with daily bulk downloads. It was not exercised in this research cycle because the search budget was exhausted before it could be, which is a notable omission — Phase 2 should make this a first-class ingestion target rather than a lookup.

Coverage: European Union, all designated platforms · from September 2023

Access (inherited): bulk_download · free

European Commission reuse policy — free reuse with attribution

Named bylines and editorial standards, but access-dependent: much reporting relies on company briefings and unnamed sources. Syndication means the same story appears across many domains, which can create a false impression of independent corroboration — under this contract that is one source, not several.

LICENSING NOTE for Phase 2: The Information is the highest-signal outlet for this sector's private-company reporting (ByteDance, OpenAI consumer strategy, platform product plans) and is fully paywalled. If budget exists for one trade subscription in this sector, that is the one. Note also that syndication mirrors such as dnyuz.com reproduce Business Insider content and must be attributed to the original publisher, not the mirror.

Coverage: Global, US-centric · decades

Access (inherited): public_page · freemium

Publisher copyright; substantial paywalled content. The Information is entirely subscription-based.

No commercial interest. Genuine limitation: monthly changes carry wide sampling error (July 2026 total change was -0.5% +/-0.8%, i.e. not statistically significant) and prior months are revised substantially. Category definitions are the sector's biggest trap: data-centre construction is reported inside the 'Office' line, which is why Census office construction rose 21.3% y/y in July 2026 while office CRE remained distressed.

Release is first business day of each month. The single most load-bearing series for this sector. Pull the detailed value-put-in-place tables, not the press release, to separate data centres from conventional office.

Coverage: United States, national with limited category detail · back to 1993 (current series); some categories to 1964

Access (inherited): bulk_download · free

US Government work, public domain. Citation of Census as source expected. No redistribution restriction.

None commercial. Multifamily (5+ unit) starts have enormous confidence intervals - the July 2026 y/y change was -7.1% +/-31.8%, which is meaningless on its own. Any pipeline story built on a single month of multifamily starts is noise. Permits are the more reliable leading series.

Joint Census/HUD release ~17th of month. Always carry the confidence interval; this series is routinely over-read in trade press.

Coverage: United States, four census regions · back to 1959

Access (inherited): bulk_download · free

Public domain US Government work.

None commercial. Median vs average price diverge sharply when builders shift mix toward smaller homes - in July 2026 median fell 0.9% y/y while average rose 5.4% y/y. Reporting only one of the two produces opposite conclusions about the same market.

The 'for sale by stage of construction' table (completed / under construction / not started) is the best public read on builder spec inventory risk.

Coverage: United States, four census regions · back to 1963

Access (inherited): bulk_download · free

Public domain US Government work.

None commercial. Headline construction payrolls mix residential and nonresidential, which in 2026 move in opposite directions; the specialty-trade split is needed to see it. Does not identify immigration status, so the labour-supply effect of the immigration decline is not directly measurable here.

Pair with JOLTS construction job openings and the CPS construction unemployment rate, which hit a record-low 3.1% in August 2026 - the cleanest evidence that labour is the binding constraint.

Coverage: United States, state and metro detail via SAE/CES · back to 1939

Access (inherited): api · free

Public domain. BLS Public Data API requires free registration above 25 queries/day.

None commercial. The key analytical point is that this index is decoupled from consumer inflation: construction inputs ran +8.9% y/y in August 2026 against core CPI of 2.4%. Trade bodies re-publish this series with advocacy framing; go to the BLS series directly.

Series to watch for tariff pass-through: copper and brass mill shapes, softwood lumber, steel mill products, and switchgear/switchboard apparatus - the last of which is also the data-centre/industrial contention point.

Coverage: United States · back to 1947 for major groups

Access (inherited): api · free

Public domain.

Freddie Mac is a GSE in federal conservatorship with an institutional interest in mortgage volume. The survey samples lender-offered rates, not locked rates, so it understates what a marginal or low-credit borrower actually pays. Adequate as a directional benchmark, not as a borrower-level affordability measure.

Released Thursdays. The single most important price in this sector: 6.76% on 2026-09-10, up 41bp year over year.

Coverage: United States · back to 1971 for the 30-year fixed

Access (inherited): public_page · free

Free use with attribution to Freddie Mac.

NAR is a trade association for real estate agents with a direct commercial interest in transaction volume and in an optimistic reading of the market. The underlying MLS-derived counts are the industry standard and are reliable; the accompanying chief-economist commentary is advocacy and should be quoted as opinion, never as analysis. NAR also has a documented history of litigation over commission structures that colours its framing of transaction costs.

Release ~20th of month. Months-of-supply is the number that matters more than the sales pace: 4.9 months in August 2026 versus 4.0 in August 2019 means the market is looser than pre-pandemic despite depressed sales.

Coverage: United States, four regions; metro detail quarterly · back to 1968

Access (inherited): public_page · freemium

Headline figures freely citable with attribution. Detailed metro and historical datasets are member-restricted or paid.

NAHB is a homebuilder trade and lobbying association. The HMI is a sentiment survey of its own members, so it measures builder mood, not transactions, and NAHB uses it to argue for regulatory relief and against tariffs and impact fees. The price-cut and incentive shares are the genuinely useful series because they are behavioural rather than attitudinal.

Released ~16th, one day before housing starts. Eye on Housing (the blog) carries free derived analysis - build time, custom-home premium, regulatory cost - that is not otherwise published.

Coverage: United States, four regions · back to 1985

Access (inherited): public_page · freemium

Headline index free to cite with attribution to NAHB/Wells Fargo. Detailed series sit behind the paid Housing Economics Plus subscription.

CBRE is a brokerage that earns fees on the leasing and investment sales it measures, so its framing consistently emphasises tight supply and rising rents. It is nonetheless the primary collector of the underlying inventory and vacancy data - there is no government substitute. The critical caveat is definitional: CBRE counts only primary-market capacity in active construction, which is why it reported 7,481 MW under construction in H1 2026 where JLL reported over 66 GW. Neither is wrong; they count different things, and the gap must be stated whenever either is cited.

The most useful single source for data-centre vacancy, preleasing share, asking rates per kW-month and - newly in 2026 - a market-by-market account of moratoria and local opposition.

Coverage: North America primary and secondary markets; companion global report · semiannual back to roughly 2015

Access (inherited): public_page · freemium

Public summaries and press releases are free to cite with attribution. Full datasets and market-level detail require a client relationship. No redistribution of report PDFs.

Same structural conflict as CBRE - JLL is a transaction intermediary reporting on its own market. Its data-centre capacity figures use a much broader definition than CBRE's, including frontier markets and earlier-stage capacity, producing headline numbers roughly an order of magnitude larger. Useful as the upper bound of a range, dangerous as a point estimate.

Run against CBRE as a deliberate pair. The divergence between the two is itself the monitorable signal - it measures how much of the pipeline is contracted versus announced.

Coverage: North America; separate EMEA and APAC editions · semiannual back to roughly 2017

Access (inherited): public_page · freemium

Public summaries citable with attribution; full reports gated behind registration.

Trepp sells data and analytics to CRE lenders and investors, so it has an interest in distress being interesting. The data itself is loan-level and auditable. The important methodological point is that Trepp's headline delinquency rate excludes matured loans that remain current on interest - in July 2026 including them would have raised the rate from 7.86% to 9.62%. That 176bp wedge is the most direct available measure of extend-and-modify behaviour.

Covers only securitised debt, roughly 15-20% of US CRE mortgage debt. Bank-held CRE loans - where most office debt actually sits - are visible only through call reports and are far less granular. Do not treat CMBS delinquency as the whole distress picture.

Coverage: United States private-label CMBS · back to the 1990s at loan level

Access (inherited): public_page · enterprise

Headline monthly rates are published free and widely syndicated. Loan-level data and the full dataset are enterprise-licensed; redistribution prohibited. Phase 2 licensing budget should assume this is a paid feed if loan-level detail is needed.

Green Street sells research to REIT investors. The index is derived from estimates of REIT portfolio values, not from closed transactions or appraisals, so it is timely but reflects public-market sentiment and is levered to REIT share prices. It systematically leads appraisal-based indices and can overshoot in both directions. Valuable precisely because its method differs from NCREIF's - the two together bracket the uncertainty.

Released within days of month-end - the fastest CRE pricing read available. Must be monitored alongside NCREIF, never instead of it.

Coverage: United States, Canada, Europe; 25+ countries, 12 core sectors · 27 years

Access (inherited): public_page · enterprise

Monthly headline percentage changes are published free. Index levels, sector series and history are subscription-only; redistribution prohibited.

NCREIF members are institutional owners who report their own appraised values, which creates a structural incentive toward smoothing and lag. The index measures what owners' appraisers say assets are worth, not what buyers paid - the capital-value component fell roughly 20% from Q2 2022 to Q4 2024 without a recession, and largely after transaction evidence had already moved. It is the sector's canonical example of appraisal lag and should always be labelled as appraised, never as market value.

The appraised leg of the appraised-versus-transacted problem. Pair with Green Street (REIT-implied) and MSCI/RCA (transacted) to measure the spread between the three, which is more informative than any one of them.

Coverage: United States institutional-grade property · back to 1978

Access (inherited): licensed · paid_high

Headline quarterly returns published free. Full index data requires membership or a data licence; redistribution restricted. A licensing line item for Phase 2.

AGC is a contractor trade and lobbying association. Its releases are analyses of BLS and Census data (Tier A underneath) wrapped in advocacy for immigration expansion, tariff relief and infrastructure funding. The data reprocessing - state and metro construction employment, construction unemployment rate, input-cost breakdowns - is genuinely useful and not published elsewhere in that form. The interpretive claims are lobbying positions.

In 2026 AGC is the most explicit public source connecting data-centre construction to crew scarcity - its 2026-09-03 release states workforce shortages remain acute despite soft market conditions because data centres are straining labour supply.

Coverage: United States, all states and ~360 metro areas · rolling archive, roughly 15 years online

Access (inherited): public_page · free

Press releases free to quote with attribution. Member survey microdata not released.

ABC is an open-shop contractor association, politically opposed to AGC on labour questions; its member base skews non-union and toward the South. The Backlog Indicator is a member survey, so it reflects a self-selected sample and should be read as a direction indicator rather than a level. Monthly input-price commentary is BLS PPI re-cut with advocacy framing.

Backlog was 8.8 months in June 2026, essentially flat year over year - a useful corrective to the impression that the data-centre boom has lifted contractor pipelines generally. It has not; it has lifted a specific set of contractors.

Coverage: United States, four regions and by revenue band · back to 2009

Access (inherited): public_page · free

Free to cite with attribution.

No commercial interest. Note the 2024 restructuring: the standalone UK House Price Index bulletin was discontinued and merged into this publication, which breaks naive time-series scrapes built against the old URL. Northern Ireland is reported quarterly, not monthly, so its headline figures are not directly comparable with Great Britain's.

The best non-US comparator in this sector and the cleanest test of whether the US housing freeze is a rate phenomenon or a US-specific one. UK prices rose 2.0% y/y in June 2026 with rents up 3.7% - a materially different pattern from the US.

Coverage: United Kingdom - England, Wales, Scotland, Northern Ireland, regions and local authorities · house prices back to 1968, private rents back to 2005

Access (inherited): api · free

Open Government Licence v3.0 - free reuse including commercial, with attribution.

University research centre, but funded in part by industry sponsors including homebuilders, lenders and REITs, and its policy framing consistently favours increased housing production and subsidy. Methodology is transparent and peer-reviewable. Its immigration-and-household-formation work is the most rigorous public modelling of the demand-side shock and should be cited as an estimate with the modeller named, not as fact.

The blog is higher-frequency and more useful for trend detection than the annual report. Daniel McCue's 2026-08-11 post quantifying the immigration-driven collapse in household formation is the key demand-side input for this sector in 2026.

Coverage: United States, with metro-level interactive data · annual report since 1988

Access (inherited): public_page · free

Free to download and cite with attribution. Underlying tabulations of Census microdata are released alongside.

Qualitative and anecdotal by design - it reports what District contacts say, not what they do, and contact panels are not statistically representative. It is nonetheless the only systematic, geographically disaggregated, near-real-time read on construction labour availability, materials cost pass-through and CRE lending conditions, and it surfaces turns weeks before the quantitative series.

Published eight times a year, two weeks before each FOMC meeting. Search the District summaries for 'data center' - District-level mentions of data-centre construction crowding out other work are an early indicator of the labour-contention trend.

Coverage: Twelve Federal Reserve Districts covering the United States · back to 1970

Access (inherited): public_page · free

Public domain.

A small, sponsor-supported industry body, not an audited data provider. Its category boundary for 'proptech' is self-defined and shifts - construction robotics, mortgage fintech and data-centre developers may or may not be counted in a given period, which makes year-over-year comparisons unstable. Treat totals as estimates with the modeller named. Its numbers diverge from Crunchbase and PitchBook for the same reasons those two diverge from each other.

Recorded $4.53bn across 231 rounds in H1 2026 against $16.7bn tracked for full-year 2025. Cross-check every figure against Crunchbase and PitchBook before relying on it; no single proptech tracker is authoritative.

Coverage: Global, US-weighted · back to roughly 2019

Access (inherited): public_page · freemium

Headline figures free to cite with attribution; full research reports gated.

Advertising- and sponsor-supported trade press with named bylines and generally sound sourcing. Sponsored content is labelled but sits adjacent to editorial. Coverage skews toward large general contractors and toward technology vendors that market to them, so contech funding rounds get more coverage than their economic weight warrants.

The fastest public source for construction-technology funding rounds and M&A, and for contractor-level accounts of data-centre labour competition. Use for discovery, then verify every figure against a primary source.

Coverage: United States, some international · rolling archive back to 2015

Access (inherited): rss · free

Free with registration. Headlines and short quotes citable; full-text redistribution prohibited.

Vendor-sponsored trade publication serving the data-centre industry. Named bylines and technically competent, but it reports announced projects, announced capacity and announced leases with limited scepticism about whether they are contracted or financed. Announcement volume in this outlet is an attention measure, not an adoption measure, and should be scored as velocity only.

Best early-warning source for site-selection announcements, local moratoria and campus-scale projects. Treat headline megawatt and dollar figures in announcements as marketing claims until a filing, financing or utility docket confirms them.

Coverage: Global, North America-weighted · back to 2015

Access (inherited): rss · free

Free to read; attribution required for quotes.

A commercial project-lead database selling subscriptions to contractors and suppliers; it has a direct interest in the pipeline looking large. Its starts measure is based on projects entering its database, which captures announced and bid projects rather than verified groundbreakings, so it leads Census put-in-place spending and overstates in categories with high cancellation rates. Its data-centre series is nonetheless the only public project-level count with cost-per-square-foot detail.

Source of the $84.1bn year-to-date data-centre starts figure and the $818.20 per square foot cost measure, both of which are widely quoted. Cross-check its starts against Census put-in-place before treating either as the level.

Coverage: United States and Canada, state and metro level · database back to the 1990s

Access (inherited): public_page · enterprise

Monthly economic commentary published free. Project-level data is an enterprise subscription; redistribution prohibited. A Phase 2 licensing candidate if project-level tracking is required.

US federal statistical agency; methodological choices (owners' equivalent rent, hedonics) are contested but disclosed. No commercial interest.

The single most important consumer-side series for this sector. Fetches cleanly. Always pair with PPI: the two diverged by two points in August 2026.

Coverage: United States, urban consumers · back to 1913 for all items; food-at-home detail series back to 1952

Access (inherited): api · free

US Government work, public domain; BLS requests citation. Public API (api.bls.gov/publicAPI/v2) is rate-limited without a registration key.

Same agency as CPI; final demand includes trade-services margins, which is routinely mis-compared with CPI.

The leading indicator for shelf-price inflation two to four quarters ahead. Final demand ran 5.4% against CPI 3.4% in August 2026.

Coverage: United States · back to 1913 (selected series)

Access (inherited): api · free

US Government work, public domain.

US government forecasts with published prediction intervals; a policy agency forecasting prices its own department influences.

Only source giving category-level food price forecasts with intervals. The 2027 food-at-home interval of -5.7% to +10.5% is itself a finding about forecast uncertainty.

Coverage: United States · annual forecasts since 1974; current file covers 2024-2027

Access (inherited): bulk_download · free

US Government work, public domain. Excel and CSV downloads are the machine-readable route; the HTML summary page is the human one.

Producer-facing agency; forecasts are used to justify farm support appropriations, so there is an institutional interest in the level.

Released roughly February, September and December. The September 2026 release carried the single most important sector fact: government payments +69.8% to $47.4bn. Programme composition is NOT broken out on the landing pages and requires the underlying data files.

Coverage: United States · income statement series back to 1910

Access (inherited): bulk_download · free

US Government work, public domain.

Market-moving official forecast; historically subject to intense scrutiny over methodology, not over independence.

Price-forming for global grain and oilseed markets. Released around the 11th of each month at 12:00 ET.

Coverage: Global, with US detail · monthly since 1973

Access (inherited): bulk_download · free

US Government work, public domain; strict pre-release embargo. The dated PDF path (wasde0926.pdf pattern) fetches reliably.

Survey-based official statistics; USDA announced a 2026 plan to add satellite, administrative and AI/ML methods, which may change series continuity.

Cattle Inventory, Acreage, Grain Stocks and Crop Production are the price-forming reports. Watch this feed specifically for further discontinuations: selected programmes were dropped in 2025.

Coverage: United States, to county level · Census of Agriculture back to 1840; major surveys back decades

Access (inherited): public_page · free

US Government work, public domain. QuickStats API available at quickstats.nass.usda.gov/api.

UN agency; index is export-weighted international quotations, so it is not comparable with national CPI food indices and is frequently misreported as if it were.

Released on the first Friday of each month. Sub-indices (cereal, vegetable oil, dairy, meat, sugar) are where the signal is: sugar moved 11.9% in a single month in August 2026.

Coverage: Global · monthly back to 1990

Access (inherited): bulk_download · free

FAO open data under CC BY-NC-SA for most datasets; attribution required, commercial redistribution restricted. Check the licence before commercial republication.

Multilateral institution; no commercial interest. Index base year 2010 differs from FAO's, which causes level confusion.

The best single file for fertiliser, energy and food commodity levels in one place. The monthly PDF URL hash changes, so resolve it from the commodity-markets landing page each month rather than hard-coding it.

Coverage: Global · monthly historical data back to 1960 in the accompanying XLSX

Access (inherited): bulk_download · free

World Bank open data under CC BY 4.0 with attribution.

US federal statistical agency. Advance estimates are revised; sales are nominal and must be deflated to read volume.

The cleanest public read on food-services versus grocery channel growth. The direct PDF path fetches reliably where the navigation pages do not.

Coverage: United States · monthly back to 1992

Access (inherited): bulk_download · free

US Government work, public domain; Census Data API available.

Official but political: framing is explicitly administration-aligned ('most pro-rancher administration in history'). Facts are checkable; framing is not neutral.

Where programme announcements appear first: Ranchers First and the BRAND insurance endorsement (2026-08-31), agricultural data modernization (2026-09-01). Individual dated release URLs fetch reliably.

Coverage: United States · current administration archive

Access (inherited): public_page · free

US Government work, public domain.

Regulator; publishes what it has finalised, which means proposed rules that stall leave no 'withdrawn' marker. Absence of a final rule must be read carefully.

MIXED FETCHABILITY, important for Phase 2: the FSMA 204 and front-of-package topic pages fetch cleanly, but /food/hfp-constituent-updates, /food/cfsan-constituent-updates, /food/food-ingredients-packaging/fda-food-chemical-safety and the ultra-processed-foods path all returned 404. Federal Register is the reliable alternative for this agency.

Coverage: United States · rulemaking history back to the 1990s online

Access (inherited): public_page · free

US Government work, public domain.

Primary legal record; no editorial position. Volume is high and requires filtering.

PHASE 2 PRIORITY. This is the reliable substitute for FDA and FSIS pages that block or 404. It is the only way to track proposed-versus-final status and comment-period changes systematically, which is the core of trend T-19-09.

Coverage: United States · back to 1994 in full text

Access (inherited): api · free

US Government work, public domain; documented JSON API with no key required.

Regulator; recall notices are factual and timely.

BLOCKED AT TIME OF WRITING: returned HTTP 403 to an automated fetch. Phase 2 needs either an allowed user agent, the FSIS recall API at fsis.usda.gov/fsis/api/recall/v/1, or Federal Register as the fallback. Record as an access-negotiation item.

Coverage: United States · recall archive back to the 1990s

Access (inherited): public_page · free

US Government work, public domain.

EU institutional perspective; trade releases emphasise EU competitiveness. Data are robust.

The single best non-US source for this sector and it fetches reliably, including individual dated article URLs. Carried three material 2026 items: the EUR 540m fertiliser package, H1 trade data, and the AgRI 2040 strategy.

Coverage: European Union, 27 Member States · news archive several years; agri-food trade series longer

Access (inherited): public_page · free

European Commission reuse policy, Decision 2011/833/EU: reuse permitted with acknowledgement.

Explicitly policy-directional in the 2025-2030 edition, which describes itself as a 'reset'. Treat the framing as policy, the recommendations as authoritative.

Updates once every five years but each edition resets the specification for federal feeding programmes and reformulation. The full text fetches at cdn.realfood.gov/DGA.pdf.

Coverage: United States · editions every five years since 1980

Access (inherited): public_page · free

US Government work, public domain.

Primary filings; figures are as-filed and audited. No editorial layer.

CONFIRMED WORKING and the most reliable route to quantitative company data for this sector. NOTE THE LIMITATION FOUND: data.sec.gov/submissions/CIK##########.json is too large for the fetch tool and comes back with the form[] array truncated, so filing discovery by form type fails. Use companyconcept for numbers; use a different route for narrative disclosure.

Coverage: US-listed issuers worldwide · back to roughly 2009 for XBRL-tagged facts

Access (inherited): api · free

US Government work, public domain. SEC requires a descriptive User-Agent header with contact details; rate limit 10 requests/second.

Issuer-authored; volume and organic-growth definitions are non-GAAP and issuer-specific. GLP-1 effects have an obvious disclosure disincentive.

MAJOR PHASE 2 GAP. IR index pages are JavaScript-rendered and were unusable; only direct release URLs worked, and only for Nestle, Coca-Cola, Starbucks and Tyson (via a q4cdn PDF redirect). PepsiCo, General Mills, McDonald's, Chipotle, Nutrien, Mosaic, Deere and Kraft Heinz could not be retrieved. Phase 2 should buy or build a release-URL index rather than rely on discovery.

Coverage: Global · typically five to ten years of releases per issuer

Access (inherited): public_page · free

Issuer copyright; quotation with attribution is standard practice. No scraping behind logins.

Commodity-group advocacy: positions on biofuel policy, tariffs and trade aid reflect member interest. Factual claims (RIN volumes, credit mechanics) are checkable.

Useful early indicator on RFS, 45Z and small refinery exemptions, which set the marginal price of vegetable oil. Listing page fetches; individual article bodies were not served in full.

Coverage: United States · several years

Access (inherited): public_page · free

Association copyright; attribution expected.

Regulator; volume decisions are litigated and politically contested in both directions.

STALE AT TIME OF WRITING: the annual-standards table was populated only through 2025 while the programme homepage referred to a finalised 'RFS Set 2' for 2026-27. Use the Federal Register entry for the actual volumes; do not rely on the summary table.

Coverage: United States · annual standards back to 2006

Access (inherited): public_page · free

US Government work, public domain.

Brazilian government agency; national perspective and Portuguese-language primary source. Periodically diverges from USDA FAS estimates for Brazilian crops, which is itself a signal.

Essential non-US primary source: Brazil is the largest soybean supplier. The landing page redirects from conab.gov.br to gov.br/conab and serves only headline figures; the Boletim PDF is the real dataset and Phase 2 should resolve that path directly.

Coverage: Brazil · monthly surveys back many seasons

Access (inherited): public_page · free

Brazilian government open data; attribution expected.

Advocacy organisation for alternative proteins with an explicit interest in the category appearing to grow. Its methodology is published, which is more than most category trackers offer, but it is not a neutral party.

NOT RETRIEVABLE THIS CYCLE: the state-of-the-industry and investment-data paths returned 404. This is the main reason no alternative-protein funding figure appears in this dossier. Treat any circulating alt-protein funding total as unverified until this source is re-established, and never cite it as Tier A.

Coverage: Global · annual reports since 2019

Access (inherited): public_page · free

GFI publishes under open licences for most reports; check per-report terms.

Industry trade body; forecasts and commentary favour the industry's outlook. Underlying data are often licensed from third parties (Technomic, Black Box).

PAYWALL WARNING FOR PHASE 2 BUDGETING: restaurant traffic and same-store-sales trackers (Black Box Intelligence, Technomic, Circana) are the sector's most useful demand data and are all licensed. The free page showed nothing newer than September 2025. Census food-services sales is the free substitute but is nominal and aggregate only.

Coverage: United States · several years of commentary

Access (inherited): public_page · freemium

Association copyright; much of the underlying traffic and same-store-sales data is paid and may not be redistributed.

None material. Primary legal text of record. The risk is analyst error in reading consolidated versus original versions, not publisher bias.

The single highest-value source in this sector. Directive (EU) 2026/470 and Directive (EU) 2025/794 both retrieved successfully via the legal-content/EN/TXT/HTML pattern. NOTE: eur-lex search.html is robots-disallowed - use direct legal-content URLs or the EP Legislative Observatory to find act numbers first.

Coverage: EU · back to 1952

Access (inherited): public_page · free

Reuse permitted under Commission Decision 2011/833/EU with attribution; the authentic version is the PDF OJ.

None material. Procedural record only; does not editorialise on content.

Best way to resolve an EU file from proposal to final act number and OJ date. Procedure 2025/0045(COD) gave the full Omnibus I timeline and the Directive 2026/470 identifier in one fetch. NOTE: oeil.secure.europarl.europa.eu 302-redirects to oeil.europarl.europa.eu - follow the redirect.

Coverage: EU · back to 1994

Access (inherited): public_page · free

EU institutional reuse policy; attribution expected.

EFRAG is the Commission's technical adviser and has an institutional interest in the ESRS continuing to exist; its framing of simplification is correspondingly positive.

The root domain fetches reliably; deep news-item URLs frequently 404. Fetch https://www.efrag.org or /en/news-and-calendar/news and read the listing. Source of the 3 July 2026 revised ESRS adoption, the 28 August 2026 datapoints draft list, the ESRS-40a consultation, and the State of Play 2026 sample of 900 assured statements.

Coverage: EU · back to 2021 for sustainability standards

Access (inherited): public_page · free

Standards texts are freely downloadable; check the copyright notice on each standard before redistribution.

Commission communications present simplification as burden reduction and understate the scope of what was removed. Cross-check against the directive text.

Useful for timeline and intent; the operative figures must come from EUR-Lex. Confirmed the 9 December 2025 political agreement and the 1,000-employee framing.

Coverage: EU · back to 2021

Access (inherited): public_page · free

Commission Decision 2011/833/EU reuse with attribution.

Official administering body; presents CBAM as a climate measure rather than a trade measure. Both readings are correct and the framing matters for how exporters respond.

Definitive regime confirmed as applicable from 1 January 2026 across cement, iron and steel, aluminium, fertilisers, electricity and hydrogen. The de minimis threshold and certificate sale schedule were NOT on the landing page and need the legislation and guidance sub-pages in Phase 2.

Coverage: EU and all exporters into the EU · back to 2023 transitional phase

Access (inherited): public_page · free

Commission reuse policy with attribution.

A specialist tracker whose commercial interest is in the durable CDR market being taken seriously; it nonetheless publishes the delivery shortfall prominently, which cuts against that interest. Coverage is self-reported and announcement-based, so purchases are better captured than quiet cancellations.

The only accessible source distinguishing tonnes SOLD from tonnes DELIVERED, which is the central analytical distinction in carbon removal. Source of 49,468,711 t sold, 1.68 Mt delivered, 3.4% delivery rate, USD 12.5bn, 858 suppliers, 1,240 purchasers and the Microsoft 75.1% concentration. Needs a Tier A cross-check in Phase 2 - none was available within this research budget.

Coverage: Global · back to 2020

Access (inherited): public_page · freemium

Public dashboard; commercial redistribution of the underlying dataset likely requires a licence. Check terms before bulk use.

The ICVCM exists to certify integrity and has an institutional interest in the CCP label mattering. Its rejections are the more informative half of the record.

Structured, dated table of approved and rejected methodologies - 44 approved as of 12 August 2026. The absence of any approved renewable-energy, landfill-gas or REDD+-specific methodology is the single most useful fact on the page. NOTE: icvcm.org/news/ returned a newsletter confirmation page rather than news; use /assessment-status/ directly.

Coverage: Global · back to 2023

Access (inherited): public_page · free

Public assessment records; attribution expected.

A registry publishing its own firsts. Every 'first ever' claim is self-declared and should be treated as a signal of direction rather than a verified market milestone.

Best available early-warning feed for CORSIA and Article 6 operationalisation. Source of the first large-scale airline CORSIA retirement (24 March 2026), first CORSIA political risk underwriter (7 July 2026) and the joint Verra Article 6.2 tool (20 August 2026). Cross-check transaction claims against ICAO.

Coverage: Global · back to 2003

Access (inherited): public_page · free

Press content; attribution expected for reuse.

Academic and independent of any registry. No commercial interest. Selection is limited to six registries, so newer or smaller registries are out of scope.

The highest-quality free carbon market dataset in existence and the correct answer to 'what is the VCM actually doing'. Release v2026-06 is current. CRITICAL FOR PHASE 2: the landing page carries no statistics - all figures live in the downloadable Excel files, so the pipeline must ingest the workbook, not scrape the page. This is why this dossier carries no VCM issuance/retirement totals.

Coverage: Global · back to registry inception, all issuances and retirements

Access (inherited): bulk_download · free

Free academic use with citation; check the accompanying licence file before redistribution.

Multilateral, methodologically transparent. Coverage of crediting mechanisms is less complete than coverage of taxes and ETSs.

Source of 47 carbon taxes, 40 ETSs and 34 crediting mechanisms in 2026. Revenue and emissions-coverage percentages sit inside interactive charts and were not extractable by fetch; the companion State and Trends of Carbon Pricing 2026 report returned HTTP 403 through the research proxy and needs alternative access in Phase 2.

Coverage: Global · back to 1990

Access (inherited): public_page · free

CC BY 4.0 for most World Bank data; verify per dataset.

The agency is simultaneously the data publisher and the party proposing to narrow the dataset. Read programme pages, not the climate-change index pages, which are stale.

Facility-level emissions for roughly 8,000 US facilities and the calibration layer under most US climate analytics. Phase 2 must monitor the docket for the 12 September 2025 proposal to remove 46 source categories - if finalised, the historical series ends and every product built on it degrades. Page last updated 30 July 2026 and still shows the rule as proposed.

Coverage: US · back to 2010 reporting year

Access (inherited): bulk_download · free

US government work, public domain.

Same dual role as above. Different EPA pages update on different schedules and contradict each other on status - always check the page's own last-updated stamp.

Tracks the 11 June 2025 proposal to repeal all power-sector GHG standards under CAA Section 111. As of the 27 February 2026 update the final action had not gone to OMB. NOTE: epa.gov/climate-change/climate-change-regulatory-actions-and-initiatives was last updated 25 September 2025 and lists NO 2025-26 actions - a live date-laundering trap.

Coverage: US · back to 2015

Access (inherited): public_page · free

US government work, public domain.

Official record. Framing reflects the sitting Commission's policy posture.

Press release 2025-58 (27 March 2025) is the authoritative record of the SEC abandoning its climate disclosure rule. ACCESS NOTE: the research proxy logged 403 CONNECT rejections for www.sec.gov and efts.sec.gov, but WebFetch on individual press-release and Archives PDF URLs succeeded - use direct document URLs, and data.sec.gov/submissions for filing lists, per the macro brief retrieval playbook.

Coverage: US · back to 1990s

Access (inherited): public_page · free

US government work, public domain.

Statistically rigorous and the closest thing to a neutral US emissions forecast. Forecasts are model output and revise materially.

The measured-outcome counterweight to every corporate target claim in this sector. September 2026 STEO gives US energy-related CO2 at 4,821 MMmt in 2026 and 4,816 MMmt in 2027 - flat. EIA has a proper open API and should be wired first among US government sources.

Coverage: US · back to 1997

Access (inherited): api · free

US government work, public domain; EIA API requires a free key.

Intergovernmental with member-state interests; historically conservative on renewables growth. Methodology is published, which is what distinguishes it from vendor market reports.

The only source that cleanly separates climate-specific capital (CCUS at just over USD 5bn) from energy capital (USD 3.4tn total). RETRIEVAL NOTE: the HTML report pages return navigation only - fetch the blob PDF at iea.blob.core.windows.net, which parses correctly and contains all headline figures.

Coverage: Global · back to 2016

Access (inherited): bulk_download · freemium

Free PDF download; redistribution restricted, attribution required. Some datasets are paid.

A commercial market-intelligence firm selling into the climate tech ecosystem, so it has an interest in the sector appearing healthy. Its own taxonomy decisions - counting data-centre deals as Built Environment climate tech - materially inflate the headline total, and it discloses this.

The most-quoted climate tech funding series and therefore the one most worth monitoring critically. Its H1 2026 release is the key evidence that climate venture funding growth is a classification artefact. MUST be triangulated against PitchBook, Net Zero Insights or Cleantech Group in Phase 2 - that cross-check could not be completed within this research budget.

Coverage: Global · back to 2020

Access (inherited): rss · freemium

Newsletter content; commercial reuse of the underlying database requires a licence.

SBTi's relevance depends on companies continuing to seek validation, which creates pressure toward accommodating standards. Version 2.0's 'options to reflect business reality' framing should be read in that light.

Updated every Thursday by 12:00 GMT and downloadable, which makes it one of the best structured corporate-commitment datasets available. Current totals: 14,174 with targets or commitments, 11,948 validated, 2,790 net-zero, 2,387 active commitments. The 'commitment removed' status is a filterable category but no aggregate count is published - Phase 2 should derive it from the export, because it is the single best measure of target abandonment.

Coverage: Global · back to 2015

Access (inherited): bulk_download · free

Public dashboard with an Excel export; attribution expected.

Self-reported corporate data with well-documented selection bias toward larger and more engaged companies. Scores measure disclosure quality, not emissions performance.

22,100+ companies disclosed in 2025; 899 A List from roughly 20,000 evaluated; 27 Triple A. The year-on-year comparison was NOT retrievable - cdp.net/en/insights returned an empty landing page and the historical press release 404'd. Phase 2 needs the prior-year figure to establish whether disclosure participation is growing or falling, which is a live question.

Coverage: Global · back to 2003

Access (inherited): public_page · freemium

Full response data requires a licence; scores and A List are public.

UN body advocating for greater climate action; the framing is directional but the underlying finance and emissions accounting is peer-reviewed and methodologically documented.

The authoritative annual quantification of the adaptation finance gap - USD 310-365bn needed by 2035 against USD 26bn delivered in 2023, down from USD 28bn. Both reports publish in late October and early November, so this is a seasonal rather than continuous monitoring target.

Coverage: Global, with a developing-country focus · Adaptation Gap back to 2014; Emissions Gap back to 2010

Access (inherited): bulk_download · free

Free download; UN attribution and non-commercial reuse conditions apply.

Negotiated text reflects the lowest common denominator of 190+ parties; the gap between decision language and implementation is the standing analytical problem.

COP30 concluded in Belem on 22 November 2025 with a Belem Political Package. RETRIEVAL NOTE: unfccc.int/cop30 is a directory page and does not carry decision content - Phase 2 must fetch the individual COP 30 / CMP 20 / CMA 7 decision documents. Article 6 transaction volumes were not retrievable and remain the largest unfilled gap in this sector's carbon market coverage.

Coverage: Global · back to 1995

Access (inherited): public_page · free

UN reuse terms; attribution required.

The largest voluntary registry, with a direct commercial interest in the credibility of its own credits and in issuance volume. Registry data is factual; programme-page framing is not neutral.

Programme page gives 2,579+ VCS projects and over one billion tonnes cumulative, but is undated. CRITICAL RETRIEVAL NOTE: registry.verra.org is a JavaScript application and returns no data to a fetch - Phase 2 must use the Berkeley bulk database (S-20-09) or a registry API for VCU issuance and retirement figures.

Coverage: Global · back to 2006

Access (inherited): public_page · free

Registry data is public; bulk extraction may be restricted by terms.

Commission framing emphasises achievability; the operative targets sit in the regulation's annexes rather than the summary page.

Confirmed Regulation (EU) 2025/40 in force 11 February 2025 and applicable 12 August 2026, with Directive 94/62/EC repealed the same day, plus the December 2025 Environmental Omnibus proposal. The numerical recycled-content and reuse targets were NOT on the page and must come from the regulation's annexes on EUR-Lex - a known gap in this dossier.

Coverage: EU · back to 1994 for packaging

Access (inherited): public_page · free

Commission Decision 2011/833/EU reuse with attribution.

Historically the authoritative free US disaster-cost series. Its apparent discontinuation is itself the signal; do not assume continuity.

The citation page shows a 2025 dataset with a publication date of 12 June 2025 and no release since. Phase 2 must confirm whether the product was formally retired and identify a replacement - candidates are Gallagher Re, Aon and Swiss Re sigma, all of which are commercial and use insured rather than total economic loss. This is a material degradation of free US climate loss data.

Coverage: US · back to 1980

Access (inherited): public_page · free

US government work, public domain.

Commercial data vendor whose products become more valuable as free public alternatives disappear. Its free pages are stale marketing surfaces - the Net-Zero Tracker page still shows data as of 31 August 2024.

Flagged for the Phase 2 LICENSING BUDGET. MSCI now gates First Street's research library and sells the weekly carbon credit price index and carbon market forecasts. Together with Sylvera, BeZero and Abatable this is where VCM price and volume data actually lives, and none of it is free. If Phase 2 needs carbon prices, this is the line item.

Coverage: Global · varies by product

Access (inherited): licensed · enterprise

Licensed; redistribution prohibited. Free pages are promotional only.

State regulator; the most significant surviving mandatory US corporate climate disclosure requirement after the SEC withdrawal.

BLOCKED: ww2.arb.ca.gov is robots-disallowed and could not be fetched. SB 253 and SB 261 deadlines, thresholds, covered-company counts and litigation status are therefore entirely absent from this dossier. Phase 2 must obtain access by another route - the California rulemaking docket, the Office of Administrative Law, or a licensed legal-tracking service. This is the highest-priority access gap in the sector.

Coverage: US (California, with extraterritorial reach to companies doing business there) · back to 2023

Access (inherited): public_page · free

US state government work; generally public domain.

None material. Registrant-prepared but audited and legally attested; management discussion is self-serving by construction and should be read separately from the financial statements.

THE primary source for this sector. Confirmed working pattern: data.sec.gov/submissions/CIK##########.json to enumerate filings, then fetch the document under /Archives/edgar/data/<cik>/<accession-no-dashes>/<primary>.htm. browse-edgar is robots-disallowed and efts.sec.gov full-text search 403s through proxies. Key CIKs: Coinbase 1679788, Circle 1876042, Strategy 1050446, Core Scientific 1839341, TeraWulf 1083301.

Coverage: US registrants including foreign private issuers · back to 1993/1994

Access (inherited): api · free

Public domain US government work. Fair-access policy requires a descriptive User-Agent and rate limiting to 10 requests/second.

Agency communications: framing reflects the sitting Commission's policy posture. Crypto policy direction changed materially between administrations, so compare releases across years rather than reading any single year as settled law.

Filterable by year and month via query parameters (?year=2026&month=8), which works reliably where the unfiltered index paginates badly. Release URLs are slugged (e.g. /newsroom/press-releases/2026-76-sec-proposes-new-regulation-crypto-assets); the bare numeric form 404s.

Coverage: US · back to 1990s

Access (inherited): public_page · free

Public domain US government work.

Prosecutorial framing: these are the SEC's allegations plus settlements, not adjudicated findings on the merits in most cases. Settlements are typically without admission or denial.

The only systematic public record of crypto market manipulation, wash trading and fraud in this sector. Essential counterweight to promotional metrics. Individual release URLs take the form /enforcement-litigation/litigation-releases/lr-NNNNN; the older /litigation/litreleases/lr-NNNNN path 404s.

Coverage: US · back to 1995

Access (inherited): public_page · free

Public domain US government work.

Agency communications. Staff letters and advisories are not Commission rules: they bind no one beyond the named recipient and can be withdrawn without notice-and-comment. Treat them as durable only with that caveat explicit.

Individual releases fetch reliably at /PressRoom/PressReleases/NNNN-NN. The combined listing page does not respond to a 'combine' keyword filter the way SEC's does, so poll the index and match on title. Highest-value items in 2026: 9192-26 (SEC MoU), 9197-26, 9200-26, 9239-26, 9241-26.

Coverage: US · back to 1990s

Access (inherited): public_page · free

Public domain US government work.

Central bank communications; institutionally cautious and understated. Dissenting governor statements are published separately and are often where the substance is.

Year-indexed pages fetch reliably. Individual releases follow /newsevents/pressreleases/<type>YYYYMMDDa.htm (bcreg, other, monetary). Captured the 2026 stablecoin CIP proposal, the payment account proposal and the tokenised securities capital clarification.

Coverage: US · back to 1996

Access (inherited): public_page · free

Public domain US government work.

Explicitly labelled independent staff research reflecting only the authors' views and not policy deliberation. That disclaimer is real and should be quoted when citing; these are not Board positions.

The highest-quality free analytical source on stablecoins anywhere. Six 2026 notes on stablecoins and digital money, including the only public comparison of USDT and USDC reserve quality. Note the URL trap: the year index links use .html while most other Fed pages use .htm, and the .htm form 404s.

Coverage: US with international comparison · back to 2013

Access (inherited): public_page · free

Public domain US government work.

Institutional perspective of central banks. Systematically sceptical of private digital money and supportive of a two-tier system with central bank money at the base; that is a defensible analytical position but it is a position. Read alongside industry sources, not instead of them.

The 2026 AER chapter 'Anchoring trust in money: innovation beyond stablecoins' is the single best analytical framing available. Chapter URLs take the form /publ/arpdf/ar2026eN.htm. The working-paper and other-publication list pages (work.htm, othp_list.htm) 404'd during this research; go straight to known document URLs.

Coverage: Global · back to 1930s

Access (inherited): public_page · free

Free to use with attribution; redistribution of full texts restricted.

Compiled from national competent authority submissions, so completeness depends on 27 different regulators' reporting discipline. Lag between national authorisation and register appearance is not documented.

STRUCTURALLY THE BEST DATASET IN THIS SECTOR: five weekly CSVs giving a complete, machine-readable picture of who is legally authorised to operate in the EEA, plus a public non-compliant-entities list. Phase 2 should wire this first, alongside the EU DSA Transparency Database flagged by sector 17. WARNING: the direct CSV path could not be resolved during this research (the dated /sites/default/files/ URL 404'd); Phase 2 must first scrape the current download links from the MiCA landing page rather than hard-coding a path. Migration to ESMA's formal IT systems was planned for mid-2026 and will change URLs.

Coverage: EEA · from 2025

Access (inherited): bulk_download · free

ESMA legal notice permits reuse with source acknowledgement.

Commercial analytics business serving the tokenisation industry, so the incentive runs toward larger headline numbers. Publishes both 'distributed asset value' and 'represented asset value' for the same market, which differ by roughly ninefold; commentary routinely quotes the larger one. Deeper metrics are behind a login.

The best available view of tokenised Treasuries by product and issuer, which is cross-checkable against fund administrator NAVs. ALWAYS record which value definition is being quoted. Its reported holder counts moved +111% in 30 days during this research, which is not organic adoption and should never be used as an adoption metric.

Coverage: Global · from 2022

Access (inherited): public_page · freemium

Terms require attribution; bulk extraction and redistribution restricted. API access is commercial.

Open-source and non-commercial relative to peers, but TVL is a price multiplied by a quantity and moves with token prices rather than deposits. Protocol teams can influence inclusion and labelling. Volume figures include incentive-driven and wash activity.

Free public API is the practical reason to use it over commercial peers. Best used for open interest (which requires posted margin and is expensive to fake) rather than volume. Its own home dashboard and category pages disagree slightly on aggregates, so pin one endpoint per metric. Never cite TVL without decomposing the price component.

Coverage: Global · from 2020

Access (inherited): api · free

Open data with a free public API; attribution requested.

Commercial analytics vendor serving crypto firms and increasingly traditional financial institutions. Tracks a wider stablecoin set (155 tokens) than peers, which produces a systematically higher total; that is a definitional choice, not an error.

Publishes active-address and transaction-count series that are more useful than supply for judging real usage. Its total stablecoin supply reads about 4% above DefiLlama's and RWA.xyz's on the same date; carry all three and report the spread rather than picking one.

Coverage: Global · from 2021

Access (inherited): public_page · freemium

Attribution required; API and bulk access commercial.

Published by an incumbent card network with a commercial interest in not overstating a competing rail — but the adjustment methodology is disclosed in detail, which is more than any crypto-native source provides. The conservatism is documented rather than hidden.

THE reference for separating genuine stablecoin payments from bot, bridge, MEV and exchange-internal traffic. Adjusted criteria exclude addresses exceeding 1,000 transactions or $10m in 30 days, CEX flows, mint/burn and infrastructure operations. CAVEAT: the dashboard is JS-rendered and returned methodology text but no figures on automated fetch. Phase 2 needs a headless browser or the underlying Allium data; budget for it, because this is the only credible adjusted series in the sector.

Coverage: Global · from 2023

Access (inherited): public_page · free

Public dashboard; attribution to Visa and Allium required.

Issuer-published but computed by the fund administrator and struck daily under the Investment Company and Securities Acts. Marketing copy on the same page is promotional; the net asset and performance table is not.

THE cleanest adoption signal in this sector: custodied, daily-marked, independently administered assets. Confirmed working by direct fetch. Poll one page per product across IBIT, ETHA, FBTC, GBTC, BITB, ARKB and the Solana/XRP products. This is what should be used instead of any third-party flow aggregator.

Coverage: US listed products · from each fund's inception (IBIT 2024-01-05, ETHA 2024-06-24)

Access (inherited): public_page · free

Issuer terms of use; data is published for investor information. Attribution expected.

Company-authored and selectively timed. Lease and financing announcements are usually accompanied by an 8-K, which is the version to rely on; the press release is the version to read for counterparty names and megawatt figures the 8-K may redact.

Retrieval lesson: IR index pages on the standard IR platforms fetch reliably as listings but individual detail-page URL patterns vary and frequently 404 when constructed by hand. Scrape hrefs from the listing rather than guessing. Watch for company renames as a signal in their own right - ciphermining.com now 302s to cipherdigital.com.

Coverage: US listed miners · typically 5-10 years

Access (inherited): public_page · free

Company terms; press releases are issued for public distribution.

Issuer-controlled. Circle's reserves sit in an SEC-registered 2a-7 fund and are corroborated by its 10-Q; Tether publishes attestations rather than audits and its reserve quality has been independently questioned by Federal Reserve staff. Do not treat the two as equivalent evidence.

CAVEAT FOUND IN THIS RESEARCH: both Tether's and Circle's live transparency dashboards are JS-rendered and returned descriptive text but no figures on automated fetch. The monthly attestation PDFs are the fetchable artefact. For Circle, the 10-Q is strictly better evidence and should be preferred. Phase 2 needs a headless browser for the live balances or should rely on on-chain supply instead.

Coverage: Global · from 2021 for attestations

Access (inherited): public_page · free

Issuer terms; attestation PDFs are published for public reliance.

Project sponsor communications. Timelines are aspirational and conditional on legislation the ECB does not control; they have slipped repeatedly since 2021. Cost estimates are the ECB's own and are contested by banking associations.

The falsifiable checkpoint for the entire digital euro thesis is whether EU legislation is adopted during 2026. Monitor that, not the rulebook drafts.

Coverage: Euro area · from 2020

Access (inherited): public_page · free

Reuse permitted with source acknowledgement.

Regulator communications, consumer-protection oriented. Standing guidance pages are not always synchronised with the latest policy statement, which produced an unresolved conflict during this research over retail access to crypto exchange-traded notes.

Key dates: final rules published 2026-06-30, FSMA cryptoasset permissions from 2027-10-25. IMPORTANT: always cross-check the standing guidance page against the most recent policy statement; this source registry records a live conflict between the two on retail cETN access.

Coverage: UK · from 2018

Access (inherited): public_page · free

Open Government Licence.

Regulator communications. The English site is a partial and lagging translation of the Japanese original; substantive working-group reports often appear in Japanese first and sometimes only.

REGIONAL SOURCE for this sector. Japan has run a statutory crypto exchange regime since 2017, longer than any Western jurisdiction, so its loss-and-recovery experience is genuinely informative. The 2026-02-16 Financial System Council working group report is the key 2026 document. CAVEAT: the English index lists items but individual English detail-page URLs 404'd during this research; Phase 2 should poll the Japanese index at fsa.go.jp/news/ and translate.

Coverage: Japan · from 2000s

Access (inherited): public_page · free

Reuse generally permitted with attribution under Japanese government terms.

Regulator and promotion agency in one: MAS both supervises and actively markets Singapore as a digital asset hub, so its tokenisation programme communications have a promotional dimension that a pure supervisor's would not.

ACCESS BLOCKED during this research: mas.gov.sg news and media-release pages returned HTTP 403 to automated fetch on every attempt. Phase 2 needs either the RSS endpoint (untested) or a different user-agent/access arrangement. Recorded here because Project Guardian is the most substantive multi-bank tokenisation programme outside the EU and its absence is a real coverage gap.

Coverage: Singapore and ASEAN · from 2000s

Access (inherited): public_page · free

Singapore Open Data Licence for most published material.

None. Authoritative record of proposed and final rules, comment deadlines and effective dates.

The definitive way to track whether the GENIUS Act regime is actually in force, as distinct from proposed. CAVEAT: /documents/search is robots-disallowed and the /api/v1/ endpoint was PROXY_REJECTED (HTTP 403) during this research. Phase 2 should test direct API access from a non-proxied path; this is the highest-value unexercised structured source for this sector.

Coverage: US · back to 1994

Access (inherited): api · free

Public domain US government work.

Agency communications. Charter approvals are conditional and the conditions are often the substance; read the approval order, not the press release.

ACCESS BLOCKED: every occ.gov and occ.treas.gov path attempted during this research returned ROBOTS_DISALLOWED, with robots.txt itself failing to fetch or timing out through the proxy. This is a significant gap because the OCC is the chartering authority for non-bank stablecoin issuers. Phase 2 must resolve access or fall back to the Federal Register for OCC rulemakings and to sector 07's coverage for charter decisions.

Coverage: US · back to 1990s

Access (inherited): public_page · free

Public domain US government work.

Internal research product, not an external source. Its own sourcing is Tier A but it is a secondary compilation and should be treated as a pointer to primary sources rather than as evidence in its own right.

Owns the boundary: GENIUS Act rulemaking from the bank side, tokenised deposits including JPMorgan Kinexys and the 17-bank Clearing House network, and the deposit-disintermediation debate. Sector 21 covers the token-market side. Read together, especially sections 8-10.

Coverage: US and EU · 2026 research cycle

Access (inherited): public_page · free

Internal use.

Regulator plus hub-promotion role, similar to MAS. Hong Kong's digital asset policy is also an instrument of broader positioning relative to the mainland, which shapes what is announced and when.

ACCESS BLOCKED: every HKMA and SFC page attempted returned an empty body (JS-rendered) or 404 during this research. Hong Kong's Stablecoins Ordinance licensing regime and Project Ensemble are material to the Asian picture and are entirely missing from this dossier as a result. Phase 2 needs a headless browser or the RSS endpoint. Flagged as a priority gap.

Coverage: Hong Kong SAR · from 1990s

Access (inherited): public_page · free

Reuse permitted with attribution.

Management narrative is promotional; the exhibits and financial statements are not. In this sector specifically, management commentary on 'institutional adoption' routinely lacks any disclosed volume figure, while the same filing's revenue lines are precise.

The quarterly cadence of COIN, CRCL and MSTR is the sector's most reliable ground truth. Retrieval lesson: fetch the accession directory listing first to identify the primary document, because the submissions JSON frequently fails to expose form types through summarising fetch tools and adjacent accession numbers are often Form 4s rather than the 10-Q.

Coverage: US registrants · back to 1990s

Access (inherited): public_page · free

Public domain filing wrapper; exhibit content is company copyright but quotable.

Central bank research; no commercial interest. Underemployment definition (graduates in jobs not typically requiring a degree) is a methodological choice that drives the headline 42% figure - understand it before citing.

THE single most decision-relevant source in this sector. Unemployment and underemployment update in February, May, August and November; wages and outcomes-by-major update annually in February. The interactive charts are JS-rendered, so the underlying figures are hard to extract by fetch - Phase 2 should locate the backing data files or use the API-less quarterly text. Current values: 5.6% unemployment, 42% underemployment (2026:Q2).

Coverage: United States · Series back to the early 1990s

Access (inherited): public_page · free

US government work, freely reproducible with attribution.

None material. Household-survey sample sizes make age-cell estimates such as 20-24 volatile month to month - do not read single-month moves as trend.

Table A-4 gives unemployment by educational attainment (bachelor's+ 2.7%, Aug 2026); Table A-10 gives age cells (20-24 at 7.1%, from 9.2%). Also the only routine read on local government education payrolls (+42,000 in Aug 2026). Fetches cleanly at the .htm table URLs.

Coverage: United States · Back to 1948

Access (inherited): api · free

US government work, public domain. BLS API requires a free registration key for high-volume use.

Model-based forecasts, not observations. BLS assumptions about AI adoption are judgemental and revised; always label output as forecast, never as fact.

The 2025-2035 round (released 2026-08-27) is the key artefact: total growth of just 3.5% versus 10.9% in 2015-25, with explicit generative-AI attribution. The occupations-with-largest-declines table is the single best structured view of which entry-level white-collar roles are projected to shrink. Static .htm tables fetch reliably.

Coverage: United States · Projections published biennially/annually since the 1970s

Access (inherited): bulk_download · free

US government work, public domain.

Establishment survey; industry detail is in supplementary tables not the summary release, and educational services detail requires pulling the tables directly.

The hires rate is the leading indicator for the entry-level thesis - a low-hiring, low-firing market (hires 3.2%, layoffs 1.0%, July 2026) is precisely the regime in which entry-level compression happens without showing up as unemployment.

Coverage: United States · Back to December 2000

Access (inherited): api · free

US government work, public domain.

Institution self-reported; lags by roughly 1-2 years. Definitional changes across collection years break some longitudinal series.

The census of US postsecondary institutions - enrolment, finance, staffing, completions, and the only authoritative basis for a current college-closure count. Phase 2 should build the closure series from IPEDS directly, because no current published closure count could be found (SHEEO and Chronicle figures are both pre-2021).

Coverage: United States · Back to 1980 (predecessor HEGIS to 1966)

Access (inherited): bulk_download · free

US government work, public domain.

Non-profit with near-census coverage but not complete - a small share of institutions do not participate, and coverage rates differ by sector. A known methodological error affected Fall 2024 freshman figures, later corrected.

Reports enrolment 6-18 months ahead of IPEDS, which makes it the sector's best early-warning instrument on the demographic cliff and on the certificate-versus-degree mix shift. Only Spring 2025 was retrievable at this research date; Fall 2025 and Spring 2026 reports should be checked first on any re-run.

Coverage: United States · Back to 2011

Access (inherited): public_page · free

Free for non-commercial use with attribution; redistribution of underlying data is restricted.

Projections, not observations - but built on already-born cohorts, which makes them unusually reliable for a forecast. State-level figures carry more error than national ones.

The definitive US high-school graduate projection: peak 2025, -13% through 2041, declines in 38 states. Published roughly every four years, so this is a low-frequency but extremely high-value source. Downloadable state-level data files.

Coverage: United States · Editions since 1979; projections run to 2041

Access (inherited): bulk_download · free

Free to use with attribution.

Produced with US State Department support, which gives it official standing but also a public-diplomacy framing that favours headline growth numbers. The headline total includes OPT participants, who are workers not enrolled students.

CRITICAL TRAP for Phase 2: track new enrolments and OPT separately from the total. AY 2024/25 showed total +5% while new enrolments fell 7% and OPT rose 21%. A pipeline that ingests only the headline will produce the wrong sign on international demand.

Coverage: United States (inbound and outbound) · Back to 1948

Access (inherited): public_page · free

Free with attribution; bulk data access requires IIE permission.

Administrative counts, no methodological bias, but publication is irregular and has lagged materially. Visa issuance is not enrolment - students may be issued a visa and not enrol.

The fastest available leading indicator on international student flows - F-1 issuance leads enrolment by roughly one to two quarters. ACCESS PROBLEM: the ICE SEVIS PDF returned HTTP 403 and the State Department topic page returned 403 through the proxy; the monthly issuance page is public but was only current to September 2025 at this research date. Phase 2 needs a resolved access route here.

Coverage: United States · Monthly files by fiscal year

Access (inherited): bulk_download · free

US government work, public domain.

Gold-standard assessment methodology. The risk is not the data but its political framing in secondary releases - proficiency thresholds are frequently misdescribed.

The only credible measure of US K-12 learning outcomes. ACCESS PROBLEM: report and highlights pages returned redirect loops at this research date; only the 2025 Long-Term Trend headline was retrievable. The NAEP Data Explorer API is the better Phase 2 target than the report pages.

Coverage: United States · Main NAEP to 1990; Long-Term Trend to 1971

Access (inherited): public_page · free

US government work, public domain.

Primary source for policy facts (dates, rules, approvals) but carries explicit administration framing on outcomes and attribution. Use the numbers, discount the narrative.

The authoritative feed for Workforce Pell approvals, which is currently the highest-signal policy series in the sector: five programmes across three states between 2026-08-04 and 2026-09-14. Individual release detail pages fetch reliably; the index page is usable.

Coverage: United States · Archived by administration

Access (inherited): public_page · free

US government work, public domain.

Audited and legally binding; no bias. Coverage is shrinking as K-12 infrastructure vendors go private (Instructure, PowerSchool) and 2U exited via bankruptcy.

Covers CHGG, COUR, DUOL and Pearson's US filings. RETRIEVAL NOTE confirming the macro-brief playbook: browse-edgar is robots-disallowed and data.sec.gov was rejected via curl through the proxy (connect_rejected), but WebFetch on data.sec.gov/submissions/CIK##########.json and then on /Archives/ document URLs works. Chegg CIK 0001364954.

Coverage: United States · Back to 1993

Access (inherited): api · free

US government work, public domain. SEC requires a declared user agent; rate limits apply.

Commercial database with a commercial interest in funding activity appearing dynamic. Systematically undercounts undisclosed and non-US rounds; totals differ materially from other trackers (see the Crunchbase/KPMG divergence in the macro brief).

The only regularly updated public read on edtech funding: ~$2.8bn global and ~$1.2bn US in 2025 YTD. Treat totals as one tracker's count, never as the market. Pair with a second tracker before asserting any annual figure.

Coverage: Global, US-weighted · Back to 2007

Access (inherited): rss · freemium

Editorial content free to read and cite with attribution; underlying database is licensed and may not be redistributed.

Transparent methodology and published field dates and sample sizes. Some studies are funded by partners (Lumina, Walton, NewSchools) whose interests should be noted; partner-funded studies are still methodologically sound but topic selection is not neutral.

Maintains the longest continuous series on perceived value of college (75% in 2010 to 35% in 2025) and on K-12 satisfaction (record-low 35%). Also the source for teacher AI use (60% using, 30% weekly, June 2025) - which I could only obtain second-hand and which Phase 2 should verify at source.

Coverage: United States · Some series back to the 1970s

Access (inherited): public_page · free

Free to cite with attribution; full datasets are licensed.

Non-partisan, publishes field dates, sample sizes and margins of error, and releases topline questionnaires. Measures attitudes and self-reported behaviour, not outcomes - do not treat perception series as employment data.

The authoritative nationally representative measure of teen AI use (54% for schoolwork) and of public expectations on AI and jobs (73% of under-30s expect fewer jobs). Datasets are downloadable after free registration, which makes this unusually good for Phase 2 reanalysis.

Coverage: United States · Back to 2000

Access (inherited): rss · free

Free to reproduce with attribution under Pew's terms.

Working papers, some not yet peer reviewed; versions are revised without prominent version notices, so always record the revision date. Depends on proprietary ADP data that other researchers cannot fully replicate.

Source of the 19% entry-level divergence finding (revised 2026-08-12). The single highest-value academic source for this sector's central question. Phase 2 must capture the authors' own causality caveat alongside the number, because secondary coverage consistently strips it.

Coverage: United States · Lab founded 2020

Access (inherited): public_page · free

Free to read and cite; underlying ADP microdata is not public.

Harmonised across member states, which is its strength; harmonisation also means national definitions are adjusted and may differ from countries' own published figures.

The best non-US structured source in this sector and the one regional source that fetched reliably. 2025: 44.8% tertiary attainment among 25-34s against a 45% 2030 target, with an 11.3pp female-male gap. Has a genuine REST API - Phase 2 should wire it.

Coverage: European Union and EFTA · Back to the 1990s for most series

Access (inherited): api · free

Free reuse with attribution under the Commission's reuse policy.

Statutory regulator; no commercial bias, but a regulator has institutional reasons to present sector risk carefully. England only - Scotland, Wales and Northern Ireland have separate regulators.

The authoritative read on English HE financial distress and international-recruitment dependence. Financial sustainability of higher education providers in England 2026 was published 2026-05-14. ACCESS NOTE: publication landing pages fetch but the substantive figures sit in PDFs that did not retrieve - Phase 2 should fetch the PDF and Annex data files directly.

Coverage: England · Back to 2018 (predecessor HEFCE earlier)

Access (inherited): public_page · free

Open Government Licence - free reuse with attribution.

Authoritative UK-wide official statistics. Some detailed tables are behind a paid data-service licence, and open tables lag the reference year by roughly 12 months.

BLOCKED AT THIS RESEARCH DATE: returned HTTP 403 through the proxy. This is the authoritative UK student and staff dataset and its absence is the largest single non-US gap in this dossier. Phase 2 needs either a resolved fetch route or a HESA data licence - flag for the licensing budget.

Coverage: United Kingdom · Back to 1994/95

Access (inherited): bulk_download · freemium

Open tables under Open Government Licence; detailed datasets require a paid HESA data licence and prohibit onward redistribution.

Cross-nationally comparable by construction, which requires definitional adjustments that diverge from national figures. OECD framing reflects member-government priorities.

BLOCKED AT THIS RESEARCH DATE: every landing-page URL pattern attempted returned HTTP 404, so no OECD figure appears in this dossier. This is the second major non-US gap. The OECD.Stat / SDMX API is the more reliable Phase 2 route than the publication pages.

Coverage: OECD member and partner countries · Back to 1992

Access (inherited): bulk_download · free

Free reuse with attribution under OECD terms; some publications are paywalled in full-text form.

Nationally representative panel of US public schools with transparent methodology. Topic rotation means any given measure (staff vacancies, technology use) is collected only intermittently.

The only regular federal instrument on school staffing vacancies and technology use. LIMITATION: the most recent collection retrievable was February 2024, and technology/AI questions were last fielded in August 2023 - so there is currently no federal measure of AI adoption in US schools. That absence is itself a finding and a Phase 2 priority.

Coverage: United States · Launched 2021

Access (inherited): bulk_download · free

US government work, public domain.

Administrative counts, no methodological bias. Sponsored study visa grants are not the same as enrolments, and dependant rules changed materially in recent years, breaking comparability across the series.

The UK counterpart to SEVIS and the key instrument on UK international student demand. Year ending June 2026 was published 2026-08-27. ACCESS NOTE: the release landing page fetches but the topic sub-pages returned 404; use the immigration-system-statistics data tables bulk files instead of the narrative pages.

Coverage: United Kingdom · Back to 2005

Access (inherited): bulk_download · free

Open Government Licence - free reuse with attribution.

Issuer-prepared. Non-GAAP measures (adjusted EBITDA, non-GAAP net income) are selected by management; acquisition effects are not always separated from organic growth in headline percentages.

Confirms the macro-brief retrieval playbook: IR index pages are JS-rendered and return navigation only, but individual press-release detail pages and static-file URLs fetch reliably. Go straight to the release URL. CAVEAT: Coursera's Q2 2026 release carried no printed publication date on the retrieved page - Phase 2 should capture release dates from the corresponding 8-K rather than the IR page.

Coverage: United States (global operations) · Since each company's IPO

Access (inherited): public_page · free

Public company disclosure, freely citable.

Membership-funded consortium of vendors and institutions; specifications reflect member interests, and certification is a commercial product.

Determines integration cost across the sector and, through the Comprehensive Learner Record and Open Badges, the technical feasibility of portable sub-degree credentials - the infrastructure precondition for the Workforce Pell credential tier to actually stack.

Coverage: Global · IMS Global founded 1997

Access (inherited): public_page · free

Specifications free to read; conformance certification requires paid membership.

Employer and career-services membership body; respondents self-select and skew toward large employers with formal campus recruiting programmes, so it overstates the structured end of the graduate labour market.

The employer-side counterpart to NY Fed graduate outcome data - hiring intentions rather than realised outcomes, which makes it a leading indicator for the entry-level thesis. Most content sits behind membership; Phase 2 should budget for a membership or rely on the free press releases only.

Coverage: United States · Job Outlook series back to the 1970s

Access (inherited): public_page · freemium

Press summaries free; full reports require membership or purchase.

Issuer self-reporting. LVMH discloses revenue and recurring profit by business group but gives only qualitative regional commentary in the press release, which systematically understates how much regional detail the sector actually needs.

CRITICAL RETRIEVAL NOTE: the /en/publications index page is JavaScript-rendered and returns no items to automated fetching. The English sitemap at https://www.lvmh.com/en/sitemap/en-us.xml DOES list every publication URL and is the reliable discovery route. Individual release detail pages fetch cleanly.

Coverage: Global, France-domiciled · press releases archived back over a decade

Access (inherited): public_page · free

Copyright LVMH; press releases reproducible with attribution. No scraping restriction observed on individual release pages.

Issuer self-reporting. Unusually granular: gives revenue by geographic area AND by metier with constant-rate growth for both, which most peers do not.

The press-release listing page fetches cleanly and exposes direct PDF URLs on assets-finance.hermes.com, which are the highest-value artefacts. Best single source for supply-constrained luxury demand by region.

Coverage: Global, France-domiciled · press releases and PDFs back several years

Access (inherited): public_page · free

Copyright Hermes; PDFs public. Attribution expected.

Issuer self-reporting under SIX ad hoc disclosure rules (Art. 53 LR), which raises the compliance bar. Fiscal year to 31 March means its data leads calendar-year peers by a quarter.

Highest-value single luxury source in this sector: it is the only large group that discloses operating margin BY DIVISION, which is what exposes the jewellery-versus-watches divergence. Results pages and PDF links both fetch reliably.

Coverage: Global, Switzerland-domiciled · results archive back many years

Access (inherited): public_page · free

Copyright Richemont; ad hoc announcements are regulatory disclosures and freely quotable.

Issuer self-reporting. L'Oreal routinely characterises its own growth relative to 'the beauty market' without publishing the market denominator, so its outperformance claims are unfalsifiable from the release alone.

RETRIEVAL NOTE: the English landing page lists releases but the working detail URLs are under the FRENCH path /fr/communique-de-presse/<slug>. Best source for the divisional split of beauty growth (Dermatological Beauty and Professional versus Luxe and Consumer).

Coverage: Global, France-domiciled · releases archived back many years

Access (inherited): public_page · free

Copyright L'Oreal. Press releases quotable with attribution.

Issuer self-reporting, additionally constrained by SEC disclosure obligations. Discloses mainland China organic growth separately, which very few beauty issuers do.

The /investors/press-releases-and-news page yields working detail URLs; the /news-and-media/newsroom/press-releases page does not list the investor items. Also the best single source on tariff cost versus tariff refund in prestige beauty.

Coverage: Global, US-domiciled · releases by year back several years

Access (inherited): public_page · free

Copyright ELC; also filed with the SEC as 8-K exhibits, which are public domain as government records once filed.

Issuer self-reporting. Headline growth rates are inflated by the rhode acquisition and by tariff refunds, and the release does not always separate organic from acquired growth.

The single clearest disclosure anywhere in this sector of how large IEEPA tariff refunds are as a share of gross margin - roughly 1,050 of 1,400 basis points in Q1 FY2027.

Coverage: US and international · quarterly releases back several years

Access (inherited): public_page · free

Copyright e.l.f. Beauty; also SEC-filed.

Issuer self-reporting. NIKE stopped disclosing revenue by geographic segment in the earnings release body, giving only directional commentary, which materially reduces its usefulness for the China question.

RETRIEVAL NOTE: investors.nike.com investor-news detail URLs 404 to automated fetching; about.nike.com/en/newsroom/releases/<slug> works. The FY2026 release quantified a $986m IEEPA tariff recovery, the largest single such disclosure found in this sector.

Coverage: Global, US-domiciled · releases back many years

Access (inherited): public_page · free

Copyright NIKE; also SEC-filed as 8-K exhibits.

Issuer self-reporting. Reports currency-neutral growth as the headline, which flatters comparison against peers reporting in reported currency. Excludes potential tariff refunds from guidance, which is conservative.

The financial-reports index page fetches cleanly and exposes direct PDF URLs on res.cloudinary.com, which then fetch reliably. Currently the single best counter-example to the 'Western sportswear is in decline' framing.

Coverage: Global, Germany-domiciled · results publications back a decade

Access (inherited): public_page · free

Copyright adidas AG. PDFs public.

Issuer self-reporting during a severe downturn, so commentary is unusually candid on volume decline but gross margin is flattered by tariff refunds booked through cost of sales.

Index page exposes stable PDF paths at about.puma.com/sites/default/files/financial-report/<year>/. Quantifies tariff refunds in euros, which makes it a useful cross-check on the US-reporting brands.

Coverage: Global, Germany-domiciled · results back a decade

Access (inherited): public_page · free

Copyright PUMA SE. PDFs public.

Issuer self-reporting. The Q2 FY2026 release reports gross margin expansion in the headline while the underlying margin contracted once tariff refunds are removed - a good worked example of why refunds must be stripped before comparison.

Listing page gives dates and titles; detail URLs are /newsroom/press-releases/<year>/<id>. The clearest single case of a court-ordered refund masking a demand reversal.

Coverage: Americas, China, international · releases by year

Access (inherited): public_page · free

Copyright lululemon; also SEC-filed.

Issuer self-reporting. H&M gives almost no quantified tariff disclosure - a single unquantified note - which is itself informative about how European retailers are handling the topic relative to US peers.

Stable PDF paths at hmgroup.com/wp-content/uploads/<year>/<month>/. Fiscal year runs 1 December to 30 November, so its quarters do not align with calendar-quarter peers - a common comparison error.

Coverage: Global, Sweden-domiciled · interim reports back many years

Access (inherited): public_page · free

Copyright H&M Group. PDFs public.

Issuer self-reporting, and Zalando is commercially selling the AI content platform whose savings it reports, so the 90-95% efficiency figures are vendor-adjacent and should be labelled accordingly.

The most quantified public disclosure of generative-AI production economics in fashion anywhere: content production time, cost, articles processed per day, assistant users, robotic picks. Results pages at /en/financials/zalando-qN-YYYY-results fetch cleanly.

Coverage: Europe · quarterly results back many years

Access (inherited): public_page · free

Copyright Zalando SE.

Voluntary disclosure by a private company. The figures are not independently audited in the public domain, and the accompanying 'members saved EUR21.6bn' style claims are marketing, not accounting.

The only scaled resale platform disclosing a net profit, and therefore indispensable to any assessment of resale economics. Publishes annually in April. Also the sector's most useful non-French, non-Italian European regional source.

Coverage: Europe, Australia · annual results posts since 2023

Access (inherited): public_page · free

Copyright Vinted. Publicly posted.

Issuer self-reporting. Note that ThredUp's separately published annual Resale Report is a vendor-sponsored market-size document with no disclosed methodology and must be treated as Tier C marketing, not as data.

RETRIEVAL NOTE: use ir.thredup.com, not investors.thredup.com (which does not resolve). The GAAP loss against a 79.9% gross margin is the key counterweight to Vinted's profitability.

Coverage: US · quarterly releases since 2021 IPO

Access (inherited): public_page · free

Copyright ThredUp; also SEC-filed.

Primary legal text with no editorial position. The risk is user error: recitals and articles are easy to confuse, and long regulations truncate when fetched as HTML.

RETRIEVAL NOTE: the HTML view truncates long regulations and often returns only recitals. The PDF view at /legal-content/EN/TXT/PDF/?uri=OJ:L_YYYYNNNNN reliably returns the FINAL articles, which is where the transposition and application dates live. This is how the 17 April 2028 textile EPR deadline was confirmed.

Coverage: EU · complete EU legal corpus

Access (inherited): public_page · free

Reuse of EU legal texts permitted under Commission Decision 2011/833/EU with attribution; the official version is the one published in the Official Journal.

Official but promotional in tone: describes what legislation will achieve rather than what it currently requires, and often omits the operative dates.

Useful for identifying which instrument to go and read on EUR-Lex; not sufficient on its own because it omits article-level dates. Confirms Directive (EU) 2025/1892 entered into force 16 October 2025.

Coverage: EU · current policy only; little historical versioning

Access (inherited): public_page · free

Commission Decision 2011/833/EU reuse terms with attribution.

Official legislative record. French-language primary source; the dossier tracks passage but does not carry implementing decrees, which is where the operative thresholds will appear.

RETRIEVAL NOTE: rate-limits aggressively (HTTP 429) - space requests. Confirmed Loi n. 2026-602 promulgated 2026-07-08. The follow-on to monitor is Legifrance for the decrets en Conseil d'Etat, which are the actual trigger for the advertising ban.

Coverage: France · full legislative history of each dossier

Access (inherited): public_page · free

French public-sector information, reusable with attribution under the Licence Ouverte.

Statistically rigorous and non-commercial. The caveat is methodological: apparel CPI is quality-adjusted, and core CPI at 2.4% sits far below core PCE at roughly 3.3%, so the choice of index materially changes the story.

The definitive measure of tariff pass-through into consumer apparel and personal-care prices. Public API available at api.bls.gov/publicAPI/v2. Detailed tables at /news.release/cpi.t02.htm fetch cleanly.

Coverage: United States · apparel and footwear series back to 1947

Access (inherited): api · free

US Government work, public domain. BLS asks that data be attributed and not presented as official BLS analysis.

Government forecast, revised monthly. Projections are model outputs and should be labelled forecast, not fact; the same report's import-share data is observed rather than modelled.

RETRIEVAL NOTE: the publication landing page does not contain the data; take the PDF at ers.usda.gov/media/<id>/cws-26<letter>.pdf and the accompanying XLSX. Carries both the cotton price forecast and US cotton-product import shares by supplier country, so it does double duty as a sourcing-shift tracker.

Coverage: United States and world · monthly outlooks over many years

Access (inherited): bulk_download · free

US Government work, public domain.

Authoritative but incomplete as a timeline source: the overview page states obligations without their compliance dates, and rulemaking status changes without the page being restructured.

Carries a 'content current as of' date (08/10/2026 at time of research), which should be recorded. For actual deadlines, the Federal Register is the better feed; this page is the index to what is outstanding - GMP and fragrance-allergen rulemaking both remain unfinished.

Coverage: United States · MoCRA material since 2023

Access (inherited): public_page · free

US Government work, public domain.

Primary regulatory record with no editorial bias. Coverage is limited to US-listed issuers, which excludes every large European luxury and fast-fashion group in this sector.

RETRIEVAL NOTE confirmed again in this sector: browse-edgar is robots-disallowed and direct curl to data.sec.gov was refused by the egress proxy; the working routes are WebFetch on data.sec.gov/submissions/CIK##########.json and directory listings under /Archives/edgar/data/<CIK>/. Covers NKE, EL, ELF, LULU, TDUP, REAL, GIII, COTY.

Coverage: US-listed issuers · filings back to 1994

Access (inherited): api · free

US Government work, public domain. SEC requires a declared user agent and applies rate limits.

Industry association with a clear interest in presenting Swiss watchmaking favourably, but the export series itself derives from Swiss customs data and is robust.

BLOCKED IN THIS RESEARCH: fhs.swiss returned robots.txt fetch failures to every automated attempt, so the watch-cycle reversal is corroborated only by issuer disclosure. PHASE 2 LICENSING TARGET - this is the only independent monthly demand series for hard luxury by destination country, and its absence is a real gap in the China question.

Coverage: Switzerland exports by destination market · monthly export series back decades

Access (inherited): public_page · freemium

Detailed statistics are subscription-gated; headline monthly press releases are free. Redistribution of the detailed series requires a licence.

Official customs-derived trade data with no commercial interest. Category definitions are technical and easy to misapply across HTS revisions.

BLOCKED IN THIS RESEARCH: otexa.trade.gov returned a robots.txt connect timeout, so sourcing-shift evidence had to be taken from the narrower USDA cotton-product import series instead. This is the single highest-value unexercised dataset for this sector and should be wired first in Phase 2, ideally via bulk download rather than page fetch.

Coverage: US imports and exports by country and category · monthly series back many years

Access (inherited): bulk_download · free

US Government work, public domain.

Issuer self-reporting from the largest fast-fashion group. Reports in euros with a February-January fiscal year, so its halves do not align with calendar-year peers.

BLOCKED IN THIS RESEARCH: every inditex.com path returned robots.txt fetch failures, so the sector's largest fast-fashion company contributed no 2026 data to this dossier. Phase 2 should route around this via the Spanish regulator CNMV, which holds the same regulated filings.

Coverage: Global, Spain-domiciled · results back many years

Access (inherited): public_page · free

Copyright Inditex; also filed with the CNMV as regulated information.

Issuer self-reporting. Kering's brand-level disclosure (Gucci, Saint Laurent, Bottega Veneta separately) is the most granular of the luxury groups and is the key input on whether the Gucci turnaround is working.

BLOCKED IN THIS RESEARCH: kering.com returned a hard site block to every automated fetch, so Kering's 2026 half-year results are absent from this dossier and its position is inferred only from the L'Oreal Gucci beauty licence announcement. Phase 2 alternative route: the French regulated-information store at info-financiere.gouv.fr, or the AMF.

Coverage: Global, France-domiciled · results back many years

Access (inherited): public_page · free

Copyright Kering; also filed as regulated information with the AMF.

Owned and funded by the major record companies. Counts trade (wholesale) revenue for IFPI-reporting labels; systematically under-counts unreported independent and self-released revenue, and excludes live and merchandise. Framing of AI is advocacy-led: the report is written partly to support a licensing-not-exception policy position.

Single most-cited global music revenue series. 2026 edition (published 2026-03-18) reported US$31.7bn 2025 trade revenue, +6.4%, 837m paid subscribers. Phase 2 needs the paid Premium edition for per-market detail; budget for it.

Coverage: Global, 50+ markets in the paid Premium edition · Annual series back to the 1990s; consistent format since 2016

Access (inherited): public_page · freemium

Headline State-of-the-Industry PDF is free to download with attribution. The Premium 'Data and Analysis' edition covering 50+ markets is a paid licence (gmr.ifpi.org) and may not be redistributed.

US-only and label-funded. Publishes BOTH estimated retail value and wholesale value, and press coverage routinely conflates the two — this is the single most common misquotation in the sector. Excludes publishing/songwriter income entirely.

Mid-year 2026 released 2026-09-01: ~$6.0bn retail, +6.9%. The same release cycle carries a wholesale figure (~$4.9bn) — always state which basis. Full-year 2025 wholesale was a record $11.5bn.

Coverage: United States · Interactive database back to 1973

Access (inherited): public_page · free

Free public use with attribution to RIAA; database terms prohibit bulk redistribution of the underlying series.

Panel-based: covers roughly 1,200 reporting US publishers. Panel composition changes between years, so year-over-year comparisons are restated rather than raw. Excludes most self-publishing and Amazon-exclusive titles, which is a large and growing blind spot.

Calendar 2025 annual (released 2026-08-10): $33.4bn total, +2.5%; trade $21.7bn. June 2026 month +11.0%, YTD +3.9% to $5.6bn. AAP is also the most active litigant-adjacent trade body in AI copyright and its news feed is the best single tracker of book-publishing AI litigation.

Coverage: United States · Monthly series since 2014

Access (inherited): public_page · free

Free with attribution. AAP asks that the panel caveat be reproduced alongside the figures.

An expert regulator's view, not binding law; courts have repeatedly declined to follow it. Part 3 remains a pre-publication draft, which is itself a political fact worth tracking.

Part 2 concluded prompts alone do not confer authorship and recommended NO legislative change. Part 3 pre-publication analysed market dilution as a fourth-factor harm and stopped short of recommending compulsory licensing. Whether Part 3 is ever finalised is a live indicator for Phase 2.

Coverage: United States · Series began 2023; Part 1 (digital replicas) 2024, Part 2 (copyrightability) 2025-01, Part 3 (training) pre-publication 2025-05

Access (inherited): bulk_download · free

US Government work, public domain. No restrictions.

Administrative counts, not economic measures. Registration volume tracks formality compliance and fee changes as much as creative output.

FY2025: 510,003 claims received, 415,780 registrations issued, $40.6m fee receipts, 12,310 recordation documents covering 5,704,306 works. Registration timeliness is now economically load-bearing because AI settlement eligibility turns on it.

Coverage: United States · Annual reports back to the 1890s; comparable electronic-era series from 2008

Access (inherited): bulk_download · free

Public domain US Government work.

A free mirror of PACER metadata, not the documents themselves. Coverage of newly filed cases lags by days; consolidated MDL member cases are indexed inconsistently, so a case can appear to vanish when it is in fact centralised.

CourtListener is robots-disallowed through this session's proxy, so Justia was the working substitute. Phase 2 should licence RECAP/CourtListener API access directly — this sector's single most important variable is litigation posture and free mirrors are not enough.

Coverage: US federal courts · Federal district dockets back to roughly 2004

Access (inherited): public_page · free

Docket metadata is public record. Justia's terms prohibit systematic scraping; use the public search interface or licensed PACER/RECAP access.

Only one of the three majors reports publicly and in this detail. Fiscal year ends 30 September, so WMG quarters do not line up with UMG's or Sony's calendar reporting — a frequent source of bad comparisons.

The single most useful Tier A window into major-label economics. Q3 FY2026 (quarter to 2026-06-30) disclosed $16m of digital revenue from copyright infringement settlements and an increase of the 'Beethoven Financing' catalogue-acquisition facility from $500m to $750m. Use data.sec.gov/submissions then fetch under /Archives/ — browse-edgar is robots-disallowed.

Coverage: Global (US-listed issuer) · Back to the 2005 IPO; continuous since the 2020 relisting

Access (inherited): api · free

Public domain filings. SEC fair-access policy requires a declared user agent and rate limiting.

Reports in euros; ARPU and subscriber figures are company-defined and have been redefined more than once. Spotify does not disclose per-stream rates and has never done so.

Q2 2026 (filed 2026-08-04): revenue EUR4.777bn +14%, 300m premium subscribers, 777m MAU, gross margin 33%, operating income EUR655m. Ad-supported revenue grew just 1% against 14% MAU growth — the most important number in the filing and the one least reported.

Coverage: Global · Back to the 2018 direct listing

Access (inherited): api · free

Public domain filings; SEC fair-access user-agent policy applies.

Marketing-adjacent transparency: Spotify chooses which cuts to publish and publishes only gross payouts to rightsholders, not what reaches artists after label and distributor splits. Cumulative 'paid out' figures are rounded and cannot be differenced year over year.

2026 edition (2025 data): >$11bn paid in 2025, nearly $70bn cumulative, >1,500 artists above $1m, >13,800 above $100k, the 100,000th-ranked artist above $7,300. Treat every figure here as label-and-distributor gross, not artist net.

Coverage: Global · Annual since 2021

Access (inherited): public_page · free

Free to cite with attribution; underlying data is not released.

Deezer is the only streaming service publishing AI-upload statistics, which makes it indispensable and unverifiable at the same time: there is no second platform series to triangulate against. Deezer also sells the detection technology it reports on, so it has a commercial interest in the numbers being large.

2026-07-21: ~90,000 fully AI-generated tracks uploaded daily, exceeding 50% of all daily uploads at peak in June 2026, 13.4m AI tracks tagged in 2025 — but only 1-3% of total streams, with up to 85% of streams on those tracks identified as fraudulent. The highest-value single disclosure in the sector. Note the domain 302-redirects https to http; fetch the http form.

Coverage: Global, France-weighted · AI-detection series since January 2025

Access (inherited): public_page · free

Press releases free to quote with attribution.

The only pure-play Western-listed webtoon issuer, so it is over-used as a proxy for a Korea- and Japan-centred industry. Reports in USD while most revenue is won- and yen-denominated, making headline growth heavily FX-distorted — always read the constant-currency line.

Q1 2026: revenue $320.9m, DOWN 1.5% year over year (+0.2% constant currency); IP adaptations -22.8%; adjusted EBITDA margin 3.0%; ~145m MAU. Japan ARPPU $22.5 versus Korea $7.8 and Rest of World $6.8.

Coverage: Korea, Japan, Rest of World · Back to the June 2024 IPO only — a short series

Access (inherited): api · free

Public domain filings.

Korean issuer reporting in won; the 'Contents' line bundles Webtoon with other content businesses so webtoon-only performance cannot be isolated. English materials are a translation of the Korean primary and occasionally lag it.

Q2 2026: group revenue KRW3,388.8bn +16.2%, operating profit KRW520.3bn -0.2%, margin down to 15.4% from 17.9%. Contents segment KRW464.3bn, +0.5% year over year. Required non-English/regional source for this sector.

Coverage: Korea, Japan, global · Quarterly series back to 2013

Access (inherited): public_page · free

Free to cite with attribution; IR PDFs are served from a download API endpoint rather than stable URLs.

Korean-language primary with an English deck; the 'Story' line (webtoon and web novel, including Piccoma in Japan) is reported without volume or user detail, so price/mix versus volume cannot be separated.

Q2 2026: group revenue KRW2,098.5bn +9.4%, operating profit KRW277.0bn +35.9%. Content segment KRW868.2bn +0.7%, within which Story (webtoon/web novel/Piccoma) fell 15.8% year over year while Music grew 7.8%. The cleanest available evidence that Korean webtoon is contracting while Korean music is not.

Coverage: Korea, Japan · Quarterly series back to 2014

Access (inherited): bulk_download · free

Free to cite with attribution. Fact sheets are published as PDF and XLSX on a CDN.

A small-cap acquirer, so its multiples are not the same as those paid for trophy catalogues by Blackstone- or KKR-backed vehicles. Investor-presentation multiples are company-calculated and unaudited.

FY2026 (to 2026-03-31): revenue $176m +11%, adjusted EBITDA $74m at 42% margin, ~$102m deployed on catalogue in the year and ~$980m since inception, 15.5x average entry multiple, net debt $430m. The only public, quarterly, auditable window on catalogue acquisition pricing.

Coverage: US, UK, MENA · Back to the 2021 SPAC listing