1 — Mission and Definitions

1 — Mission and Definitions

Phase 1 deliverable · research date 2026-09-15

1.1 The problem this platform exists to solve

There is no shortage of information about what is changing in the world. There is a severe shortage of information about which changes are real, how much they matter, and how confident anyone should be. The gap is not coverage. It is calibration.

Phase 1 research produced a concrete illustration. Across 25 sectors we recorded 201 contradictions between credible sources — not errors, but genuine methodological disagreements that the ordinary flow of trade press resolves by silently picking one number. Crunchbase and KPMG differ by $50.4bn on the same half-year of global venture funding. CBRE and JLL differ by roughly 9x on North American data-centre capacity under construction. IATA's press release and its own market-analysis report give opposite-signed figures for Asia-Pacific traffic in the same month. Each of these numbers circulates as fact.

Meanwhile the loudest claims turned out to be the weakest. Of the 50 trends our analysts classified as overhyped, the evidence base was systematically thinner than for trends of comparable prominence: the autonomous AI security operations centre has attracted over $300m of funding in twelve months and a deliberate search returned zero non-vendor efficacy evidence. "The creator economy is worth $X hundred billion" could not be traced to any nameable methodology, while the auditable figures are an order of magnitude lower.

The mission: to be the place where a decision-maker can find out not just what is happening, but how well it is known. Every claim carries its evidence, its date, its sources, its confidence, and — when sources disagree — the disagreement itself rather than a false resolution.

1.2 How the platform helps users

Each capability below maps to a specific mechanism in the Phase 1 frameworks, not to an aspiration. The mechanism is what makes the claim testable.

1.3 Definitions

These distinctions are not pedantry. Each one is a field in the database, and conflating any two of them is how trend research goes wrong. The examples are drawn from the Phase 1 research and are traceable to the dossiers.

The core distinction: trend vs. signal vs. fad

Trend — A directional change in a measurable quantity, sustained across multiple independent observations over a period long enough to exclude noise, with an identifiable causal mechanism. The mechanism requirement is what separates a trend from a correlation.

Example: Global industrial robot installations have plateaued near 542,000 units while China's domestic vendors moved from 30% to 57% of their home market between 2020 and 2024. Measurable, sustained, multi-source, with an identifiable mechanism (industrial policy plus cost position).

Emerging signal — An observable indicator that may precede a trend, detected before it is measurable in aggregate data. A signal is a hypothesis with evidence attached, not a small trend. Most signals do not become trends, and a platform that cannot say so is selling noise.

Example: Bitcoin miners converting to AI data-centre landlords with named AI-lab tenants. Real and documented, but not yet established as a durable sector reallocation.

Fad — A rapid rise in attention without a corresponding rise in adoption, revenue or capability, typically with a single dominant promoter class and no independent demand. The diagnostic is divergence: attention up, adoption flat.

Diagnostic in our scoring: high velocity, low adoption, low persistence, low evidence_quality. The metaverse-as-general-purpose-computing-platform scores this shape — Meta's Reality Labs lost $8,647m in H1 2026 on $833m of revenue, roughly 10:1, and no credible market model carries a separate metaverse revenue line.

Timeframe and causality distinctions

Market cycle — A recurring oscillation around a trend line, driven by capacity, inventory, credit or sentiment, which reverts. The critical property is that it has happened before and will happen again; the mistake is to read the upswing as a structural shift. Memory semiconductors have run four boom-bust cycles since 2007.

Short-term event — A discrete, dated occurrence. It is not a trend, but it can start, accelerate, or kill one. Events belong in timelines and are the raw material of trend detection; they are never trends themselves.

Example: The closure of the Strait of Hormuz. An event — which then reset jet fuel to 31.4% of airline operating cost, up from 25.4%, halved industry net profit to $23bn, and invalidated the entire February 2026 container-shipping bear case.

Structural shift — A change in the underlying rules of a system that does not revert: a change in cost curves, market structure, regulation, or physical constraint. The test: what would have to happen for this to reverse, and is that plausible?

Example: The flat 30% app-store commission was dismantled across the US, EU, Japan, China, Korea, UK and Brazil inside nine months. Restoring it would require reversing multiple independent legal and regulatory processes. That is structural.

Consumer behaviour change — A shift in what people actually do, measured in behavioural data rather than stated intent. Survey-reported intent is an estimate; transaction, traffic and usage data are fact. The two diverge constantly and travel is the worst offender.

Technological development — A change in what is technically possible or economically feasible. Distinct from adoption: a capability can exist for years before deployment, and the gap between demonstration and deployment is where most technology forecasting fails.

Example: Humanoid robots. Unitree has shipped 5,632 units cumulatively; Agility disclosed $1.8m of revenue against a $140m operating loss in its S-4. The capability is advancing; the deployment is not. Both statements are true and must be held together.

Regulatory trend — A directional change in rules, enforcement posture, or legal interpretation. Note that regulation moves through four distinct stages — proposed, enacted, in force, enforced — and vendor marketing routinely collapses them.

Example: The EU AI Act's high-risk obligations for employment AI were the most-marketed compliance deadline of 2026. They were deferred from 2026-08-02 to 2027-12-02 by the Digital Omnibus. Every vendor claim built on that deadline is stale.

Cultural movement — A change in shared values, norms or identity that alters demand without a price or technology driver. Slower, harder to measure, and consequential precisely because it is not priced in.

Epistemic status distinctions

These three are the most frequently conflated, and the conflation is usually deliberate.

Forecast — A conditional, probabilistic statement about the future with stated assumptions, a stated time horizon, and identified falsifiers. A forecast without a falsifier is a prediction.

Form: "If gas turbine lead times remain above 30 months and interconnection queues do not clear, then X by 2029. The indicator that would falsify this is Y."

Prediction — An unconditional claim that something will happen. Predictions are not produced by this platform. We record other people's predictions as opinion with attribution, because the track record of predictors is itself useful data.

Opinion — An attributed judgement, valuable when the holder's position makes their view informative, and always labelled as such.

Example: "Welcome to the AGI era" (Greg Brockman, 2026-09-03) is recorded as opinion, attributed and dated, set against independent indices showing the model in question tied with a competitor at 53 and losing 64%–20% on an independent code benchmark. The opinion is the datum; the benchmarks are the facts.

The claim-type ladder (enforced in the schema)

Every evidence item in the database carries one of these. Seed database distribution:

A marketing claim may never be rendered without its label. That rule is the single most important editorial control in the system, because unlabelled vendor claims are the primary vector by which hype enters research.

1.4 What this platform will not do

Stating the limits is part of the mission, and each limit below is enforced somewhere in the frameworks rather than merely promised.

• It will not present forecasts as certainties. Six-scenario analysis with explicit falsifiers is the required format for any major trend.
• It will not let volume substitute for quality. The evidence cap is arithmetic.
• It will not resolve genuine disagreements silently. 201 contradictions are recorded as contradictions. Where the honest answer is "credible sources disagree and here is why," that is the answer.
• It will not treat attention as importance. Search interest, funding volume, media coverage and social volume are inputs to velocity only, and velocity is 7.5% of the composite score.
• It will not hide what it does not know. Every dossier carries a data-gaps section. The Phase 1 research deliberately records that no frontier AI lab discloses a gross margin, that no operator anywhere publishes robotaxi unit economics, and that no independent audited DPI benchmark exists for venture capital. These absences are findings.

More research from the library · Explore trend records