Evidence & decision layer · v1.1.0 · reviewed 2026-07-28
Methodology, uncertainty, and updates.
Claim statuses and uncertainty labels
Established result
A published result, defined construction, or documented methodology that the relevant literature treats as settled. It says what was shown, not that any broader capability claim follows.
Active research
A genuinely open or contested question. Competent researchers disagree on technical grounds and no side has closed it.
Conjecture
A precisely stated proposal motivated by evidence but not demonstrated. It may be true; it is not established.
Forecast scenario
A conditional scenario with explicit assumptions and failure modes. Not a prediction, not a probability statement, and not evidence about the world. Historical observations are never labelled this way, and forecast claims are never labelled anything else.
The public Atlas exposes only established and active-research claims. Each record also carries a limitation: this is the local uncertainty label, stating what the claim does not establish.
Source hierarchy
Technical claims rely on content-verified primary papers, benchmark documentation, or authoritative research-organization publications. Provider technical reports are labelled self-reports, not independent measurement. URL-only records and the local methodological artifact are excluded from public claim support.
Three different statements
Demonstrated capability is a result reported under specified inputs, system configuration, and conditions. Benchmark performance is a measurement on a defined task distribution under a scaffolding configuration at a date; it is not a scalar measure of general capability. Future projection is a conditional scenario or forecast claim, never evidence about the present and never a promise.
Forecasting boundary
Forecast-status records remain outside this public edition. When presented, they must state assumptions and failure modes, carry no confidence interval unless a method is published, and must not be read as predictions. Leaderboards, provider marketing, price trajectories, and unsupported capability timelines do not become claims merely by being current.
Update and correction policy
A new public claim requires a stable identifier, source resolution, status assignment, limitation, concept links, and a reviewed date. Material changes increment the Atlas version and update its modified date; a claim review updates the claim’s review date. Corrections are made on the claim page and matching JSON records together. Identifiers are not silently reused.
Publishing and educational context
The Atlas uses the site’s claim-level provenance standard and release registry so its records can be reused without losing their status or boundary. For educational methodology—not AI capability claims—see the Mayon Learning Library.