Announced
A provider or primary source says something changed.
Useful for awareness. Not treated as proof of real-world performance.Evidence layer · Public standard
Latest, without pretending.
AI changes quickly. Our job is not to repeat every announcement: it is to show what is known, what is uncertain and when the evidence should change your decision.
Seven evidence states.
These labels stop an announcement, an observation, a user report and a completed Human Bit test from collapsing into the same claim.
Observed
Availability or behaviour has been independently seen.
Conditions and limitations stay attached to the observation.Tested by us
A published, reproducible Human Bit protocol has been completed.
The model, settings, inputs, rubric, date and limitations must all be visible.Reported by users
A structured user outcome or issue has been recorded.
A report is useful field evidence, not a universal result.Inferred
A conclusion is derived from named evidence and stated assumptions.
The reasoning remains visible and can be challenged.Watching
A change may matter but is not ready to alter guidance.
It remains in view without becoming a recommendation.Unverified
Evidence is insufficient, unavailable or conflicting.
No confident guidance is built on it.Every record carries its own audit trail.
A source list is necessary, but it is not enough. Current model, capability and Signal records use the same typed evidence contract.
01
Claim
One canonical statement, with provider fact, outside observation and Human Bit guidance kept as separate typed claims.
Required field02
Trace
Each claim names its provenance, verification state, confidence and one or more registered sources.
Required field03
Fit
Freshness, readiness, audience, task, access and limits show where the claim can, and cannot, travel.
Required field04
Change
Affected guidance and a linked revision history make every later update inspectable instead of silently overwriting the record.
Required fieldWhere evidence comes from.
Source diversity helps us find claims. Source quality, transparent testing and applicability determine whether those claims become guidance.
01
Primary
Official release notes, documentation, policies and product pages
Source class02
Independent
Reputable reporting, research and documented evaluations
Source class03
Practitioner
Named field reports and repeatable community observations
Source class04
Internal
Our test method, conditions, result, limitations and review
Source classThe live register is public.
The Signal only publishes an item when it has a dated source, evidence state, human boundary, practical decision and next review date. Announced never silently becomes tested.