Evidence authority
What each class of evidence is entitled to claim — and what it is not.
Six epistemic roles
Every measurement on this site carries one of these labels. They describe what kind of question the measurement is entitled to answer.
They are deliberately not presented as a ranking from good to bad. A diagnostic scan is not a defective official result; it is a different instrument used for a different purpose. Post-hoc forensic work is not second-rate evidence; it is the only way to examine the measuring instrument itself. The failure mode this list guards against is not low quality, it is category error — reading a diagnostic observation as if it were a frozen verdict.
Official / frozen result
The output of a decision contract that was locked before the results were seen.
What it can supportA binding verdict about whether the pre-registered criteria were met.
What it cannot supportAny claim the contract did not define, and any claim added after the results were seen.
Example in this researchModel-0 STOP-v1: decision STOP, measurable axes 0/8.
Source§2.4, §5.1
Official chain — calibration / selection stage
A stage inside the official pipeline whose job is to select difficulty candidates (d*), not to decide the outcome.
What it can supportThe statement that a level was structurally usable and was therefore carried into confirmation.
What it cannot supportA final inferential verdict. Calibration did not enter the gate family.
Example in this researchThe 30 Model-0 calibration cells that selected d* candidates.
Source§5.1, §5.2, §8.1
Post-hoc forensic
Analysis performed after the fact to examine the measurement instrument itself, without altering the official result.
What it can supportStatements about how the instrument behaved: difficulty ordering, format sensitivity, scoring artefacts.
What it cannot supportReopening or relaxing a frozen decision. A post-hoc trend does not become a gate pass.
Example in this researchThe Model-0 difficulty-scale audit and the 110-cell format ablation family.
Source§2.4, §5.4, §8.3
Diagnostic-only
Exploratory measurement that was never routed into a decision gate.
What it can supportDescriptive statements about what was observed in this run.
What it cannot supportInferential claims, population claims, or anything phrased as a general law.
Example in this researchThe Pilot-0.2 eight-checkpoint Birth Map and the Pilot-0.3 checkpoint scan.
Source§2.4, §8.2
Compatibility-adapter
A decision chain that ran through an architectural compatibility adapter and is therefore not recorded as a canonical artifact.
What it can supportA description of what that chain reported, clearly marked as non-canonical.
What it cannot supportEquivalence with a canonical STOP-v1 artifact.
Example in this researchThe Pilot-0.2 final n=800 chain, recorded as canonical_stop_v1_artifact: false.
Source§2.4, §6.4
Retrospective noncanonical observation
Weight observations extracted after the fact from checkpoints that were produced before the canonical observation flow was in place.
What it can supportProvenance and read-only state facts about those checkpoints.
What it cannot supportA canonical birth record, or LAB_READY evidence.
Example in this researchLegacy Pilot-0.2 and Pilot-0.3 Observatory records.
Source§2.4, §8.8
Why these are not quality grades
The classes cannot substitute for one another. A diagnostic result does not change an official gate result, and a retrospective observation does not become a canonical birth record.
That restriction runs in both directions. An official result is authoritative only about what its contract defined; it says nothing about questions the contract never asked. The Model-0 STOP verdict is binding about whether eight axes met a frozen bar. It is silent about how the measuring instrument behaved, which is exactly why a separate forensic audit exists.
Source: Scientific Evidence Inventory v1.0 (evidence cutoff 30 August 2026), §2.4, §8.1, §8.8.