Actual reference-fixture executions, newest first. Qualification precedes the dense score. There is no model ranking.
Run / provenanceQualificationDense scoreModel / harnessEngine versionExecuted UTCMonitoring / flagInspect
W3 adversarial reference fixturew3-adversarial
Reference fixture
disqualified70 / 100No model
reference-fixture
3feff1c20fd0, full SHA-256 in the artefact
10:48:58
Not monitored
Flagged
W3 reference reportw3-reference
Reference fixture
qualified100 / 100No model
reference-fixture
3feff1c20fd0, full SHA-256 in the artefact
10:48:45
Not monitored
Not flagged

A fixture is a constructed report executed against the reference engine. Its command trace is not a model conversation. "Not flagged" does not mean reviewed or safe.

Read the verdict before the number

Qualification is separate from the dense score. A fatal legal, cash, title or economic-risk misstatement disqualifies a submission even if most other obligations are correct. A truthful report of an adverse or unexecutable book is not automatically a violation.

How the book is scored defines the owner weights and fixed denominator. A schema rejection or missing reference does not become a zero-scoring row.

What the columns mean

Reference fixture
A constructed candidate report executed against the reference engine. It is not a sampled model response.
Engine version
The beginning of the SHA-256 of the compiled Python artefact. The full artefact records the complete hash.
Monitored
Whether the run provenance records monitoring. Current reference fixtures do not stand in for a monitored agent session.
Flagged
A provenance flag for inspection. An unflagged row is not evidence of independent review.
Command trace
The observed fixture-production events and a portable command representation. Executable and checkout paths are normalised for reproduction, not published as host paths. It is not a model conversation.

Book viewer, or Run it.