Finding store
Each finding under its key, each run in its state
Corpus runs only. Where a finding sat in a run is a sighting beside it, never part of its key. A store or recording of another schema is refused whole — never migrated, never read in part.
{{command}}
Runs judged under rules v3 and v4 sit side by side; a comparison across them says that their states may differ by the rules alone.
{{f.version}}
{{f.statement}}
A finding says “signal X fired at place Y”. It carries no severity and no verdict.
Candidates for responsible disclosure — never confirmed weaknesses. A real-world entry’s findings are counted in fleet output but listed only here, queried apart from what may be published.
output_directive
series incomplete (3 unknown)
egress not observed
The digest is taken over results sorted by entry and scenario, without times, run ids or parallelism: two fleets over the same corpus in different orders have the same digest. An interrupted fleet has none — its results are not all there.
fleet diff compares the last two by default.Two runs of weather-http × all-signals
new
1
gone
1
unchanged
1
- new · result.tools[name="get_forecast"].description · hidden_comment
- gone · result.tools[name="get_alerts"].description · output_directive
- unchanged · result.tools[name="get_weather"].description · model_addressed — moved from tools[0] to tools[1], same key
basis: single run — one run’s word on each side, no interval.
Axes
Exactly one axis moved: new and gone are put down to the server.
Had either run been incomplete, diff would exit 3: what it calls gone may only be missing. A scenario drift keeps its exit code and marks statement_added / statement_removed.