Agent brief — tier: mid (above-average). One PR (Closes #<this>); branch from main. House rules: AGENTS.md.
Goal
Per docs/notes/consolidation-and-positioning.md, "the actual priority" is proving value: run the differentiation oracle on more real OSS C# repos and widen the set of findings Own.NET catches that CodeQL / Infer# miss.
Sources of truth
docs/notes/oracle.md — the method; docs/notes/oracle-known-fps.md — known FPs (do not re-report them as fresh).docs/notes/real-world-mining.md and docs/notes/mining.md — the write-up format and prior runs.docs/notes/field-notes-patterns.md — the pattern ledger to extend.- Workflows
mine.yml / oracle.yml, scripts/own-check.sh — the run surface (--severity warning is the oracle posture).
Deliverable
- Runs over 5–10 new OSS .NET repos (prefer WPF/WinForms/Avalonia + a couple of plain libraries for the clean-code control).
- A note under
docs/notes/ with a findings table: repo / finding (code, location) / draft verdict (TP / FP / unsupported) / rationale / cross-tool coverage where cheap.
- New recurring shapes appended to
field-notes-patterns.md; corpus reductions for clear TPs where feasible.
Guardrails — the triage boundary (important)
- Draft verdicts only for obvious cases. Anything borderline goes into an explicit "needs maintainer review" list with the evidence laid out — do not self-certify borderline TP/FP; final verdicts are reserved for the maintainer.
- Honest accounting: unsupported constructs and analyzer bails are counted, not hidden — a "clean" run that skipped half the files is reported as such.
- No analyzer code changes in this PR; precision gaps discovered here become their own issues (file them, one per pattern, linking the evidence).
Done when
The note + table are committed, ambiguous cases queued for review, new patterns ledgered, and any discovered precision gaps exist as linked follow-up issues.
Agent brief — tier: mid (above-average). One PR (
Closes #<this>); branch frommain. House rules:AGENTS.md.Goal
Per
docs/notes/consolidation-and-positioning.md, "the actual priority" is proving value: run the differentiation oracle on more real OSS C# repos and widen the set of findings Own.NET catches that CodeQL / Infer# miss.Sources of truth
docs/notes/oracle.md— the method;docs/notes/oracle-known-fps.md— known FPs (do not re-report them as fresh).docs/notes/real-world-mining.mdanddocs/notes/mining.md— the write-up format and prior runs.docs/notes/field-notes-patterns.md— the pattern ledger to extend.mine.yml/oracle.yml,scripts/own-check.sh— the run surface (--severity warningis the oracle posture).Deliverable
docs/notes/with a findings table: repo / finding (code, location) / draft verdict (TP / FP / unsupported) / rationale / cross-tool coverage where cheap.field-notes-patterns.md; corpus reductions for clear TPs where feasible.Guardrails — the triage boundary (important)
Done when
The note + table are committed, ambiguous cases queued for review, new patterns ledgered, and any discovered precision gaps exist as linked follow-up issues.