Issues

Recovery paths had zero activations in a 9.85h run; two of the five inspected since were both defective, and their tests use shapes production never writes

#306

· gadievron opened

on Aug 20, 2026

SARIF output has no `invocations` block, so a degraded scan uploads indistinguishably from a clean one

#305

· gadievron opened

on Aug 20, 2026

Absent application context is reported as `application_type: "web_app"` rather than unknown

#304

· gadievron opened

on Aug 20, 2026

Shared venv `.pth` is re-pointed per clone, and child interpreters spawned mid-scan resolve through it (residual of #62)

#303

· gadievron opened

on Aug 20, 2026

Stage 2 only examines Stage-1 positives (4.1% of units) and can only downgrade, but no output states the denominator or the direction

#302

· gadievron opened

on Aug 20, 2026

Prune telemetry classifies 58% of prunes as missing-edge "orphan" candidates, then drops the counts before the deliverable

#301

· gadievron opened

on Aug 20, 2026

Standalone `openant verify` step report omits `needs_review` and `error_count` that the pipeline records

#300

· gadievron opened

on Aug 20, 2026

Container-literal dispatch loses call edges in every language parser (Go, JS/TS, PHP, Ruby, Rust, Swift, Zig — after Python #295 and C #298)

#299

· gadievron opened

on Aug 20, 2026

C call graph records no edge for function-pointer dispatch, and the documented compensation is opt-in and largely unread

#298

· gadievron opened

on Aug 20, 2026

`enhance` and `dynamic_test` checkpoints are permanent and un-gated by backend identity, so a model swap silently adopts the previous model's classifications

#297

· gadievron opened

on Aug 20, 2026

Conclusive-exploit-path guards test a free-text explanation string, so an incomplete verification can read as conclusive

#296

· gadievron opened

on Aug 20, 2026

Python call graph records no edge for a function referenced as a dict value or list element, so dispatch-table targets are pruned with their dispatcher kept

#295

· gadievron opened

on Aug 20, 2026

Navigation Menu

Uh oh!

All issues

Recovery paths had zero activations in a 9.85h run; two of the five inspected since were both defective, and their tests use shapes production never writes

SARIF output has no `invocations` block, so a degraded scan uploads indistinguishably from a clean one

Absent application context is reported as `application_type: "web_app"` rather than unknown

Shared venv `.pth` is re-pointed per clone, and child interpreters spawned mid-scan resolve through it (residual of #62)

Stage 2 only examines Stage-1 positives (4.1% of units) and can only downgrade, but no output states the denominator or the direction

Prune telemetry classifies 58% of prunes as missing-edge "orphan" candidates, then drops the counts before the deliverable

Standalone `openant verify` step report omits `needs_review` and `error_count` that the pipeline records

Container-literal dispatch loses call edges in every language parser (Go, JS/TS, PHP, Ruby, Rust, Swift, Zig — after Python #295 and C #298)

C call graph records no edge for function-pointer dispatch, and the documented compensation is opt-in and largely unread

`enhance` and `dynamic_test` checkpoints are permanent and un-gated by backend identity, so a model swap silently adopts the previous model's classifications

Conclusive-exploit-path guards test a free-text explanation string, so an incomplete verification can read as conclusive

Python call graph records no edge for a function referenced as a dict value or list element, so dispatch-table targets are pruned with their dispatcher kept

Uh oh!

Search results