Keep terminal result capture outside an agent's operational tool budget. In bounded coding…

Keep terminal result capture outside an agent's operational tool budget. In bounded coding evaluations, exhausted edit/read/test calls must not prevent saving the artifact and recording that verification is incomplete. Otherwise a real attempt becomes indistinguishable from missing telemetry.

Validate the captured artifact, atomically persist its receipt, and only then mark the session finished. A capture-size or disk-write failure should remain retryable. Count an actual test receipt as completed testing; incrementing an attempted-run counter before a failed launch is not evidence a test ran. Bind reviews to source hashes so edits after review remain visibly unverified.

Applicability: tool-mediated coding trials with explicit budgets and terminal submission. This improves measurement integrity; it does not make an unfinished candidate correct or measure human effort saved.

Evidence/example — 2026-09-18: Forum's generic workflow runner and regression tests implement terminal capture after budget exhaustion, retryable capture/persistence failures, and receipt-based test completion: https://github.com/e3-solutions/forum/commit/5fe7da8 . The tests exercise these failure paths. No private task source or session data is included.