plat 075

← Pulse platform issue index

PLAT-075 — auto-evaluation starts before its target execution is finalized

Coordination	Value
Assigned agent	Codex
Ticket state	implemented — runtime reverify pending
Last synchronized	2026-08-10

• Priority: P1 — evaluators are forced to grade an execution whose authoritative interval is still open
• Owner: batch execution / auto-evaluation boundary
• Finding: LinkedIn 9495aef3dab65c42

Evidence and root cause

The full workflow-local finding says evaluators observed run_metadata.json with status=running and no completion time. The retained run later records started_at=2026-08-06T17:01:54Z and completed_at=2026-08-06T17:27:52Z; evaluation output explicitly says it had to substitute terminal timing because the metadata was unfinished.

This was deterministic ordering, not a scheduler race. In controller_batch_execution.go, a successful group called MaybeRunAutoEvaluation and only called finalizeRunMetadata after evaluation and its token persistence completed. Every auto-evaluator therefore saw the target execution as running by construction.

Fix

Finalize the target execution immediately after its execution steps and their own persistence complete, before starting the separate evaluation run. Post-evaluation persistence errors remain evaluation diagnostics and no longer rewrite the already-completed target execution's status.

TestSuccessfulTargetRunIsFinalizedBeforeAutoEvaluation pins this ordering.

Verification

• Focused regression tests: pass.
• go build ./...: pass.
• Required broad suite: exactly 22 known baseline failures; no additional failures.
• Runtime reverify: pending server restart and one evaluation-producing run.

Ready to close after runtime verification: 9495aef3dab65c42.