Measurement specs write to a pre-archive change dir, recreating a phantom active change

[robertcsakany]

Symptom

After verify-subagent-pull-under-load was archived (#548), re-running its measurement specs recreated openspec/changes/verify-subagent-pull-under-load/ containing only a partial measurements.json. That directory looks like an active OpenSpec change to tooling that enumerates openspec/changes/*, but it is a leftover artifact of an archived one.

Cause

Both specs hardcode the pre-archive path:

• tests/e2e/subagent-pull-measurements.spec.ts:46
• tests/e2e/subagent-pull-under-load.spec.ts:68

const EVIDENCE_PATH = "openspec/changes/verify-subagent-pull-under-load/measurements.json";

record() calls mkdirSync(dirname(EVIDENCE_PATH), { recursive: true }), so the directory is created rather than failing loudly once the change moves to openspec/changes/archive/<date>-<name>/.

Consequences

1. A phantom active-change directory reappears on every re-measure.
2. Worse for evidence integrity: the write silently goes somewhere other than the archived heap-evidence.md's source of truth, so a re-run appears to leave the archived numbers "unchanged" while actually recording different ones elsewhere. That happened during test(verify-subagent-pull-under-load): verify the four scenarios the payload change shipped unverified #548 — the final post-fix run's numbers landed in the stray path and were never transcribed. The recorded verdict (~36×) is unaffected in direction, but the failure mode is exactly the staleness the evidence file exists to prevent.

Suggested fix

Resolve the change dir at runtime instead of hardcoding it — prefer openspec/changes/archive/*-verify-subagent-pull-under-load/ when the active dir is absent, and fail (rather than mkdirSync) when neither exists, so a misdirected write is loud instead of silent.

Low severity: both specs are opt-in (PI_SYNTH_AGENT_TICKS=1, and PI_E2E_MEASURE=1 for the measurement file), so neither runs in normal CI.

Found while shipping #548.