[rmcdaniel]

Main's Shared Server fixtures fail schedule-policies (or schedule-catalog) intermittently, at a different step each time:

Run	Difference
main 29fb77e3 (job 119962357942)	step 47, the second POST schedules/quota/trigger, returned 404 instead of 200
#667 b2999cac9	the first policy-skip trigger returned skipped instead of triggered
#667, earlier head	both DELETE schedules/policy-skip steps returned 503

Cause. These differences come from the published PHP reference, not native. When native differs, the step log first shows {"outcome":"pass","target":"php"} and …"embedded"}; none of these failures has those lines.

A local reproduction of the CI step shows the mechanism:

• PHP on SQLite can report database is locked for a schedule write that had already committed. The started run's task is queued after the trigger's transaction.
• The probe sent that write again, so it applied twice:
  • a retried first trigger read as skipped, because its own first run was still running;
  • a retried quota trigger used up the second action, so the next trigger found the schedule deleted.

Change.

1. Safe write retries. The probe sends a write again only when its schedule reads exactly as it did before the write, which shows the write rolled back. Otherwise it records the failure, and the recorded lock replays the whole sequence in a new namespace, now up to five times. Reads are still retried. Native never reports this lock, so it never retries or replays.
2. Readable failures. A status difference now names each differing step with its response, and an outcome difference shows the trigger responses with their reasons.

Evidence, local runs of the CI step:

Probe	Result
Main's probe	PHP failed schedule-catalog after three replays
Replay-only (no write retries)	PHP failed both fixtures after five replays
Replay-only, maintenance loop at 1 CPU	Worse; contention is not CPU starvation
This head	PHP and embedded pass. PHP's schedule-policies took five safe retries and one replay; schedule-catalog passed first time. Native's only difference is the known frozen signals-before-worker; it matched both schedule fixtures on attempt 1 with no retries.

The schedule contract tests pass (20/20 and 18/18), including the new message and attempt-bound cases.

[rmcdaniel]

Riding merge train #675 with #667, #669, #670 and #674 on main 17dc12cd, so they share one CI run. This PR closes with a link once the train merges.

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud