perf(policy): investigate moving stop-condition evaluation entirely post-write

[windoliver]

Summary

The current stop-condition evaluation runs in two places:

1. Pre-write inside PolicyEnforcer.enforce() (src/core/policy-enforcer.ts:267) — runs inside the write mutex, costing one store.list() plus per-root store.thread() calls per contribution.
2. Post-write recheck in contributeOperation (src/core/operations/contribute.ts:584) — runs outside the mutex to recover the threshold-crossing case (the Nth review satisfying quorum) that the pre-write check misses.

The post-write recheck exists because the pre-write check has the threshold-crossing blind spot. So the pre-write check never detects threshold crossings, and the post-write recheck always has to run anyway. The pre-write check is doing duplicated work.

Hypothesis

If we delete the pre-write call and rely solely on the post-write recheck, we:

• Halve the stop-condition cost per contribution (one scan instead of two)
• Move all the O(n) work outside the write mutex (no contention)
• Lose nothing semantically, since the post-write path was already designed to be the source of truth for threshold crossings

The risk: are there any consumers that depend on policyResult.stopResult being populated before the write? If the PolicyEnforcer is called outside contributeOperation (lifecycle paths, grove_check_stop tool), they may need their own evaluation.

Investigation steps

1. Audit all callers of PolicyEnforcer.enforce() to see whether they read stopResult.
2. Check whether any pre-write rejection logic depends on stop conditions being met (the comment at policy-enforcer.ts:255-258 suggests no).
3. Benchmark: measure throughput of contributeOperation with and without the pre-write call against a 10K-contribution store.
4. If safe: delete the pre-write block and the skipStopConditions option (added by Issue 13A in the protocol: plan/message operations bypass policy enforcement and topology routing #228 PR), since plans/messages would no longer pay the cost in the first place.

Context

Surfaced during the #228 review (Issue 13D in the locked-in plan, deferred as a separate investigation). Issue 13A landed in the #228 PR as a tactical fix (skipStopConditions for plan + ephemeral kinds), which is a good targeted patch but not the structural fix.

References

• src/core/policy-enforcer.ts:267 — pre-write stop-condition block
• src/core/policy-enforcer.ts:255-258 — comment acknowledging the threshold-crossing limitation
• src/core/operations/contribute.ts:584 — post-write recheck
• protocol: plan/message operations bypass policy enforcement and topology routing #228 — parent review