fix(heldout): declare decision-latency repetitions per context

[seonghobae]

Stack

Draft successor of #1095 (74c27e7e). Parents #1093, #1092, #1091, #1090, #1074, and psychometric KPI owner #1067 stay open. This PR does not close #1095.

Exact head: 8ccf011dbf74f7140dd29fc2661234117c7cb36b.

Problem

Paired held-out ranking timings hid LATENCY_REPETITIONS = 200, so buyer-facing decision-latency p50/p95 used a repository-selected Monte Carlo sample size.

Change

• Require declared repetitions_per_context on _measure_paired_latency. Missing, boolean, or non-positive values fail closed.
• The harness run still writes 200 as this run's choice and records latency_repetitions_per_context.
• ADR 0047 is Proposed. Production route/conduct defaults stay locked.

Verification

Local smallest tests:

• 56 passed in 18.66s (tests/test_psychometric_routing.py, tests/test_psychometric_benchmark_boundaries.py, tests/test_planning_adr_identifiers.py)
• interrogate on the held-out script: 100%

This is declaration-contract evidence, not buyer-held-out accuracy, p95 latency, or protected merge.

Remaining no-heuristics work

Other harness sample sizes remain later slices. No production policy or release is authorized. Parent #1067 still needs independent review.

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[seonghobae]

Normal integration checkpoint b8a7b3b: retains both predecessor8ccf011dbf74f7140dd29fc2661234117c7cb36b and repaired parent318370e84aa4a2a2d280493057a6b19b23484a7e. Only the gap-baseline insertion conflicted; both entries were preserved. Required latency repetitions remain in place, while censored no-detection semantics and parent static-analysis repairs are retained. Integrated boundary and ADR identifier tests:23 passed in33.65s, exit0. This is local focused evidence, not protected-main merge, current-head hosted success, release or measured buyer latency. Draft/parent-first review boundaries remain unchanged.