@clawsweeper review

🦞🧹
ClawSweeper re-review requested.

I asked ClawSweeper to review this item again.
Action: item re-review queued (workflow sweep.yml, event repository_dispatch).
Result: the existing ClawSweeper review comment will be edited in place when the review finishes.

Re-review progress:

• State: Complete
• Detail: The targeted re-review published an exact durable review comment and queued an executing target-wide serialized router scan.
• Run: https://github.com/openclaw/clawsweeper/actions/runs/29149870247
• Updated: 2026-07-11T10:52:06.348Z

Proof update for head 2d43edb. This also addresses Copilot's policy-name comment.

• Frozen replay input: 1,780 records, 345,642 bytes, SHA-256 b1e4aa33180b7c177293eb9ed16b45e24e026d259600fba2b1b67b931b904f0b.
• Current Adaptive simulated 694 refreshes versus 1,383 for fixed 5m, or 49.8% fewer. Its p50/p95 menu staleness was 142s/1,093s versus 150s/281s.
• Exact-head validation: 83/83 focused tests, 604 selections across 51/51 full-suite groups, make check, and an ad-hoc app bundle with strict codesign verification.
• The frozen trace contains zero timer-advance evaluations. This proof does not claim accepted/rejected live-path coverage.
• The raw trace remains local. The image below contains aggregate CLI output only.

Exact-head replay output:

@clawsweeper re-review

@clawsweeper What's your review status? Do you need to restart the review process?

@clawsweeper review

Codex review: needs maintainer review before merge. Reviewed July 11, 2026, 11:33 PM ET / July 12, 2026, 03:33 UTC.

Summary
The branch adds an offline JSONL adaptive-refresh replay CLI and analysis kit, extracts the production decision table into a shared package-internal core, and adds focused tests and decision-record documentation.

Reproducibility: not applicable. This PR adds an offline developer analysis capability rather than fixing a reproducible current-main defect. Its behavior is demonstrated through exact-head CLI runs and focused deterministic tests.

Review metrics: 3 noteworthy metrics.

• Review surface: 28 files, +3,195/-146. The substantial change is mostly isolated in new replay targets and tests, with a narrow production policy adapter extraction.
• Focused test files: 10 changed. Coverage spans CLI parsing, trace parsing, shared policy boundaries, adapters, replay math, segmentation, activity policy, and schedule auditing.
• Policy ownership: 2 consumers, 1 decision table. The application and replay tooling now share one package-internal adaptive policy implementation.

Merge readiness
Overall: 🦞 diamond lobster
Proof: 🦞 diamond lobster
Patch quality: 🦞 diamond lobster
Result: ready for maintainer review.

Overall follows the weaker of proof and patch quality, so missing proof can cap an otherwise strong patch.

Risk before merge

• [P1] One exact-head x64 Linux test job is red while the matching arm64 Linux job passes; required checks should be rerun or diagnosed before merge, although the available evidence does not establish a patch defect.

Maintainer options:

1. Decide the mitigation before merge
   Ship the maintainer-sponsored offline replay tool with its single canonical policy core after the remaining exact-head test gate is green or confirmed unrelated, while keeping runtime trace collection and production policy changes outside this PR.
2. Pause or close
   Do not merge this PR until maintainers decide whether the risk is worth taking.

Next step before merge

• [P2] No repair task remains; required checks and ordinary maintainer merge controls should gate the exact sponsored head.

Security
Cleared: No concrete security or supply-chain concern was found: the patch adds no dependency, network, provider, credential, installer, downloaded-code, or runtime trace-collection path.

Review update for exact head cd96204ac197ec0c7ce48e615a608358c01ba3fb:

• Rebased onto current main; GitHub reports the PR as mergeable.
• Split runtime collection out of this PR. The current diff has no Sources/CodexBar changes, app hooks, recorder, trace writer, transcript scanning, provider calls, or storage lifecycle.
• Reduced the review surface from 32 files and +4,147/-55 to 21 files and +2,840/-7. Tests account for 1,224 additions.
• Kept AdaptiveReplayKit package-internal. The only new product is AdaptiveReplayCLI.
• Fixed and resolved both Copilot CRLF threads with Character.isNewline plus strict and tolerant multi-record regressions.

Exact-head checks passed: make check, swift package describe, swift build --product AdaptiveReplayCLI, and an author-run CLI replay over the hash-pinned local trace. The previous pushed revision, with identical source and test code, passed the full 610 selections in 51/51 groups; upstream CI is running the exact head.

Trace: 1,780 records, SHA-256 b1e4aa33180b7c177293eb9ed16b45e24e026d259600fba2b1b67b931b904f0b.

policy              refreshes  per24h  sim advances  active >5m  staleness p50/p95  constrained ok
adaptive            694        143.47  53            4/145       142s / 1093s       yes
adaptive-activity   696        143.88  53            0/145       139s / 1093s       yes
fixed-5m            1383       285.90  0             0/99        150s / 281s        NO (157)
activity telemetry: 462/733 sampled (63%), 185/462 active (40%)
recorded schedule: 46 advances, 0/0 evaluations, all mismatch counts 0
segmentation: 28 segments, 26.10h excluded, 300s grace

Runtime proof is no longer applicable because runtime collection is no longer in the diff. This remains single-machine, author-run evidence; it makes no battery, user-time, request-count, cost, or default-policy claim.

@clawsweeper re-review

🦞🧹
ClawSweeper re-review requested.

I asked ClawSweeper to review this item again.
Action: item re-review queued (workflow sweep.yml, event repository_dispatch).
Result: the existing ClawSweeper review comment will be edited in place when the review finishes.

Re-review progress:

• State: Complete
• Detail: The targeted re-review published an exact durable review comment and queued an executing target-wide serialized router scan.
• Run: https://github.com/openclaw/clawsweeper/actions/runs/29151344407
• Updated: 2026-07-11T11:45:24.141Z

🦞🧹
ClawSweeper re-review requested.

I asked ClawSweeper to review this item again.
Action: item re-review queued (workflow sweep.yml, event repository_dispatch).
Result: the existing ClawSweeper review comment will be edited in place when the review finishes.

Re-review progress:

• State: Complete
• Detail: The targeted re-review published an exact durable review comment and queued an executing target-wide serialized router scan.
• Run: https://github.com/openclaw/clawsweeper/actions/runs/29152589644
• Updated: 2026-07-11T12:32:58.571Z

Addressed fixed-interval validation in 41c6c0dc.

--policy now accepts only the seven names listed by --help. Generic fixed-<N>m parsing is gone, so fixed-0m, fixed--1m, fixed-3m, and fixed-9223372036854775807m are rejected before trace reading or policy construction. Each real CLI invocation exits 1 and prints the accepted names. FixedIntervalPolicy also converts to TimeInterval before multiplying, removing the integer-overflow path.

Validation on this head:

• 3 CLI argument tests covering 12 cases, including all seven accepted names and the four invalid interval classes;
• 67 replay/core tests across 8 suites and 4 app-adapter tests;
• swift build --product AdaptiveReplayCLI, make check, and git diff --check;
• the hash-pinned 1,780-record replay still produces the same metrics.

changes, lint, and GitGuardian are green on the new head. The Linux builds and macOS test shards are running. The mechanical finding is addressed; owner sign-off for the CLI remains the product decision.

@clawsweeper re-review

🦞🧹
ClawSweeper re-review requested.

I asked ClawSweeper to review this item again.
Action: item re-review queued (workflow sweep.yml, event repository_dispatch).
Result: the existing ClawSweeper review comment will be edited in place when the review finishes.

Re-review progress:

• State: Complete
• Detail: The targeted re-review published an exact durable review comment and queued an executing target-wide serialized router scan.
• Run: https://github.com/openclaw/clawsweeper/actions/runs/29153392775
• Updated: 2026-07-11T12:59:33.546Z

Maintainer verification at exact head b231503a53020acad599672717ef80e0207544fd:

• make test: 624 selections across 52 groups passed on the first attempt, with zero retries, recoveries, or timeouts.
• Focused adaptive replay, replay segmentation, and recorded schedule audit suites passed.
• make check: passed with zero SwiftLint violations across 1,402 files.
• Maintainer autoreview: clean after two fix/review cycles (0.90 confidence).
• Real CLI smoke produced expected table and JSON metrics; malformed JSONL exited 1 and identified line 2 plus its offending content.

The contributor's private trace was not copied or published. Provider/account live testing is intentionally out of scope for this offline, caller-supplied trace tool.

Hosted exact-head proof: lint, changes, Linux arm64, and Linux x64 are green. The first x64 attempt had one unrelated Claude planner availability race; its clean rerun passed the build, full Linux tests, and CLI smoke.