feat(scheduler): add adaptive scheduled hot-review allocation

[brokemac79]

Important

DO NOT MERGE without Martin's explicit approval. This PR is intentionally draft and will receive an independent maintainer review before any merge decision. It does not deploy or activate adaptive production behavior.

Summary

Delivers the complete adaptive scheduled hot-review package in one PR:

• deterministic capacity-aware per-repository allocation;
• real scheduler observation, shadow, canary, and full-mode integration;
• durable signed observations, decisions, readiness facts, and cursor reservations;
• offline replay and bounded public visibility;
• explicit activation approvals, cohort controls, readiness gates, and kill switch;
• operator documentation, tests, and an exact-head Docker-backed proof harness.

The merged behavior remains unchanged by default: legacy mode plus the kill switch are configured, hot fanout remains on the 20-minute containment cadence, scheduled admission remains 300/hour with burst 30, and normal fanout is unchanged.

Problem

The contained scheduler gives every selected repository the same generic candidate allowance despite large differences in eligible demand, novelty, review cost, observation health, and time since service. That wastes scarce review capacity and can starve quieter repositories. A safe replacement also needs durable evidence, replayability, quota/circuit integration, explicit rollout controls, and exact rollback semantics; an allocator alone is not an operable production design.

Implementation

Allocation and pressure signals

• Service capacity is the minimum of available review-queue candidate capacity, the global scheduled-feed token balance, and the hot-lane token balance.
• The offer budget is ceil(1.5 * service capacity), capped at 30; at most 20 repositories are selected.
• A repository receives at most 25% of the offer budget, with an absolute cap of 10 candidates.
• Deterministic rounds cover 24-hour fairness, source novelty, at most two unknown/stale probes, ordinary observed demand, then residual expansion.
• Unknown probes have a separate cursor, ordinary demand retains a reserved slot in small cycles, and priority outliers cannot reset ordinary fairness progress.
• Queue unavailability, zero capacity, scheduled throttling, and fix: collapse exact publication on GitHub quota #1102 credential circuit pressure constrain discretionary allocation. Publication depth is deliberately not a repository-ranking signal.
• An observation outage has a bounded five-repository fallback, but fallback cycles cannot count toward activation readiness.
• Signals are repository-level operational facts only; the policy uses no personal, opaque, or learned attributes.

Observation, durability, and replay

• Scheduled planning publishes versioned, signed planner observations, including zero-demand, dedupe, shedding, no-op/cache, execution-cost, retry, and outcome counters.
• The Worker stores bounded observations and decisions in SQLite-backed durable state and exposes a sanitized public snapshot through the existing read-only queue status surface.
• Public decisions are reconstructed from a strict nested allowlist before storage and again when read, so unknown workflow-supplied fields cannot enter the unauthenticated status projection.
• Observation snapshots retain the latest row per repository, lane, and policy version so policy rollovers cannot hide current evidence.
• Readiness is scoped to the current policy version and counts only dispatched cycles backed by an allocated adaptive proposal; unavailable-observation fallback and queue-unavailable decisions remain visible but do not advance rollout.
• scripts/evaluate-hot-allocation.mjs replays sanitized fixtures deterministically without GitHub, queue, or production mutation.

Execution safety and cursor settlement

• Active-mode legacy, adaptive, and probe cursor movement uses a signed two-phase reservation/commit protocol with a one-hour reservation expiry and a 24-hour/80-receipt replay ledger.
• Reservations leave all covered cursors unchanged until a matching commit; commits advance the active batch atomically and return a durable idempotent receipt after lost responses or Worker restarts.
• Committed reservation identities are one-shot. A zero-success dispatch aborts and releases its reservation unchanged.
• After any GitHub dispatch succeeds, the whole reserved cycle is committed. This intentionally provides at-most-once cursor settlement: an undispatched repository in a partially failed batch is not replayed immediately, avoiding duplicate Actions/quota consumption where repository_dispatch has no rollback or idempotency key. It remains due in inventory and is reconsidered by demand, novelty, probes, or the 24-hour fairness floor.
• Shadow mode reserves adaptive and probe cursor progress together while the authoritative legacy cursor stays outside the comparison reservation and persists independently. If shadow telemetry or reservation fails, legacy dispatch and cursor progress remain fail-open.

Rollout and rollback

• Modes are legacy, shadow, canary, and full; deterministic 10/50/100 cohorts and explicit canary repositories are supported.
• Activation requires explicit approval plus current-policy readiness: seven days/21 qualifying shadow cycles before canary, then at least 24 hours/3 qualifying cycles at each bounded rollout stage.
• Defaults are legacy plus kill switch on. Turning on the kill switch immediately makes legacy selection authoritative without changing cadence or admission limits; rollback does not require clearing durable observations or cursors.
• This PR supplies activation configuration and documentation but does not change live variables, schedules, admission, gates, workflows, queues, or production data.

Validation

Reviewed range:

• base: c145b4599285969b43e40d0816071893d5090501
• head: ceb36659a534ad77f88f0573b68e4d7f3f602563

Passed on the current head:

pnpm run check:static
pnpm run build:all
pnpm run lint
node --test [adaptive allocator/runtime/control-plane/fanout/dashboard/workflow focused suites]
git diff --check origin/main...HEAD
docker run --rm -v "$PWD:/repo" -w /repo rhysd/actionlint:1.7.12 -color=false -shellcheck= -ignore "unexpected key.*queue.*concurrency" .github/workflows/sweep.yml
codex review --base origin/main
pnpm run codex:local:check
pnpm run review -- --local-range --target-repo openclaw/clawsweeper --base origin/main --body-file <this-body>
docker build [automerge E2E base and application images]

The checked-in proof suite reports 67/67 focused tests passing: 58 allocator/runtime/control-plane/fanout tests, six Worker/durable-boundary tests, and three workflow/cadence tests.

After rebasing onto #1115 and #1116, the overlap checks passed: canonical GitHub API tests (3/3), adaptive control-plane tests (3/3), selected Worker/Durable Object tests (3/3), static checks, build, lint, and the complete exact-head behavior proof. The sole rebase conflict was import-only in dashboard/exact-review-queue.ts; both the canonical GitHub API imports and adaptive store imports were retained.

Broad-check classification:

• Exact-head native Windows pnpm run check passed static checks, formatting, build, and lint, then reached repository test surfaces that require Bash/WSL, POSIX locking/symlinks, and Unix temporary-file semantics. The observed workflow failures were /bin/bash-unavailable errors from the Docker Desktop WSL distribution; those unrelated platform failures are not treated as a request to broaden this PR.
• Pinned actionlint parsed the changed workflow after disabling repository-wide baseline shellcheck notices and ignoring the two pre-existing concurrency.queue schema gaps that reproduce on origin/main; the changed completion payload is covered by the focused workflow parser test.
• In a jq-equipped Playwright Noble Crabbox container before the final rebase, the full check reduced to three review-blob-hydration failures. The identical isolated three failures reproduced on then-current main 51ac499c741b... in the same image; the changed scheduler/Worker paths were not involved. The final exact-head focused container proof passes, and hosted CI on the rebased head remains authoritative for the repository's supported environment.

Real Behavior Proof

Claim: at the exact committed head, the complete default-disabled package authenticates and persists telemetry, survives Worker restarts, fences and atomically settles active-mode cursor batches while preserving independent shadow legacy progress, rejects reservation replay, runs the actual fanout command through zero-success abort and successful recovery, deterministically replays allocation fixtures, and exposes only bounded sanitized status.

Exercised surface: a clean checkout of the exact head inside Docker-backed Crabbox; pinned Wrangler 4.107.0; SQLite Durable Object; real signed/unsigned HTTP routes; Worker restart and persistent state; cursor reserve/commit/abort; the built dist/repair/target-fanout.js command crossing loopback TLS into Wrangler; offline replay; 67 focused tests, including a signed current-policy hot-lane allocator view beyond the 100-partition public cap, nested unknown-field decision sanitization, malformed inactive rollback controls, empty-plan dependency isolation, and repeated shadow comparison-commit failure. Production activation was disconnected.

Scenario/fixture: sanitized example/alpha and example/beta repository observations, synthetic run/reservation IDs, and a disconnected deterministic gh fixture that recorded dispatch intent and injected one zero-success failure. No GitHub API, production repository dispatch, queue, gate, schedule, comment, label, workflow, or dashboard mutation occurred.

Command/environment: Crabbox provider local-container, image mcr.microsoft.com/playwright:v1.60.0-noble, checked-in docs/proof/adaptive-hot-review/run-proof.sh. After two Crabbox rsync handshakes failed before copying any bytes, the clean exact-head checkout was seeded into the same provider from a verified read-only Git bundle and the proof ran through Crabbox with --no-sync; the script performed its own frozen install/build without Actions hydration.

Observed result: RESULT: PASS.

exact head: ceb36659a534ad77f88f0573b68e4d7f3f602563
auth boundary: HTTP 401
allocator view: signed current-policy hot lane returned 1 observation
telemetry persistence: 1 observation, 1 dispatched decision after restart
cursor fencing: unchanged before commit; 3 cursors advanced atomically
commit retry after second restart: identical durable receipt
committed reservation identity reuse: HTTP 409
direct write during reservation: HTTP 409
dispatch failure recovery: aborted unchanged; replacement reservation acquired
actual target-fanout: injected zero-dispatch failure aborted; recovery dispatched 2 and committed
shadow commit failure: legacy cursor advanced; next cycle selected the next repository
kill switch rollback: malformed inactive controls bypassed; legacy dispatch persisted
public snapshot: bounded and sanitized

Trace/artifacts: lease cbx_0d5a25303ead (tidal-crayfish), provider local-container, image above, exit 0. The local proof package contains proof-summary.json, runtime-transcript.md, focused TAP, replay JSON, and Wrangler logs; it contains synthetic values only and is not published as a repository artifact. The lease was stopped after artifacts were copied.

Limits: this proves the delivered activation machinery while deliberately leaving production activation disconnected. It does not demonstrate live GitHub repository dispatch, real fleet yield, live quota consumption, or a production canary. Those are operational rollout observations after a separately approved activation, not development-time proof authorization.

Review closeout

Accepted findings fixed during iterative dirty and committed Codex reviews include fairness/cursor starvation, global-token accounting, probe isolation, shadow cursor fencing, one-shot reservations, dispatch-abort recovery, partial-dispatch duplicate prevention, direct-write exclusion during active reservations, hot-only adaptive configuration parsing, canonical retention wiring, kill-switch bypass of malformed inactive controls, independent legacy cursor persistence across shadow comparison failures, fail-open empty-plan observation without default-branch or queue-capability hard dependencies, canonical operator-documentation links, exact snapshot parsing, policy-version observation isolation/order, excluding observation fallback from readiness, deterministically bounding the public observation snapshot to its advertised 100 partitions, separating that public cap from a signed current-policy hot-lane allocator view covering the eligible fleet, strict public decision projection, aligning the documented 25% cap with the over-offered candidate budget, running the actual fanout command through its Worker settlement boundary, preserving measured review runtime end to end instead of replacing it with lease elapsed time, and reserving the final constrained-cycle slot for ordinary demand before overdue, novelty, or probe rounds can exhaust it. The final committed review reports no discrete actionable correctness issues.

The full local ClawSweeper range loop found and drove fixes for fallback readiness, the public observation cap, strict decision sanitization, and the documentation mismatch. The final current-base artifact reviewed c145b4599285969b43e40d0816071893d5090501...ceb36659a534ad77f88f0573b68e4d7f3f602563, reported real_behavior_proof_status: sufficient, zero review comments, and Overall correctness: patch is correct. Its B/A/B overall/proof/patch rating reflects the broad central-automation scope and the still-required maintainer acceptance of the documented at-most-once settlement tradeoff, not an actionable correctness finding. Hosted @clawsweeper re-review will be requested only after the exact final body and head are on the draft PR.

Bay impact

Observer-only. The existing queue/status projection can display the bounded adaptive snapshot, but this PR adds no Bay controls, actions, activation path, or mutation surface.

Risks and maintainer decisions

• The diff is broad because safe activation, rollback, durability, proof, and observability are intentionally delivered with the allocator in this one PR.
• The default path remains legacy and the kill switch remains on; the primary merge risk is future operator misconfiguration, bounded by explicit approvals/readiness and documented rollback.
• The at-most-once partial-dispatch policy favors avoiding duplicate Actions/quota use over immediately replaying an undispatched cursor position; maintainers should explicitly accept that tradeoff.
• Maintainer/product direction is still required before any shadow activation, on the canary repository set, and before each rollout stage.
• The workflow-file change may require maintainer-owned GitHub OAuth/workflow scope to update the branch.

Related work

• #1086 — quota control
• #1087 — 20-minute hot-fanout containment
• #1088 — durable no-op/self-feedback remediation
• #1102 — quota/circuit/throttle accounting taxonomy
• #1105 — scheduler/automation documentation
• #1108 — operator contract inventory
• #1112 — canonical Bay and handoff telemetry
• #1113 — documentation correction follow-up

[clawsweeper]

🦞👀
ClawSweeper picked this up.

Pull request received. I will update this pull request when review starts.

[brokemac79]

@clawsweeper review

[clawsweeper]

🦞🧹
ClawSweeper re-review requested.

I asked ClawSweeper to review this item again.
Action: item re-review queued (workflow sweep.yml, event repository_dispatch).
Result: when the review finishes, ClawSweeper will create the durable review comment if needed or update the existing comment in place.

Re-review progress:

• State: Waiting
• Detail: A newer exact-review queue owner superseded this run. This run completed as a no-op.
• Run: https://github.com/openclaw/clawsweeper/actions/runs/31424007463
• Updated: 2026-08-10T19:29:32.891Z

[clawsweeper]

Codex review: needs maintainer review before merge. Reviewed August 12, 2026, 7:32 PM ET / 23:32 UTC.

ClawSweeper review

What this changes

This PR adds adaptive capacity allocation for scheduled hot reviews, with durable observations, cursor reservations, staged rollout controls, workflow integration, tests, documentation, and a local Worker proof harness.

Merge readiness

⚠️ Needs maintainer review before merge - 5 items remain

Keep open for explicit maintainer product direction and a current-main rebase. The branch is a substantial default-disabled scheduler feature, not a fix for an established contract; GitHub reports it as dirty, so its exact-head proof does not establish the behavior of a merge result.

Priority: P2
Reviewed head: ceb36659a534ad77f88f0573b68e4d7f3f602563
Owner decision: Required. See Decision needed.

Review scores

Measure	Result	What it means
Overall readiness	🦐 gold shrimp (3/6)	The exact-head proof is strong, but the broad, dirty cross-lane feature still needs a sponsored current-main reconciliation.
Proof confidence	🦞 diamond lobster (5/6)	Sufficient (logs): The PR body records an exact-head Docker-backed Worker, Durable Object, HTTP, cursor, and dispatch-fault proof with an observed passing result; it must be rerun after resolving the dirty merge.
Patch quality	🦐 gold shrimp (3/6)	No actionable review findings were identified.

Verification

Check	Result	Evidence
Real behavior	Verified	Sufficient (logs): The PR body records an exact-head Docker-backed Worker, Durable Object, HTTP, cursor, and dispatch-fault proof with an observed passing result; it must be rerun after resolving the dirty merge.
Evidence reviewed	6 items	Current scheduler contract: Current main uses cursor-selected hot repositories and fixed candidate batches; the adaptive behavior is a new selection mode rather than a repair to an existing failure. Existing safety boundary: Current documentation defines authenticated Durable Object cursors with monotonic revisions and fail-open persistence, which the proposed reservation protocol would materially extend. Main moved beyond the PR base: The PR base is c145b45; current main has since received queue, Worker, workflow, and recovery changes, including the queue seam refactor d8471a6 and later work through 9a25790.
Findings	None	None.
Security	None	None.

How this fits together

Scheduled target fanout selects repositories and sends review work into the durable exact-review queue. This PR inserts an adaptive planning and telemetry layer before that dispatch path while retaining legacy selection by default.

flowchart LR
  A[Scheduled target fanout] --> B[Queue and token facts]
  A --> C[Repository observations]
  B --> D[Adaptive allocation]
  C --> D
  D --> E[Cursor reservation]
  E --> F[Repository review dispatch]
  D --> G[Durable status snapshot]

Loading

Decision needed

Question	Recommendation
Should ClawSweeper adopt this adaptive scheduled-review control plane as core product direction, or keep the current legacy allocator while scheduled-demand controls mature?	Keep the legacy scheduler for now: Do not merge this large control plane until maintainers explicitly prioritize adaptive allocation and its operating model.

Why: This is a new cross-lane scheduling policy with durable state, rollout configuration, and operational consequences; current source does not define it as a broken existing contract, and VISION.md reserves product direction for human judgment.

Before merge

• Resolve merge risk (P1) - The branch is dirty against current main; resolving its Worker, queue, workflow, cursor, and dependency overlaps can change behavior beyond the exact head that was proven.
• Resolve merge risk (P1) - If activated, allocation and cursor-settlement policy can alter review fairness, throughput, and recovery behavior across the scheduled fleet.
• Resolve merge risk (P1) - The added durable control plane and workflow telemetry increase automation and availability surface despite legacy mode remaining the default.
• Complete next step (P2) - A maintainer must choose the adaptive-scheduling direction before a rebased implementation and proof refresh can be evaluated.

Agent review details

Security

None.

Review metrics

Metric	Value	Why it matters
Change breadth	43 files, +9,679 / -73 lines	The PR spans scheduler selection, durable Worker state, workflow behavior, dependency policy, tests, proof tooling, and operator documentation.

Merge-risk options

Maintainer options:

1. Rebase and re-prove the resolved branch (recommended)
   Resolve the dirty merge against current main, then rerun the Worker/fanout proof and review on the exact resulting head before considering activation.
2. Pause pending scheduler direction
   Keep legacy selection authoritative until maintainers explicitly sponsor the persistent adaptive rollout model.

Technical review

Best possible solution:

If maintainers sponsor adaptive scheduling, rebase it onto current main, reduce any unnecessary control-plane scope, and rerun the exact-head runtime proof and review on the resolved branch; otherwise retain the existing legacy scheduler and quota controls.

Do we have a high-confidence way to reproduce the issue?

Not applicable: this PR proposes a new scheduling capability rather than reporting a current-main bug with a failing reproduction.

Is this the best way to solve the issue?

Unclear: the design has strong exact-head proof, but the large deterministic cross-lane control plane needs maintainer sponsorship and a fresh proof after merging current-main changes.

AGENTS.md: found and applied where relevant.

Codex review notes: model internal, reasoning high; reviewed against 9a257905e50b.

Labels

Label justifications:

• P2: This is a bounded scheduler improvement with fleet-wide operational impact but no demonstrated emergency.
• merge-risk: 🚨 compatibility: The PR adds persisted cursor and rollout contracts that must remain compatible with existing scheduler state after reconciliation.
• merge-risk: 🚨 availability: Allocation, reservation, and dispatch changes can reduce or stall scheduled review progression if reconciled incorrectly.
• merge-risk: 🚨 automation: The diff changes the sweep workflow and durable queue integration used by scheduled automation.
• rating: 🦐 gold shrimp: Overall readiness is 🦐 gold shrimp; proof is 🦞 diamond lobster and patch quality is 🦐 gold shrimp.
• status: 👀 ready for maintainer look: ClawSweeper has no concrete contributor-facing blocker left for this PR. Sufficient (logs): The PR body records an exact-head Docker-backed Worker, Durable Object, HTTP, cursor, and dispatch-fault proof with an observed passing result; it must be rerun after resolving the dirty merge.
• proof: sufficient: Contributor real behavior proof is sufficient. The PR body records an exact-head Docker-backed Worker, Durable Object, HTTP, cursor, and dispatch-fault proof with an observed passing result; it must be rerun after resolving the dirty merge.

Evidence

What I checked:

• Current scheduler contract: Current main uses cursor-selected hot repositories and fixed candidate batches; the adaptive behavior is a new selection mode rather than a repair to an existing failure. (src/repair/target-fanout.ts:161, 9a257905e50b)
• Existing safety boundary: Current documentation defines authenticated Durable Object cursors with monotonic revisions and fail-open persistence, which the proposed reservation protocol would materially extend. (docs/scheduler.md:238, 9a257905e50b)
• Main moved beyond the PR base: The PR base is c145b45; current main has since received queue, Worker, workflow, and recovery changes, including the queue seam refactor d8471a6 and later work through 9a25790. (dashboard/exact-review-queue.ts:1, d8471a64e56b)
• Merge state and proof boundary: GitHub context reports head ceb3665 as dirty against main; the supplied Docker-backed proof is substantive for that exact head but cannot prove the eventual resolved merge result. (ceb36659a534)
• Feature provenance: The current target-fanout implementation dates to the scheduler boundary introduced in commit 621b297; later mainline queue refactors make its current owners the appropriate reconciliation reviewers. (src/repair/target-fanout.ts:107, 621b297eadd5)
• Checkout limitation: Some PR-only blobs are promisor objects unavailable in this read-only checkout because the configured remote could not resolve; this review therefore relies on supplied PR diffs/body plus inspected current-main source and history. (ceb36659a534)

Likely related people:

• steipete: Introduced the current target-fanout boundary and recently carried queue and dashboard refactors that the branch must reconcile. (role: current scheduler and queue-area contributor; confidence: high; commits: 621b297eadd5, d8471a64e56b; files: src/repair/target-fanout.ts, dashboard/exact-review-queue.ts, dashboard/worker.ts)
• brokemac79: Merged history in the supplied context connects this author to the 300/hour scheduled-admission and credential-circuit controls consumed by the proposed allocator. (role: recent scheduled-admission and quota contributor; confidence: medium; commits: 13f8ffdea32a, cd4892391195; files: .github/workflows/sweep.yml, dashboard/exact-review-queue.ts)

Rank-up moves

Optional improvements that raise the rating; they are not merge blockers.

• Obtain a maintainer decision on adopting adaptive scheduling as a core capability.
• Rebase onto current main and rerun the exact-head runtime proof and fresh review.

Rating scale

Score	Internal tier	Crab rank	Meaning
6/6	S	🦀 challenger crab	Exceptional readiness
5/6	A	🦞 diamond lobster	Very strong readiness
4/6	B	🐚 platinum hermit	Good normal PR; ordinary maintainer review
3/6	C	🦐 gold shrimp	Useful, but confidence is limited
2/6	D	🦪 silver shellfish	Proof or implementation needs work
1/6	F	🧂 unranked krab	Not merge-ready
N/A	NA	🌊 off-meta tidepool	Rating does not apply

Overall follows the weaker of proof and patch quality.
Shiny media proof means a screenshot, video, or linked artifact directly shows the changed behavior. Runtime, network, CSP, and security claims still need visible diagnostics.

Workflow

• ClawSweeper keeps one durable marker-backed review comment per issue or PR.
• Re-runs edit this comment so the latest verdict, findings, and automation markers stay together instead of adding duplicate bot comments.
• A fresh review can be triggered by eligible @clawsweeper re-review comments, exact-item GitHub events, scheduled/background review runs, or manual workflow dispatch.
• PR/issue authors and users with repository write access can comment @clawsweeper re-review or @clawsweeper re-run on an open PR or issue to request a fresh review only.
• Maintainers can also comment @clawsweeper review to request a fresh review only.
• Fresh-review commands do not start repair, autofix, rebase, CI repair, or automerge.
• Maintainer-only repair and merge flows require explicit commands such as @clawsweeper autofix, @clawsweeper automerge, @clawsweeper fix ci, or @clawsweeper address review.
• Maintainers can comment @clawsweeper explain to ask for more context, or @clawsweeper stop to stop active automation.

History

Review history (8 earlier review cycles)

• reviewed 2026-08-10T19:23:49.246Z sha 8901a08 :: needs maintainer review before merge. :: none
• reviewed 2026-08-10T19:33:02.380Z sha 8901a08 :: needs maintainer review before merge. :: none
• reviewed 2026-08-10T19:40:17.597Z sha 8901a08 :: needs maintainer review before merge. :: none
• reviewed 2026-08-11T00:24:57.768Z sha 854a29c :: needs maintainer review before merge. :: none
• reviewed 2026-08-11T03:40:52.444Z sha ceb3665 :: needs maintainer review before merge. :: none
• reviewed 2026-08-11T03:47:41.998Z sha ceb3665 :: needs maintainer review before merge. :: none
• reviewed 2026-08-11T12:13:54.916Z sha ceb3665 :: needs maintainer review before merge. :: none
• reviewed 2026-08-12T06:24:36.858Z sha ceb3665 :: needs maintainer review before merge. :: none

[brokemac79]

@clawsweeper re-review

[clawsweeper]

🦞🧹
ClawSweeper re-review requested.

I asked ClawSweeper to review this item again.
Action: item re-review queued (workflow sweep.yml, event repository_dispatch).
Result: when the review finishes, ClawSweeper will create the durable review comment if needed or update the existing comment in place.

Re-review progress:

• State: Complete
• Detail: The durable review result and its route handoff completed.
• Run: https://github.com/openclaw/clawsweeper/actions/runs/31456675772
• Updated: 2026-08-11T03:52:12.844Z

[brokemac79]

Closing by maintainer direction. No current issue signal indicates that the default-disabled adaptive scheduler package remains needed, and this draft is now stale and dirty against main. The branch and PR history are retained for selective recovery if the requirement returns.