[Dong1017]

One explicit additional pair, without clearing the first claim

Refs #248. This software correction implements the approved PR275-only $350 lifetime exception and exactly one additional same-specification 8+8 comparison. It does not itself dispatch or activate the comparison.

• Keep the generic PR cap at $200 and the repository Shanghai-day/pilot ceilings at $1,000. Preserve every historical ledger byte and the original full $162.72 claim with unknown actual cost; no refund, reset, new charge PR or partial release.
• Bind the exception to the reviewed public freeze, exact PR identity, prior claim/checkpoints and one fixed new campaign/execution. The cap context is never decoded from ledger JSON or supplied by a worker, candidate, CLI or environment.
• Check both ceilings before daily-first CAS. Partial/uncertain writes, races, changed identities and any third attempt fail closed. After a successful new admission, total retained liability is $325.44 for PR275 and $816.79 daily with the observed prior carryover; later obligations still count.
• Preserve the historical task/K/verifier/controls/metrics and equal request budgets. Add prospective exception DATA only; the old expired freeze and locator are not renewed. Fresh exact-source hosted qualification, prices, v2 freeze, same-host preflight and anchored admission remain required.

First run 41159550621 remains INCONCLUSIVE: one A2 slot consumed, not proof of a request grant or HTTP delivery; A3 did not start. This patch does not recover its omitted cause or establish billing.

Software verification and provenance

• 655 affected offline tests and 9 subtests passed; 17 inventory checks, Ruff (src tests tools) and diff checks passed. Fake controls cover exact history preservation, default-cap rejection, context/JSON injection, partial CAS, racing second admissions, full additional-pair execution, daily overflow and third-attempt rejection.
• Historical actor inputs/configs, suites and hidden trace DATA still match their frozen bytes. No task, oracle, timing gate, credential binding or workflow change.
• Independent exact-source cost/admission review passed 277 tests plus separate race/full-8+8/old-binding controls; root-bound source/security review passed 50 focused tests. Both found no material blocker. The clean full local aggregate passed on this exact source tree: 3240 passed, 12 skipped, 284 subtests passed; one expected duplicate-ZIP-member warning. Ready run 41182807900 passed all five jobs (Python 3.10/3.11 full checks and installed-package checks, plus model-free task/actor qualification); its test-merge tree equals the reviewed tree. The opened-event Ready run 41182789635 was superseded/cancelled, not counted as PASS.
• Native code review completed with no findings. Separate exact-head cost/admission and root-bound source COMMENT reviews found no blocker; neither is a formal approval.
• Independent original-artifact checks reconstructed 198 actor semantic outputs (77 hidden per arm), 20 benchmark outputs and 201 execution records, plus 792 task-control outcomes and 51 benchmark outputs. Correct, incorrect and unchanged/no-op outcomes remain distinct; runtime denial and descendant cleanup evidence passed. The original bounded ZIPs are retained, with privacy clearance for archival publication; no candidate/model replay was used to replace first-run evidence.
• Automatic Skill Eval remains planned/unexecuted with zero supported comparison cases. These software controls and qualification do not establish live delivery, actual cost, efficacy or final spending admission.
• Published head: 1566c3700079c4eab133058083bdb48c12e88333. Source tree: 371c048c6720d17192775ced6c34707704445329. The API wrapper preserves the local reviewed source; it does not claim authorship of unrelated current-main knowledge changes.

Authorship

Actual implementation and independent review: Codex / gpt-6-astra / xhigh. Software checks and operational admission remain separate; no paid call or ledger write was made while implementing this PR.

{"schema_version":1,"head_sha":"1566c3700079c4eab133058083bdb48c12e88333","human_only":false,"material_ai_authors":[{"tool":"Codex (scoped single additional KV attempt)","model_id":"gpt-6-astra","effort":"xhigh","role":"author"}]}

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-31T08:02:20.661502Z	1566c37	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[github-actions]

Skill evaluation

Automatic execution / cumulative PR budget: {"reason":"Trusted configuration or pricing missing, invalid or expired","remaining_usd":"unknown","reserved_or_unresolved_usd":"unknown","spent_or_legacy_usd":"unknown","state":"blocked"}

Advisory: obtain missing evidence

Retained evidence availability: planned / unexecuted. No approval or merge authorization is issued.

• Retained evidence is planned / unexecuted.
• No supported comparison cases were selected.
• Uncovered/unsupported scope needs a case or explicit reviewer disposition.

PLANNED: execution is not authorized by knowing this hash.
Outcomes: {"fix":0,"inconclusive":0,"missing_pair":0,"regression":0,"unchanged_correct":0,"unchanged_incorrect":0}
Earlier revision results are stale for this plan.

Selected comparisons: 0 · Estimated task calls: 0.
Uncovered/unsupported changes: 16. Gaps never count as passes.

Applicable only while these revisions and inputs match the current PR.
Paid execution requires trusted label authorization and cumulative PR reservation. Unknown accounting blocks new paid calls; planning is not execution.
Pipeline evidence only; offline fixed outputs do not establish skill efficacy.
Evidence and logs · Retention: 30 days.
Outcomes are historical; verifiable only before Unix time 1796025673 while the linked artifact is available. Expired evidence is unverified.

Plan bindings, budget, frozen exam and advisory details

Unchanged-correct is valid neutral evidence, not incremental benefit.
Content/source review and authorship admission remain separate.
Trusted regrading checks retained answers, not model/hardware provenance or a hostile executor; this is not an efficacy certification.

Per-case outcomes (discovery excluded; regressions and gaps first):

[]

Showing at most 20 of 0 pairs; the evidence link contains all outcomes, including omitted regressions/gaps.

Plan SHA-256: b68180668663ca77f01ece8d334a4f07729793d4324465eb865aab44de59dadc
Base: ac4c54bc06d44e11b067f5efb92de93c17dc4312 · Head: 1566c3700079c4eab133058083bdb48c12e88333
Trusted implementation: ac4c54bc06d44e11b067f5efb92de93c17dc4312

Model/settings: {"effort":"not_applicable","name":"claude-opus-5-5"}
Budget: {"comparisons":0,"estimated_calls":0,"limits_per_comparison":{"campaign_seconds":600,"max_task_calls":2,"repeats":1,"timeout_seconds":180,"token_stop_new_trials":200000},"provider_config_sha256":"b28d80f8194fbe0f5a9f01515efb24b218979fd45f702776560da276f4b30d04","reservation":null}

Skill	Comparison	Cases	Estimated calls

Provider/settings: {"client":{"api_version":"2023-06-01","endpoint":"https://yibuapi.com/v1/messages","kind":"anthropic-messages-full-context-v1","max_output_tokens":4096,"max_read_bytes":32768,"max_request_bytes":262144,"max_requests":1,"max_tool_calls":12,"provider":"yibuapi"},"limits":{"token_stop_new_trials":200000},"model":{"effort":"not_applicable","name":"claude-opus-5-5"},"pricing":{"currency":"USD","expires_at":1793271228,"group":"default","max_input_tokens":1000000,"model":"claude-opus-5-5","rate_evidence":"https://github.com/InferMatrixLab/InferRSI/issues/215#issuecomment-6467531341","rates":{"cache_read":"0.4","cache_write":"10","input":"8","output":"40"},"request_fee":"0","unit":"per_million_tokens"},"spend_limit":{"amount":"200","currency":"USD"}}

Uncovered / unsupported paths (16)

• uncovered: "docs/kv-live-launcher.md"
• uncovered: "evidence/kv-live-pilot-20261031/plan/SECOND-ATTEMPT.md"
• uncovered: "evidence/kv-live-pilot-20261031/plan/second-attempt-budget.json"
• uncovered: "evidence/kv-live-pilot-20261031/plan/second-attempt-integrity.json"
• uncovered: "tests/test_evidence_inventory.py"
• uncovered: "tests/test_kv_live_job.py"
• uncovered: "tests/test_kv_live_policy.py"
• uncovered: "tests/test_kv_pilot_admission.py"
• uncovered: "tests/test_skill_eval_budget.py"
• uncovered: "tests/test_skill_eval_budget_git.py"
• uncovered: "tools/evidence_inventory.json"
• uncovered: "tools/kv_live_job.py"
• uncovered: "tools/kv_live_policy.py"
• uncovered: "tools/kv_pilot_admission.py"
• uncovered: "tools/skill_eval_budget.py"
• uncovered: "tools/skill_eval_budget_git.py"

[Dong1017]

Independent cost/admission review of 1566c3700079c4eab133058083bdb48c12e88333 (tree 371c048c6720d17192775ced6c34707704445329): no material findings in this scoped software change.

• Default lifetime PR cap remains $200; the prospective $350 exception is bound to repository 1409534281 / charge PR275, the exact prior claimed binding and checkpoints, and one fixed additional 8+8 attempt. Ledger JSON cannot supply or change this trusted context.
• Old unknown claims remain fully retained. Namespace/history checks reject third attempts, identity/PR/day resets and partial reservations; daily-first CAS and per-request checks prevent a concurrent extra admission. The $1,000 day and pilot ceilings are unchanged.
• Independently ran 277 offline tests across budget core/Git adapter, admission, policy/controller and evidence inventory. Extra fake-only checks covered competing different second-attempt IDs (one winner), 16 grants, rejection of a 17th, unchanged original binding and no refunds.

This is software review only, not execution clearance or a formal approval. No provider call, credential access or ledger mutation was used in this review. New source-bound hosted qualification, fresh public pricing/group and all-ledger checks, and an exact versioned freeze/locator with sufficient admission runway remain required. With unchanged commitments, retaining $162.72 plus $491.35 and adding one $162.72 attempt would total PR275/pilot $325.44 and day $816.79; these figures are bounds, not actual billing or a new reservation.

[Dong1017]

Published-byte/root-freeze binding: scoped software LGTM at 1566c3700079c4eab133058083bdb48c12e88333, complementing the separate cost/admission review. No additional material finding.

The published tree 371c048c6720d17192775ced6c34707704445329 and all 16 changed blobs match independently reviewed local 1459fdd2; base is ac4c54bc. Reused my 50 passing focused offline policy/controller tests, without repeating the full suite.

The exception reaches the anchored PR reader only through exact owner-comment/locator verification and immutable key-bound policy. Ledger JSON, candidate/worker data, CLI and environment are not cap authority; daily-reader exception injection is rejected. Historical v1 binding 1f7051e…, frozen task/K/input/qualification bytes, locator, workflow and worker remain byte-identical. Independently calling qualification validation on the retained old plan rejects the changed source with “Actor qualification dependency closure changed”; the old receipt cannot silently qualify this patch.

Current gates and scope limits

Ready 41182807900 and native review remain in progress at this read; declarations pass. Superseded cancelled jobs are not passes. The author owns terminal CI/qualification monitoring.

This is software review, not final merge or paid-execution clearance. Fresh exact-source hosted qualification/raw-evidence review, exact PR275/$350/one-additional-pair v2 freeze and locator, current pricing/expiry, same-host preflight and fresh anchored admission remain required. The first run's omitted cause and billing remain unknown. No provider/model call, credential access, ledger write or workflow dispatch occurred in this review.

COMMENTED, not formal approval or efficacy evidence.

[Dong1017]

The current software-check wait is cleared at 1566c370; operational execution prerequisites remain open.

• Ready now passes both Python full-check jobs, both installed-package jobs and model-free KV qualification. The current-head declaration policy and separate pull_request_target plan/publication also pass; execution is skipped.
• Incremental source inspection found no additional blocker in the root-bound cap context, immutable historical accounting, daily-first CAS and one-additional-attempt guard. This complements the existing cost/admission and root/freeze reviews; their independently reported local tests are not new tests by this task.
• The existing author/integrator can assess the software merge with those scoped reviews and terminal checks. A fresh source-bound operational qualification, exact v2 freeze/locator, valid prices/expiry, same-host preflight and anchored admission remain necessary for any separately authorized execution. The old unknown claim stays retained. No execution clearance or whole-Issue acceptance is issued here.

Evidence and limits

Reviewed head: 1566c3700079c4eab133058083bdb48c12e88333; API base / observed main: ac4c54bc06d44e11b067f5efb92de93c17dc4312. Ready 41182807900 records this head/base pair. Current policy succeeds; Skill Eval 41182808359 plans zero comparisons and does not execute them. Disposition is deterministic software/control behavior only; uncovered evaluation scope is not a pass or evidence of model benefit.

Inspected all 16 changed-file diffs and relevant existing root authorization, admission and Git-ledger consumers. Source and retained hosted-evidence inspection only: no local tests, provider calls, workflow actions, credential access or ledger writes. COMMENTED, with no formal approval or new LGTM claim.