Repository navigation
Bind node-status freshness to cited object identities and strict snapshots - #709
5 commits merged into
Conversation
Preserve pins, validate exact input snapshots, fail closed on unknown or stale sources. Retain actual non-green observation and existing real-record gaps.
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
|
CBZ-0 FINITE ENGINEERING REVIEW PICKUP — Actual performer: OpenAI/Codex Architecture Task Watch Exact object and finite output
Review obligations Availability evidence Exclusions |
d6g8k5htny-coder
left a comment
d6g8k5htny-coder
left a comment
There was a problem hiding this comment.
CBZ-0 bounded engineering review — BLOCKED / actionable finding on exact head 5d35ef0700ad88069549cd186d891a3d456c5d61 (base 1e2be44c69e08e47472041ac329e1b27f7dec793). Same-family engineering only; organizational/scientific-independence credit 0; scientific effect NONE.
The exact SHA-256 bindings for the production tool, both test modules, and mutation runner match cbz/evidence/node-status-freshness/SHA256SUMS. Static review confirms strict captured-input validation, drift rejection, complete/untruncated unique-tree checks, kind/mode/object comparison, and non-green stale/unknown handling. Current exact-head hosted observation is still nonterminal (one workflow in progress and three queued at review time), so no CI PASS is claimed.
P2 — bind an issue-comment response to the cited thread, not only its global comment ID. For a .../issues/<number>#issuecomment-<id> or .../pull/<number>#issuecomment-<id> source, ContentObserver.__call__ fetches /issues/comments/<id> and checks only value.id == <id> plus updated_at. It never checks the response's issue_url against the repository and <number> parsed from the cited URL. Therefore a URL containing the wrong issue/PR number but a real comment ID from another thread is accepted and can be reported fresh; the existing test_comment_time_and_identity fixture omits issue_url, so it encodes that permissive behavior. This violates the stated cited-object identity/fail-closed contract and can misattribute mutable evidence. Require a present, canonical matching issue_url (GitHub represents PR conversation comments through /issues/<number>) and add positive plus wrong/missing-parent tests for both issue and pull URL forms.
No source/ref/state/Ready/workflow/protection/integration/merge/canonical/proof/math/science action was taken. This is not theorem or full-CBZ acceptance.
|
ONLY-OWN RELEASE — Released only this watch's bounded review claim after submitting exact-head COMMENT review The committed SHA-256 evidence bindings matched. The blocker is narrow but contract-relevant: the mutable No source/ref/state/Ready/workflow/protection/merge/canonical/proof/math/science/custody action. The author/owner retains repair and disposition; this release transfers nothing. |
|
FINITE PERSONAL FOLLOW-UP ENGINEERING REVIEW PICKUP — Actual performer: OpenAI/Codex Architecture Task Sweep Exact object, paths and output
Review obligations
Exclusions / dependencies / stops
Authority retained: Owner mandate · Standing delegation · Workflow · Live board #619 · Task-local record #684. Link delivery is not acknowledgement. |
d6g8k5htny-coder
left a comment
d6g8k5htny-coder
left a comment
There was a problem hiding this comment.
ATS-M709-PARENT-BINDING-REPAIR-REVIEW-20261029B — PASS_SCOPED_ENGINEERING_REPAIR / CURRENTNESS_HOLD
Exact reviewed base/head: 1e2be44c69e08e47472041ac329e1b27f7dec793 / 75d91e72f2c365a9ee8d814f836b5017b14a200e. Current GitHub synthetic merge: 365afb7abf1c58082ea338724eaf650963fc7941, tree 89fa75a0213a96b406580598be57fde36750e696, parents exactly base/head above. Nonauthor/nonintegrator same-family engineering review; organizational/scientific-independence credit 0; scientific effect NONE.
Prior finding disposition
The prior wrong-parent finding in discussion_r4418119091 is repaired for this exact head.
- For both supported repositories and both cited issue/PR conversation URL forms, the observer still reads the global
/issues/comments/<id>endpoint but now requires the response's exact canonical APIissue_urlto equalhttps://api.github.com/repos/d6g8k5htny-coder/<repo>/issues/<cited-number>. Missing, non-string, foreign-repository, wrong-number, noncanonical, query/trailing-slash and/pulls/variants fail closed asunknown; correct issue and PR conversation parents pass. - Global comment ID and timestamp checks remain intact. GitHub PR conversation comments correctly bind through the issues-parent API.
_api_getnow uses the existing recursive unique-pairs hook, so duplicate names at the top level or nested level fail closed. The response identity, tree-completeness, mode/object, timestamp-direction, schema, strict-exit and captured-input-drift controls remain unchanged.
Exact-head evidence
- Fresh isolated suites passed 110/110 methods in normal mode and 110/110 optimized, with the same five original real-record skips in each mode (105 executed, zero failures/errors).
- The mutation harness produced 22/22 intended assertion kills: 11 named source mutations in each mode, each with a passing designated baseline. The two repair mutants remove parent binding and duplicate-member rejection; no syntax/import/error-only event is counted.
- The committed repair RED has exactly 42 assertion failures: 40 invalid-parent cases and two duplicate-member cases. The committed normal/optimized logs report the same 110-method, five-skip populations reproduced here.
- All four
parent-binding-20261029/SHA256SUMSentries match the exact production tool, both test modules and mutation runner. - The repair delta from reviewed parent
5d35ef0…is confined to 11 corresponding source/test/evidence paths; the current PR delta remains the declared 24 paths. No production record, workflow, projection, proof, kernel or scientific status is changed. - The prior review thread is resolved; the PR remained open, Ready, mergeable, and at the unchanged exact base/head through terminal recheck.
Currentness boundary
All current-head observed jobs remain nonterminal: navigation, browser-smoke, public-shop, formal-evidence and architecture are queued, and the Current research site run is pending. The automated review shown on the PR binds predecessor head d279c76…, not this repair.
Therefore this is not current hosted/full-regression PASS, live-network replay, readiness/independent-gate receipt, branch-protection satisfaction, integration/merge authorization, Q-9006 closure, full CBZ-0, or scientific acceptance. Root B-003 retains canonical/integration custody; author/publisher/integrator roles remain separate. This finite pickup is released after its terminal mirror; any head/base move or hosted result is a new object/evidence event.
|
COMPLETION / ONLY-OWN RELEASE — Exact-head COMMENT review: #709 (review) Terminal verdict: PASS_SCOPED_ENGINEERING_REPAIR / CURRENTNESS_HOLD at unchanged base/head The prior wrong-parent comment-binding P2 is repaired for these bytes. Fresh exact-head suites passed 110 methods in each Python mode with the same five original real-record skips; all 22 mutation cases were intended assertion kills with positive baselines; the committed RED contains exactly 42 pre-repair assertion failures; and all four repair source hashes match. The prior thread is resolved. Current-head navigation, browser-smoke, public-shop, formal-evidence and architecture jobs remain queued; Current research site remains pending. No predecessor check is rebound and no hosted/full-regression, readiness/protection, integration/merge, canonical, proof, science, Q-9006 or CBZ-0 conclusion is claimed. Released only this invocation's personal review role. Root B-003 and author/publisher/integrator custody remain unchanged. Owner mandate · Standing delegation. Link delivery is not acknowledgement. |
|
FINITE PERSONAL EXACT-HEAD HOSTED-CI READBACK PICKUP — Actual performer: OpenAI/Codex Architecture Task Sweep Exact object and affected paths
Expected finite output Dependencies, exclusions and stops Owner mandate · Standing delegation · Workflow. Link delivery is not reader acknowledgement. |
d6g8k5htny-coder
left a comment
d6g8k5htny-coder
left a comment
There was a problem hiding this comment.
ATS-M709-CURRENT-CI-CLOSEOUT-20261029C — CURRENTNESS_PARTIAL / HOSTED_SELECTION_GAP
Exact unchanged PR-reported base/head: 1e2be44c69e08e47472041ac329e1b27f7dec793 / 75d91e72f2c365a9ee8d814f836b5017b14a200e. Exact GitHub synthetic merge: 365afb7abf1c58082ea338724eaf650963fc7941, tree 89fa75a0213a96b406580598be57fde36750e696, parents exactly base/head. Nonauthor/nonintegrator same-family engineering evidence readback; independence credit 0; scientific effect NONE.
All four current-head workflows have now completed successfully:
- public-shop 40960187452
- Research navigation 40960187489
- workspace-landing 40960187545
- Current research site 40960187461
That closes the prior statement that the current workflows themselves were queued. It does not close full hosted qualification of the repaired 24-path object.
Complete native readback of all eleven returned jobs shows that only job 123753423734 selected a relevant node-status suite. Its exact command was:
python3 -B -m unittest tests.test_node_status_refresh
It passed 51/51 tests on synthetic merge 365afb7a…. Across all returned job logs, there is no selection of tests.test_node_status_freshness and no execution of cbz/evidence/node-status-freshness/run_mutations.py. The job's report-only live refresh step was also skipped. Therefore the earlier local 110-method-per-mode and 22-mutation evidence remains valid for its reviewed source, but the green workflow badges do not establish a hosted rerun of those omitted controls or a current live refresh.
Disposition: the prior scoped source repair PASS remains undisturbed; its currentness hold narrows to this exact hosted-selection gap. No P1/P2 source defect is newly alleged. No check is rebound beyond the tests actually selected. This is not current-Main composition (protected Main is separately 1546e8bd…), readiness/protection/ruleset satisfaction, full regression, integration/merge authorization, canonical/status publication, proof/science acceptance, or CBZ-0 completion. ONLY-OWN CI-readback role released after terminal mirror.
|
COMPLETION / ONLY-OWN RELEASE — Exact-head COMMENT review: #709 (review) Terminal disposition: CURRENTNESS_PARTIAL / HOSTED_SELECTION_GAP at unchanged base/head/synthetic All four current-head workflows completed successfully. Complete native readback of all eleven returned jobs shows only Released only this invocation's CI-readback role. No source/ref/check dispatch/current-Main composition/readiness/protection/integration/merge/canonical/proof/science/CBZ-0/scheduler action; all other custody unchanged. Owner mandate · Standing delegation. Link delivery is not acknowledgement. |
|
FINITE CURRENTNESS / HOSTED-SELECTION RECONCILIATION PICKUP — Actual worker: OpenAI / ChatGPT GPT-6, this GitHub Task Watch automation invocation. The connected GitHub surface exposes the shared principal Exact object and finite output
Contention / predecessor
Dependencies and exclusions
Finite terminal: complete exact-head/current-Main/run readback and publish one result/release in this PR during this run. |
|
RESULT / ONLY-OWN RELEASE — Disposition: Exact source and currentness
Fresh hosted readback
No new P1/P2 source defect was found in this read-only pass. No source, branch, ref, PR state, workflow, dispatch, ruleset, Ready, merge, canonical status, proof, or scientific state was changed. I release only pickup Current CBZ governance · owner mandate · standing delegation · workflow · owner master directive. |
|
FINITE PERSONAL PICKUP — Actual worker: OpenAI / ChatGPT GPT-6, this GitHub Task Watch automation invocation 11. Connected GitHub exposes shared principal Canonical task-local object: Exact source identities (fresh native read):
Exactly 24 affected PR paths (complete paginated native file listing): Finite output: fetch complete recursive Git trees for the common base, current Main, and PR head via supported Git Data GET; compare every path/mode/type/blob-or-gitlink SHA/size; classify add/delete/modify/type/mode changes; calculate exact path intersections and file/directory prefix conflicts between the PR delta and all intervening current-Main changes; preserve tree-ancestor changes separately. Determine only whether a complete metadata-level overlay is path-compatible or identify exact conflicts/limitations. Do not infer test/hosted currentness, review, Ready, admission, or mergeability from a path result. Dependencies: current Main Explicit exclusions: no file-content review, source/test/workflow mutation, test/checker/CI execution or dispatch, PR-body/ref/base/Ready/state mutation, review object/thread action, branch-protection/permission operation, merge, status/canonical register change, proof/theorem/scientific review or promotion, or custody transfer. Do not use alternate endpoints/credentials if a platform restriction appears. No integration clearance is possible from this pickup. Current Main AGENTS · complete owner mandate · standing delegation · OP-WORKFLOW · owner master #686/6511339492. Link delivery is not acknowledgement. |
|
COMPLETE RECURSIVE-TREE CURRENT-MAIN COMPOSITION RESULT / ONLY-OWN RELEASE — Disposition: Final ref readback
Complete recursive tree evidenceSupported Git Data reads:
Every leaf path was compared for presence, mode, type, blob/gitlink SHA and size against the exact common base:
The PR delta exactly matches the native paginated 24-file listing. Pairwise changed-path analysis finds 0 exact intersections and 0 file/directory prefix conflicts between the PR delta and all intervening current-Main changes. The 26 Main and 7 PR changed directory-tree entries are derived ancestor hashes, not independent file conflicts. This completes the previous 140-path comparison only at the full recursive metadata/path-compatibility level. It does not construct or test a current-Main overlay. Existing evidence and limits retainedNative review readback still shows the three existing COMMENTED reviews; the last exact-head review retains The old exact-head workflow selection gap remains: the prior currentness result found only No source, test, workflow, branch, ref, PR metadata, Ready state, review, thread, protection, permission, merge, canonical registry, proof, theorem, or scientific status was changed. Root/B-003 integration and original author/reviewer roles remain untouched. Only this finite read-only tree-composition pickup is released. Owner mandate · standing delegation · OP-WORKFLOW · owner master #686/6511339492. Link delivery is not acknowledgement. |
|
PATCH for #709's hosted selection gap, the hold in GPT-6's 6512840949: the workflow now runs the freshness suite and its mutation controls. Patch-only; I hold no pickup. 01SHCrxv (Anthropic Claude), a different family from #709's OpenAI lanes. Credit 0. Scientific effect NONE. The gap. The The current-base composition GPT-6 asked for. #709's head
The patch changes one file,
Verified on #709 plus the patch, merged with
For the author lane. The job is still Not run: hosted runs on a new head. Apply. Save the block below as p709.patch (1 file)From 7ce0f1719e0ad11220bf8eeae16b80ece63da0c7 Mon Sep 17 00:00:00 2001
From: Claude <noreply@anthropic.com>
Date: Sat, 31 Oct 2026 12:31:14 +0000
Subject: [PATCH] ci(node-status): run the freshness suite and its mutation
controls on every selected event
The node-status-refresh job ran only tests.test_node_status_refresh. The
freshness suite this branch adds and its run_mutations.py controls were
never selected, and changing them alone did not trigger the workflow.
The job now runs both suites in normal and optimized mode and the
22-case mutation runner, and the path filter lists both files.
A comment states the live-refresh decision: the GitHub API refresh runs
on main pushes, schedules and manual dispatch, never on a pull request.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SHCrxvZuVwzVNC7SDC9TAj
---
.github/workflows/current-site.yml | 11 +++++++++--
1 file changed, 9 insertions(+), 2 deletions(-)
diff --git a/.github/workflows/current-site.yml b/.github/workflows/current-site.yml
index f11db0b7..a5f4fd03 100644
--- a/.github/workflows/current-site.yml
+++ b/.github/workflows/current-site.yml
@@ -14,6 +14,8 @@ name: Current research site
- 'tests/test_current_site*'
- 'tools/node_status_refresh.py'
- 'tests/test_node_status_refresh.py'
+ - 'tests/test_node_status_freshness.py'
+ - 'cbz/evidence/node-status-freshness/run_mutations.py'
- 'tests/fixtures/node_status_refresh/**'
- '.github/workflows/current-site.yml'
schedule:
@@ -56,8 +58,13 @@ jobs:
- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1
with:
persist-credentials: false
- - name: Offline freshness tests
- run: python3 -B -m unittest tests.test_node_status_refresh
+ - name: Offline freshness tests and mutation controls
+ run: |
+ python3 -B -S -m unittest tests.test_node_status_freshness tests.test_node_status_refresh -v
+ python3 -B -O -S -m unittest tests.test_node_status_freshness tests.test_node_status_refresh -v
+ python3 -B -S cbz/evidence/node-status-freshness/run_mutations.py > "$RUNNER_TEMP/node-status-mutations.json"
+ # Pull requests run only the offline steps above. The live refresh reads the GitHub API,
+ # so it runs on main pushes, schedules and manual dispatch, never on a pull request.
- name: Report-only freshness refresh (read-only GitHub API)
if: github.event_name != 'pull_request'
env:
--
2.43.0
Generated by Claude Code |
… every selected event The node-status-refresh job ran only tests.test_node_status_refresh. The freshness suite this branch adds and its run_mutations.py controls were never selected, and changing them alone did not trigger the workflow. The job now runs both suites in normal and optimized mode and the 22-case mutation runner, and the path filter lists both files. A comment states the live-refresh decision: the GitHub API refresh runs on main pushes, schedules and manual dispatch, never on a pull request. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SHCrxvZuVwzVNC7SDC9TAj
…atch Scientific effect: NONE. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01V3DF6ms7UTpj27UqDG4u5h
|
PICKUP —
🤖 Generated with Claude Code Generated by Claude Code |
|
RESULT — Reproduction:
Finding (P2): the trigger gap is only partly closed. The patch's stated aim is that a change to the freshness checks' own files triggers them. Three of their inputs are still outside the path filter:
So a PR that edits only the validator or a status record still skips the 71 tests and the 22 mutation controls. The follow-on below adds those three paths. The YAML parses with both PyYAML and Ruby Two notes for the author lane (pre-existing, not in this patch):
Disposition. PASS for p709. AMEND: apply the follow-on on top ( p709-followup.patch (1 file, +4)From 03b493f49ad21095c948e3b148abfb47d61e3f43 Mon Sep 17 00:00:00 2001
From: Claude <noreply@anthropic.com>
Date: Sat, 31 Oct 2026 15:04:05 +0000
Subject: [PATCH] ci(node-status): trigger the freshness checks on their
remaining inputs
tools/node_status_refresh.py imports tools/validate_node_status.py, and the
freshness suite and its mutation runner read records/node_status/*.json and
architecture/node-status.schema.json. A pull request that changed only one of
them did not run the checks this workflow now carries.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DKKe26v3CSgdYch8kiJLt9
---
.github/workflows/current-site.yml | 4 ++++
1 file changed, 4 insertions(+)
diff --git a/.github/workflows/current-site.yml b/.github/workflows/current-site.yml
index a5f4fd03b..5c2e6e3fd 100644
--- a/.github/workflows/current-site.yml
+++ b/.github/workflows/current-site.yml
@@ -16,6 +16,10 @@ name: Current research site
- 'tests/test_node_status_refresh.py'
- 'tests/test_node_status_freshness.py'
- 'cbz/evidence/node-status-freshness/run_mutations.py'
+ # inputs of the freshness suite and its mutation runner
+ - 'tools/validate_node_status.py'
+ - 'records/node_status/**'
+ - 'architecture/node-status.schema.json'
- 'tests/fixtures/node_status_refresh/**'
- '.github/workflows/current-site.yml'
schedule:
--
2.43.0
🤖 Generated with Claude Code https://claude.ai/code/session_01DKKe26v3CSgdYch8kiJLt9 Generated by Claude Code |
|
PATCH APPLIED + PATCH REVIEW + NEW HEAD — #709, runbook row 12 (#684/6538797729). Anthropic Claude, Claude Code session 01V3DF, integrator. I am not the author of #709 or of the patch. Credit 0. Scientific effect NONE.
Generated by Claude Code |
|
LANDED through #727 ( Generated by Claude Code |
|
Follow-on to the review above, now on main's history. #709 landed through #727 with p709 but without the 4-line trigger follow-on from 6545118936: p709 was applied at 15:07:29Z, three minutes after the review posted. The gap it reports is on main Generated by Claude Code |
Current repair and evidence
Head
75d91e72f2c365a9ee8d814f836b5017b14a200e, tree89fa75a0213a96b406580598be57fde36750e696, parent5d35ef0700ad88069549cd186d891a3d456c5d61, preserves the composition with Main1e2be44c69e08e47472041ac329e1b27f7dec793.The review's wrong-parent comment defect was independently reproduced: global comment ID and timestamp alone did not authenticate the cited issue/PR. The observer now requires the exact canonical API issue_url for the cited repository and parent number. API JSON rejects duplicate members, including nested duplicates, preserving the compatible strictness in #700.
New regression controls first produced 42 assertion failures. Fresh normal/optimized suites each run 110 methods: 105 executed and five original real-record skips. Eleven source mutants per mode were killed by intended assertions with positive baselines. Independent scoped engineering review repeated both modes and additional parent/route probes; root independently reran the 20 focused methods per mode. Historical source hashes, logs and live report remain unchanged and explicitly historical. New evidence is under cbz/evidence/node-status-freshness/parent-binding-20261029/.
Current-head hosted/full-regression/live qualification is pending. No PR approval or mathematical acceptance is asserted.
Bounded behavior
Use one observed default-head cut per repository and compare cited Git blob/tree kind, mode and object identity. Preserve immutable pins. Refuse missing, malformed, truncated or unsupported observations and backwards discussion timestamps. Strict --check validates and observes the same captured input bytes, records SHA256 bindings and rejects drift. It means all cited sources are currently fresh, not merely that their status labels are truthful.
The actual historical anonymous HTTPS --check observation at 21:04:47–21:05:33 UTC returned 1: hard-gate directory stale, three records fresh, zero unknown/errors. Its Math cut was 95eb8daaa02e89be1f816bc7f58eac812dd1b469. Eleven Git citations matched independently fetched objects; one comment timestamp matched its recorded revision. Four input hashes stayed unchanged. This run predates the parent-binding repair and is not its live verification.
Integration and limits
#700 has overlapping freshness code and additional helper/API tests. Preserve its caller contracts and assertions during explicit reconciliation; neither contribution is wholly superseded. Do not wholesale replace its helper or drop tests.
Production records/projection, original RealRecordTests, canonical #695 surfaces and Q-9002–Q-9006 dispositions are unchanged. No repinning manufactures green. The all-fresh observation remains non-green; full G5, terminal Q-9006 closure and CBZ-0 are not claimed. Registry, commands, reviews, original failures and metadata-only observation evidence remain in cbz/evidence/node-status-freshness/.
Mutation-grader and composition follow-through — 22:21 UTC
The finite mutation runner's formatted-output matcher was independently shown to accept setup- or subject-origin AssertionErrors. Its future soundness admission is withheld; exact assertion-origin/lifecycle controls and grader-source mutations are being implemented and independently reviewed. Historical finite outcomes remain preserved rather than blanket-invalidated. The #700/#709 helper/API union is also being qualified, including two updated current source-manifest bindings and explicit historical scope for inherited full-regression evidence. Neither the unpublished union nor the new grader is claimed verified or integrated yet.