upstream-drift: record the provenance convention entry and add the cleaned-state CI gate

Refs #3465. Slice 3 of 3. Blocked by slice 2 (the disposition sweep).

Problem

#3465 ends with a convention engagement that fires at sweep completion, a narrow CI gate that keeps the cleaned state clean, and two items recorded at the same firing: the hash-store designed-issue trigger and the dead-pointer round-trip wiring. The engagement was drafted in docs/specs/provenance-convention-engagement.md, which #5719 deleted; its content survives only in the #3465 body ("The convention engagement"). The upstream-drift convention is now 2.0.0 with a new record shape.

Fix

Read the slice-1 and slice-2 comments on #3465 as the evidence. Then, in one PR:

Convention entry. Add ONE entry to docs/conventions/upstream-drift/CHANGELOG.md, graded against convention 2.0.0, with the three parts from the provenance sweep sub-topic: repo-wide sweep, upstream-drift convention engagement, and the cleaned-state gate #3465 body: (a) re-derive the chore(conventions): plugins restating upstream harness facts are not adopters of upstream-drift, so stale claims never surface #2273 adoption trigger ("a detector is demonstrated that separates an upstream restatement from an in-repo one without a suppression list") from the closure count, measured precision/recall at the stated n, and the carve-out record; if per-instance suppressions accumulated, record that the trigger has not fired, with the count; (b) the enforceability table: major bump only if an enforceability verdict changes, otherwise a minor entry with the as-of date refreshed; (c) the trigger-less-stamp row moves to "built, off-by-default, repo-override enabled". Never rewrite earlier entries, never declare the plugin-eval deferral trigger fired, never turn a report-only tier into a deterministic claim.
Hash-store trigger. In the same entry, evaluate the hash-store designed-issue trigger (docs/conventions/upstream-drift/README.md, the deferred hash store) against the sweep's fetch telemetry, and state fired or not fired with the numbers.
Dead-pointer wiring. Route dead links found by .github/workflows/maintenance-check-links-online-claude-code-plugins.yml to the demotion path in plugins/attribution/skills/audit/reference/dispositions.md ("The demotion path when a pointer later dies"), as consuming-repo integration (the plugin file says the consuming repo owns the trigger).
Cleaned-state gate. Add a narrow CI step in .github/workflows/pr-require-checks.yml that runs the deterministic half of the attribution audit (stamps and fingerprint, no model call) on changed markdown and fails on a new fix-eligible finding. Keep it suppression-free; report-only tiers stay report-only.

Done when

git diff origin/main...HEAD -- docs/conventions/upstream-drift/CHANGELOG.md | grep -c '^+## ' prints 1
actionlint .github/workflows/pr-require-checks.yml .github/workflows/maintenance-check-links-online-claude-code-plugins.yml exits 0
The new gate step's script has a .test.sh with a fixture that fails the gate and one that passes; bash <that test> exits 0
ci-status is green on the PR head

Proposed work class: work-class: scoped
Basis: automated intake triage; the lanes App applied this class.

Size: Medium
Complexity: Medium
Uncertainty: Knowable
Flags: none

Understood: this is slice 3 of #3465. In one PR it adds a single upstream-drift CHANGELOG entry, which includes the hash-store trigger evaluation. It also routes dead links from the online link-check workflow to the demotion path, and adds a suppression-free CI gate step in pr-require-checks.yml with a pass fixture and a fail fixture. The issue says it is blocked by slice 2 (the disposition sweep), so the evidence from the #3465 slice comments has to exist first. No likely duplicates found among the open issues.

_{Automated triage, applied by the lanes App.}

Not admitted to the issue lane yet: blocker #9755 is not closed as completed by a listed human or a merge.

Cordoned until the disposition sweep in #9755 runs. Issue Triage session.

This slice's evidence is the slice-2 sweep comment on #3465: the closure count, the measured precision and n, and the carve-out record. Three of its four steps wait on that sweep:

the convention entry re-derives the chore(conventions): plugins restating upstream harness facts are not adopters of upstream-drift, so stale claims never surface #2273 adoption trigger from that comment;
the hash-store trigger is evaluated against the sweep's fetch telemetry;
the cleaned-state gate fails on "a new fix-eligible finding".

Step 3, routing dead links to the demotion path, does not need the sweep: reference/dispositions.md makes that wiring the consuming repo's integration, and #3465 ties it to the sweep only by scheduling ("recorded at the same firing"). It stays in this slice because the slice ships as one PR.

The owner ruled on 2026-10-30 that no class is fix-eligible yet. The owner-decisions record (docs/owner-decisions.md in melodic-software/architecture, row "Q1 and Q2 (Backlog HITL numbering), 2026-10-30") records the owner's reply, verbatim: "go w/ recommended". The decision: "(a) not yet; re-measure on a larger sample before any class is declared fix-eligible". Until a class is fix-eligible, the sweep cannot run and the gate has nothing to fail on. #9755 is cordoned on that condition.

What reopens this: #9755 closes as completed, and #3465 carries a comment headed ## Disposition sweep, <date>. Then take this issue off After zero and send it back through intake with a **Decision recorded.** comment. Its blocked-by edge to #9755 stays as it is. At re-intake, also repoint the body's hash-store trigger from docs/conventions/upstream-drift/README.md to docs/contracts/upstream-drift/README.md, where that section now lives.

Status. Out of After zero: both reopen conditions in the 2026-11-01T15:38Z cordon comment are met. Routed to Automation Lanes' builder, because Fix steps 3 and 4 edit .github/workflows files, which the issue lane cannot push. Stays at status: confirmed, no agent-ready.
Agent choice, not an owner answer.

attribution: run the disposition-applying provenance sweep once a class is fix-eligible #9755 closed as completed at 2026-11-04T14:59:04Z.
provenance sweep sub-topic: repo-wide sweep, upstream-drift convention engagement, and the cleaned-state gate #3465 carries ## Disposition sweep, 2026-11-04 (2026-11-04T14:58:50Z).

For the builder:

The 2026-11-04T15:25Z note on this issue says the sweep's zero dispositions are not evidence the repository is clean. State those limits next to the closure count in the convention entry's trigger.
The owner ruled no class fix-eligible yet (owner-decisions, "Q1 and Q2 (Backlog HITL numbering), 2026-10-30"). The cleaned-state gate ships able to fail, with its failing fixture, but has nothing to fail on today.
The cordon comment asks that the body's hash-store trigger path move from docs/conventions/upstream-drift/README.md to docs/contracts/upstream-drift/README.md, where that section now lives. Read the body's path as the contracts one.

docs(upstream-drift): state #3567's reopen condition instead of claim…

docs(upstream-drift): say how the entry reads the undefined term flee…

Owner decisions Q-9759-1 and Q-9759-2, 2026-11-05, in session. The owner pasted the question lines and wrote, verbatim:

Go with recommended here, as long as its in line with everything above:

Q5 (Q-9759-1): I recommend (a), a ci-workflows link-check output plus a local filer.

Q6 (Q-9759-2): I recommend (b), turn the gate on only after a sweep applies dispositions.

Decided:

Step 3 (dead-pointer routing): (a). The ci-workflows link-check reusable workflow emits its dead pointers as an output or artifact, and a local workflow that holds only issues: write files them.
Step 4 (gate predicate): (b). The gate is deferred until a sweep applies dispositions; on 11-04, 0 were applied.

fix(ci): label new dead-link issues needs-triage and say when to clos…

fix(ci): keep one rollup issue for dead links instead of one per link

test(ci): disable SC2016 file-wide in the dead-link filer test

fix(ci): never close the dead-link rollup on bad data, and cap its size

docs(upstream-drift): correct the gate deferral reason and trigger label

Status: steps 1–3 shipped, step 4 deferred, and this issue stays open as the gate's tracker. Step 1 (the convention entry) and step 2 (the hash-store trigger) shipped in #13461. Step 3 (dead-pointer routing) shipped in #13549 with melodic-software/ci-workflows#993. Owner decision Q-9759-2 (b) defers step 4, the cleaned-state gate with its failing and passing .test.sh fixtures, until a disposition sweep applies at least one disposition. #13595 records that decision and its recheck trigger in docs/contracts/upstream-drift/README.md. pr-verify ruled that closing this issue would drop the tracked gate work, so it stays open until that trigger fires and the gate is built.

Closed: steps 1–3 shipped, and the step-4 gate moved to the hardening pass.

Steps 1 and 2 (the convention entry and the hash-store trigger): docs(upstream-drift): record the provenance convention engagement as 4.2.0 #13461.
Step 3 (dead-pointer routing): feat(ci): file one issue per dead link the online link check finds #13549 with feat(maintenance-check-links-online): upload dead links as a short-lived artifact ci-workflows#993.
Step 4 (the cleaned-state gate with its failing and passing .test.sh fixtures): owner decision Q-9759-2 (b) defers it until a disposition sweep applies at least one disposition. docs(upstream-drift): record the cleaned-state gate deferral and its trigger #13595 records that decision and its recheck trigger in docs/contracts/upstream-drift/README.md.

The owner's 2026-11-05 direction, relayed by the GitHub Support hub: "If there are any issues that are tied to hardening, close them because we'll reopen and do those one at a time. That'll come from rediscovery." The deferred gate is recorded in the hardening seed list (.work/gh-monitor/hardening-pass-seed.md, a local file), with that trigger event.

Navigation Menu

Repository navigation

upstream-drift: record the provenance convention entry and add the cleaned-state CI gate #9759

Problem

Fix

Done when

3 remaining items

Metadata

Assignees

Labels

Type

Fields

Projects

Milestone

Relationships

Development

upstream-drift: record the provenance convention entry and add the cleaned-state CI gate #9759

Description

Problem

Fix

Done when

Activity

github-actions commented on Oct 24, 2026

melodic-automation-lanes commented on Oct 24, 2026

kyle-sexton commented on Nov 1, 2026

3 remaining items

kyle-sexton commented on Nov 5, 2026

kyle-sexton commented on Nov 5, 2026

kyle-sexton commented on Nov 5, 2026

kyle-sexton commented on Nov 6, 2026

Metadata

Metadata

Assignees

Labels

Type

Fields

Projects

Milestone

Relationships

Development

Issue actions