[cemililik]

Summary

• Compare Markdown with the event's immutable base snapshot so squash merges and shallow reruns cannot invalidate its merge base. Keep full-tree fallback and the tooling census regression.
• Preserve the first native constructor-stop test failure and bounded cleanup observations. Wait for the preceding native operation's actual private cleanup before issuing the fixture's single extra same-generation hint.
• Bound stalled browser dependency acquisition with an exclusive APT fragment and verified effective settings. Allocate 17 minutes to outer setup/proof and reserve three minutes for bounded tail steps; retain the 20-minute job, 12-minute proof and 30-second native cleanup ceilings.

Tests

• Five real-Git controls execute the workflow link-audit shell; the old implementation fails them.
• Sixty-three offline native launcher/parser controls, seven corpus/Docker-trait cases, zero-warning/error integration Release builds and scoped C# formatting pass on the unchanged backend correction sources.
• Six real PostgreSQL/native-worker handoff cases pass. Removing readiness guards consumes a real premature hint and intentionally fails the regression; actual joins and exact owned-container absences are verified.
• Twenty-two actual browser/deadline/tail workflow controls pass. Full frontend validation passes 1,857 cases in two packages with zero skips/todos; typecheck, lint and scoped formatting pass.
• Fifteen real-Git snapshot controls pass, including actual foreground maintenance completion and the planted graph-write failure. The combined shard-owner/aggregate suite passes 67 controls; closed gate diagnostics retain identical refusal behavior.
• All 19 actual Outbox host/processor cases pass without skips. The fixture disables its unowned collector and observes actual startup before disposal; an injected startup delay distinguishes the join from a deliberately abandoned task. Real adapter output retains only finite result/completion facts. All owned cleanup observations are confirmed.
• Every correction completes two fresh independent review rounds. Offline shell checks establish command arguments and failure propagation, not hosted signal timing or fit.

Frozen head: dada277bdd66a51dd10ef8c4ea66ebbb6f50b170; tree: 46e742d9d262b070542ee55a99335b723956e0f3.
Full B attempt 2 passes independent producer/public/identity verification. Full A attempt 3 has a verified failing native handoff case; complete same-source A/B qualification remains open and this PR stays draft.

Earlier observations remain distinct: source 9eecf967 passes complete B but its A attempts fail browser APT setup. Source e2954be9 passes complete A (27 jobs, 5,921 backend cases); B passes all backend cases and both aggregates but outer-deadline cancellation leaves 26/27 public audit report pairs. Successful natural cleanup does not qualify those failed whole attempts.

At source 58d9ecdf, all 20 producers and public demo pass, but whole A is 24/27: a synthetic Git-cleanup failure blocks OpenAPI and both aggregates refuse a producer gate. The fixture now waits actual maintenance; closed diagnostics preserve the original gate observation. The first aggregate API response remains unavailable, so projection lag is unproven. That diagnostic correction adds no polling or acceptance relaxation.

At source 95edb219, whole A is again 24/27: two Outbox startup cases fail in ordinary-10 and both aggregates refuse producer state. The baseline fixture independently reproduces those two failures against an owned held collector; this verifies a mechanism, not that hosted exception. Closed producer-state diagnostics now identify the refused owner/state before the unchanged failure. The original hosted exception and backend API response remain unavailable.

At source ed223aca, all twenty producers (5,921 cases) and the other five checks pass, but both aggregates observe a required producer still in_progress and refuse: whole A is 25/27. Metadata acquisition now permits at most 30 seconds, sixteen complete reads and two-second pacing for explicit present-null pending states. Every snapshot is independently validated; terminal/malformed/identity-invalid evidence refuses immediately, and snapshots never combine. A single complete successful snapshot remains mandatory before unchanged artifact checks. New source full A/B is pending.

Current frozen source dada277b, run 39805658821 attempt 1: both aggregates and all twenty producers pass. Independent archive/result recomputation confirms 3,350 non-Docker + 2,571 Docker cases and both aggregate receipts. Whole A remains unqualified, 26/27: Ubuntu package downloads reach the unchanged five-minute browser-installer limit before any public proof. Full B attempt 2 now passes all 27 jobs and independent verification: 5,921 backend cases, 1,857 frontend cases, 27 public JSON/HTML report pairs, 9 screenshots, 134 API completions and natural confirmed cleanup. Its backend summary SHA-256 is 6dec960a69056da115ec6691a3cedee8e50670da505d1ea7e43c2e07a866e79d; public receipt SHA-256 is ced9f6afa005f33f164071b9d1f4022dacf16bf298c2a17677c70fd3dcefc845. Full A attempt 3 uses the same frozen source but fails backend ordinary-8 (job 119943000095): 243 assigned cases finish, with one failure in Native_fixture_start_waits_for_previous_actual_cleanup_before_one_same_generation_hint. Its retained result records cancellation at the provider-construction wait, 31.379 seconds, after the preceding cleanup and hint helper have returned. The same case passed in A1 and B2 in 3.472 and 2.682 seconds. Two independent source/evidence inspections confirm the boundary; the artifacts do not establish the post-hint cause. This failed whole attempt cannot qualify A or be combined with successful subsets. No further behavior or deadline change, or new workflow rerun, has been made for this failure.

PR #34 is merged and P02b-6's four implementation steps are complete. This PR is the separate CI closeout; P02b-7 and P02b-8 remain pending.

Risk

The metadata allocation consumes time inside the same five-minute aggregate; its sixty-second upload reserve remains fixed. Thirty seconds is not a guaranteed API propagation bound.

Application, realm and credential policies, shard plan, case identities, required contexts, aggregate API identity checks and permissions remain intact. The outer allocation changes from 16/4 to 17/3 within the existing ceilings. Sandbox cleanup and receipt generation use 15+5-second timeout wrappers; NuGet uses restore-only caching at the same pinned revision. Tail allocations are nominal scheduling allowances, not a worst-case OS guarantee. APT settings do not guarantee dependency availability or total setup duration.

The earlier 32-second native cancellation remains causally unexplained. The controlled fixture weakness is reproduced and corrected independently. Native success, elapsed timeout and forced cleanup never substitute for actual private cleanup. Partial/failed-job reruns never qualify a run.

Related

• PR feat(identity): reconcile development realm and prove real issuance #34
• CI correction, native failure attribution, owned fixture handoff and public-demo qualification records in the P02b-6 roadmap
• ADR-0056; ADR-0063

Prepared with Codex.

Summary by Sourcery

Harden CI evidence and fixture ownership so audits remain reliable across reruns while native and hosted failures retain their original causes.

New Features:

• Add regression coverage for immutable-snapshot Markdown link auditing across squash merges and shallow reruns.
• Add bounded browser dependency setup and tail-step handling for the public demo workflow.
• Add native handoff coverage that verifies actual private cleanup before issuing a same-generation fixture hint.

Bug Fixes:

• Preserve the original native test failure while recording bounded cleanup diagnostics.
• Prevent native fixture hints from racing preceding private cleanup.
• Prevent hosted Markdown audits from failing on invalid merge-base availability after squash merges.
• Ensure Outbox startup tests observe task completion before fixture disposal.
• Keep aggregate gate refusals while exposing only sanitized producer and gate diagnostics.

Enhancements:

• Refine CI timing allocation and use restore-only NuGet caching for the public demo job.
• Strengthen Git maintenance fixture cleanup and regression coverage for foreground completion.
• Expand native, aggregate, browser setup, and workflow-tail validation with bounded, sanitized evidence.

CI:

• Bound APT acquisition and workflow cleanup commands while retaining existing job, proof, and native cleanup ceilings.
• Add workflow controls for link auditing, browser installation, tail command failure propagation, and producer refusal diagnostics.

Documentation:

• Update testing, contribution, architecture, and roadmap documentation with post-merge qualification rules and evidence limits.

Tests:

• Add real-Git Markdown audit controls, native handoff/cleanup tests, Outbox startup ownership tests, browser setup controls, aggregate diagnostics controls, and Git maintenance regressions.

[sourcery-ai]

Reviewer's Guide

This PR fixes closed-PR Markdown audit reruns by comparing the checkout against the immutable event base commit rather than a later shallow base-branch tip, eliminating merge-base failures while retaining full-tree fallback auditing. It adds workflow-level real-Git regressions and extensively documents qualification timing, evidence boundaries, and the unchanged aggregate identity controls.

Sequence diagram for immutable-base Markdown link audit

sequenceDiagram
    participant Workflow
    participant Git
    participant Audit

    Workflow->>Workflow: Read PR_BASE_SHA
    Workflow->>Git: git diff --name-only PR_BASE_SHA HEAD -- *.md
    Git-->>Workflow: Changed Markdown files
    Workflow->>Audit: Check relative links in changed files
    Audit-->>Workflow: Pass or report broken links

Loading

Flow diagram for Markdown audit source selection

flowchart TD
    A[Determine event type] --> B{Pull request event?}
    B -->|Yes| C[Use immutable pull_request.base.sha]
    B -->|No| D[Use github.event.before]
    C --> E{Base commit available?}
    D --> E
    E -->|Yes| F[Compare snapshots with git diff base HEAD]
    E -->|No| G[Audit all tracked Markdown files]
    F --> H[Audit relative Markdown links]
    G --> H

Loading

File-Level Changes

Change	Details	Files
Switch the Markdown link audit from branch ancestry comparison to immutable event-commit snapshot comparison, while preserving full-tree fallback behavior.	Use the pull request event’s base SHA instead of fetching the current base branch. Use two-tree diffing that does not require a merge base. Retain unresolved-base and push fallback audits over all tracked Markdown files. Add five real-Git regression scenarios covering squash/shallow histories, broken links, pushes, fallback bases, and empty ranges.	.github/workflows/ci.yml scripts/test-markdown-link-audit.py
Document the post-merge rerun failure, correction scope, evidence limits, and requirement to complete qualification before PR closure.	Require complete CI qualification while the PR remains open because closed PRs may lose API identity projection. Record PR feat(identity): reconcile development realm and prove real issuance #34 merge status, failed post-merge aggregate identity checks, local validation, and independent review evidence. Clarify that hosted same-source A/B qualification remains pending and that identity checks, permissions, job names, and ceilings are unchanged.	.github/CONTRIBUTING.md CLAUDE.md README.md docs/roadmap/README.md docs/roadmap/phase-02b-events-auth.md docs/standards/06-testing.md

---

Tips and commands

Interacting with Sourcery

• Trigger a new review: Comment @sourcery-ai review on the pull request.
• Continue discussions: Reply directly to Sourcery's review comments.
• Generate a GitHub issue from a review comment: Ask Sourcery to create an
  issue from a review comment by replying to it. You can also reply to a
  review comment with @sourcery-ai issue to create an issue from it.
• Generate a pull request title: Write @sourcery-ai anywhere in the pull
  request title to generate a title at any time. You can also comment
  @sourcery-ai title on the pull request to (re-)generate the title at any time.
• Generate a pull request summary: Write @sourcery-ai summary anywhere in
  the pull request body to generate a PR summary at any time exactly where you
  want it. You can also comment @sourcery-ai summary on the pull request to
  (re-)generate the summary at any time.
• Generate reviewer's guide: Comment @sourcery-ai guide on the pull
  request to (re-)generate the reviewer's guide at any time.
• Resolve all Sourcery comments: Comment @sourcery-ai resolve on the
  pull request to resolve all Sourcery comments. Useful if you've already
  addressed all the comments and don't want to see them anymore.
• Dismiss all Sourcery reviews: Comment @sourcery-ai dismiss on the pull
  request to dismiss all existing Sourcery reviews. Especially useful if you
  want to start fresh with a new review - don't forget to comment
  @sourcery-ai review to trigger a new review!

Customizing Your Experience

Access your dashboard to:

• Enable or disable review features such as the Sourcery-generated pull request
  summary, the reviewer's guide, and others.
• Change the review language.
• Add, remove or edit custom review instructions.
• Adjust other review settings.

Getting Help

• Contact our support team for questions or feedback.
• Visit our documentation for detailed guides and information.
• Keep in touch with the Sourcery team by following us on X/Twitter, LinkedIn or GitHub.