Tracking: streamline Zebra's software delivery lifecycle

[gustavovalverde]

Purpose

Track implementation of the accepted improvements to Zebra's review, merge queue, CI, changelog, and release workflows.

Decision-making predecessors

• SPIKE: Best approach for Zebra release process #10138 — release-process investigation
• Rebuild CI #9815 — CI rebuild investigation
• Tracking: Zebra test and build optimizations #9509 — test and build optimization tracker

Summary by area

Area	Target outcome	Primary workstream
Review and merge queue	Maintainers can finish approved PRs; source metadata is checked correctly; the queue recovers quickly and becomes the routine integration boundary	#10963
CI execution	Each check has one lifecycle owner; floating checks are advisory; obsolete work is cancelled; affected checks are selected safely	#10966
Changelog authoring	Source PRs add validated fragments instead of editing shared changelog sections	#10965
Release readiness and recovery	Releases use explicit triggers, reproducible inputs, readiness checks, an exact manifest, and resumable publication	#10964
Immutable artifacts	The same reproducible, signed artifact is promoted through release stages	#10156, informed by #10614

Execution plan

Phase 0: active unblockers

• fix(ci): Docker build broken — pinned Debian Trixie packages no longer available #10957 — repair the currently broken Docker build.
• Run check-no-git-dependencies CI job when A-release label is added late #10897 / PR fix(ci): rerun release dependency check on label #10955 — run the release dependency check when A-release is added late.
• Confirm the final missing-testnet-checkpoint and MSRV-classification outcomes and record them in devops: make releases reproducible, ready, and resumable #10964.

Phase 1: remove immediate delivery friction

• devops: simplify PR completion and stabilize the merge queue #10963 — simplify PR completion and stabilize the merge queue.
• devops: assign CI checks to lifecycle stages and add change-aware selection #10966 — assign CI responsibility, move floating checks to advisory schedules, cancel obsolete work, and start the impact planner in shadow mode.
• devops: do not run clippy on MSRV #10937 — stop running Clippy on MSRV.
• devops: clippy beta should not be required #10642 — stop requiring beta Clippy.
• ci: move link-check to a scheduled weekly workflow #10632 — move link checking to a scheduled workflow.

Phase 2: remove recurring authoring and release failures

• devops: replace shared changelog edits with validated fragments #10965 — adopt changelog fragments.
• devops: make releases reproducible, ready, and resumable #10964 — add release readiness, an exact manifest, and resumable publication.
• Workflow for publishing Docker images fails #10317 — resolve false or real failures in Docker publishing.
• update release checklist regarding crate releases #9701 — update the crate-release checklist.

Phase 3: enable optimizations after proof

• Require the focused queue check only after devops: simplify PR completion and stabilize the merge queue #10963 meets its queue-candidate and latency thresholds.
• Enable affected-package selection one check family at a time only after devops: assign CI checks to lifecycle stages and add change-aware selection #10966 completes its shadow period without a missed failure.
• Remove repeated queue or main checks only after their replacements prove equivalent coverage.
• Complete reproducible, signed, immutable artifact promotion in devops: reproducible Docker image builds with signed provenance #10156.

Related existing issues

Review and merge policy

• Pull requests with unresolved comments were merged #10034 — unresolved review comments must block merging.
• Commits in main inherit name of PR's last commit instead of PR title #10089 — squash-merged commits should use the PR title.
• devops: Update CODEOWNERS for DevOps review rotation #10139 — DevOps review ownership and CODEOWNERS rotation.
• Enable merge commits #10196 — merge commits proposal; related but explicitly excluded because this plan keeps squash merging.

CI architecture, speed, and coverage

• Rebuild CI #9815 — broader CI rebuild investigation.
• Tracking: Zebra test and build optimizations #9509 — existing test and build optimization tracker.
• devops: clippy beta should not be required #10642 — beta Clippy should not be required.
• devops: do not run clippy on MSRV #10937 — do not run Clippy on MSRV.
• ci: move link-check to a scheduled weekly workflow #10632 — move link checking to the weekly schedule.
• ci: no CI job builds or tests the vendored RocksDB engine #10933 — preserve coverage for the vendored RocksDB engine.
• task: investigate and catalogue flaky tests across the suite #10790 — catalogue flaky tests and inform scheduled detection.
• Automatic reporting of failures of scheduled integration tests doesn't work #10093 — repair scheduled-test failure reporting.
• Scheduled workflow logs are missing #10081 — restore scheduled workflow logs.
• Update Continuous Integration book page #9043 — update the CI book after workflows stabilize.
• devops: avoid maintaining extra .patch workflow files #8861 — avoid unnecessary duplicate patch workflows.

Release reliability and artifacts

• fix(ci): Docker build broken — pinned Debian Trixie packages no longer available #10957 — broken Debian package pins in Docker builds.
• Run check-no-git-dependencies CI job when A-release label is added late #10897 and PR fix(ci): rerun release dependency check on label #10955 — release-label dependency validation.
• Workflow for publishing Docker images fails #10317 — Docker publishing workflow failure.
• devops: reproducible Docker image builds with signed provenance #10156 — reproducible Docker images and signed provenance.
• Evaluate deterministic builds for Zebra (stagex or alternatives) #10614 — deterministic-build evaluation.
• update release checklist regarding crate releases #9701 — crate-release checklist update.

Automated incident issues may be used as evidence, but they are not separate roadmap workstreams unless they reveal an untracked root cause.

PR strategy

Aim for one substantive PR per coupled workstream, plus small follow-up PRs that only enable behavior after a proof gate passes. Do not split PRs by individual decision.

Urgent active failures such as #10957 should land independently rather than waiting for a broader refactor.

Tracking rules

• Every implementation PR must link its workstream issue.
• Every workstream must have an owner before implementation starts.
• Record GitHub settings changes in the relevant issue even when no repository PR is possible.
• Copy proof-period measurements into the workstream before changing enforcement.
• Add newly discovered existing issues here instead of creating duplicates.
• Do not close a workstream until the behavior is live and its acceptance measurements are verified.

Definition of done

• Every accepted decision maps to a workstream or existing issue.
• Every workstream has an owner and implementation PRs.
• Queue and CI proof gates have passed with recorded evidence.
• Required GitHub settings and repository workflows agree.
• Changelog fragments are the normal authoring path.
• A release completes successfully using readiness checks and resumable publication.
• Release artifacts are reproducible, signed, and promoted by immutable identity.
• CI and release documentation describe the final workflow.