Queued CI/CD Adoption
Brings operations, common-module, and infrastructure onto the Queued CI/CD model already running in documentation and arda-frontend-app: a GitHub merge queue with ALLGREEN grouping, changelog entries authored in the PR body, and post-merge assembly into CHANGELOG.md.
The three target repositories differ from the two live ones on one structural point that shapes the whole project: CHANGELOG.md is their build-time version source. The adaptation inverts the order — assemble the release block first, then build and publish from the assembly commit.
Status
Section titled “Status”| Aspect | Value |
|---|---|
| Status | In progress — operations is live on the merge queue |
| Started | 2026-08-04 |
| Repos affected | operations, common-module, infrastructure, qualify-build-action, gradle-build-pipeline-action, synthesize-changelog-entry, cicd-testbed |
| Owner | Miguel (jmpicnic) |
| Parent ticket | PDEV-1408 |
Documents
Section titled “Documents”- goal.md — purpose, scope, repositories, success criteria.
- requirements.md — the numbered requirements the design implements, each traced to its source, with the open ones called out.
- design.md — the adapted pipeline: structure, behavior, verification, operations impact.
- decision-log.md — the decisions that shaped the design, with rejected alternatives.
What the rollout changed about the design
Section titled “What the rollout changed about the design”The design survived first contact only in outline. Rounds 4 to 6 of the decision log were driven by production behaviour rather than design review, and three of their findings changed shared components rather than this project’s own code:
- No component owned exactly one decision (DQ-008). Entry resolution was implemented three times across two repositories, and the copies disagreed. It is now one component,
synthesize-changelog-entry. - An empty input could not mean “none” (DQ-010), so every ordinary build fell back to a derivation that rejects exactly what a merge queue stages.
- The component’s contract omitted its hardest caller (DQ-011), which assembly worked around by reaching into its implementation.
Each was found by running the pipeline, not by reading it. Two had passed review.
Prior art
Section titled “Prior art”- Queued CI/CD — the canonical specification of the target model.
- Phase 2 — Queued PRs and Tiered Gates — the
arda-frontend-approllout, completed 2026-04-28. - Frontend PR Process — the contributor how-to that will need a backend counterpart.
Copyright: (c) Arda Systems 2025-2026, All rights reserved
Copyright: © Arda Systems 2025-2026, All rights reserved