tracking: re-evaluate coverage reporting / badge after OSS launch #133

Status: deferred until repo flips public.

What was removed

The repo previously published a coverage SVG via a badges branch (generated by vladopajic/go-test-coverage's Docker action, plumbed through a workflow_run-triggered workflow) and referenced it from README. That setup was removed in #129 because:

• The badge image URL pointed at raw.githubusercontent.com, which requires authentication for private repos. Anonymous viewers saw a broken image.
• The badges branch existed solely to side-step the ruleset's block-direct-pushes-to-main rule. Once the badge wasn't serving its purpose, the whole apparatus was overhead.

Coverage data is still computed by make ci and published in the per-run job-summary panel (the "Test Coverage" block). See .github/workflows/ci.yml's "Coverage summary" step and .testcoverage.yml for the threshold spec consumed by scripts/cov.

Decision to make at OSS launch

Two reasonable paths once the repo is public:

1. Resurrect the old setup. The existing badges-branch publish workflow lives in git history (ci: kill coverage Docker pull on PR runs + tighten claude-review group #129's commit predecessors) — could be cherry-picked back. README badge URL would just start working because raw.githubusercontent.com is unrestricted for public repos. Free, no vendor dep, only as fancy as a static SVG.

2. Codecov (or similar SaaS). Provides more than a badge: PR-comment coverage diffs ("this PR drops coverage by 2.3%"), sunburst graphs, trends, file-level annotations on the PR diff. Industry standard for public Go projects. Costs: vendor account, workflow integration, one more thing to manage. Free for OSS. Trust caveat: Codecov was the target of a real supply-chain compromise in April 2021 (bash uploader exfiltrated CI secrets for ~2 months); they've hardened since but it's a known incident worth weighing.

Inputs for the decision

• How much we end up caring about coverage diff visibility on PRs versus an absolute number on a README. PR-diff is genuinely useful on bigger contributor pools.
• Whether we trust the Codecov pipeline post-2021 (they're widely used today, so general industry verdict is "yes, with reasonable precautions").
• Whether we want a single dashboard or are fine with per-run summaries.

What this issue should produce

A short ADR (in docs/adr/ or wherever feels right) capturing the chosen approach and the rationale, plus a PR implementing it. Block this issue on the OSS-launch milestone — there's no point deciding this before we have to.

Related

• ci: kill coverage Docker pull on PR runs + tighten claude-review group #129 — removed the previous coverage badge / publish workflow.
• docs(ci): mermaid renders broken (Playwright missing) + missing Astro build cache #132 — separate docs-build issue (Playwright mermaid + Astro cache), filed alongside this one.