Make the required build-deploy check always report

[d-morrison]

Requested by Ezra · project thread

Before: preview.yml had a workflow-level paths: filter, so PRs touching only other files (workflows, lychee.toml, ...) never started build-deploy, and the required check stayed pending.

After: the paths: filter is gone.
A first step in the job diffs base...head and sets relevant using the same path set as the old filter; the setup, render, highlight, DOCX and banner steps run only when relevant == 'true' and the PR is not closed.
build-deploy therefore always reports, and finishes in seconds on irrelevant PRs.
The deploy step still runs on closed for every PR, so preview removal keeps working (it no longer needs the R/Quarto setup, which is now skipped on close).
A failing git diff fails the step instead of reading as "not relevant".

Considered Morrison-Lab/gha's reusable preview.yml: it is the build half of a build/deploy split (job build, deploy via workflow_run), so migrating would rename the required check and restructure the workflow.
That is a separate, larger change; this PR is the minimal fix the issue recommends.

How verified: actionlint clean; the relevance regex checked against sample paths (chapters, _quarto*.yml, root *.qmd/*.css, latex-macros match; lychee.toml, other workflows, inst/WORDLIST, nested *.qmd do not).
This PR itself touches only preview.yml, which is in the relevant set, so it exercises the full path.

Closes #97

🤖 Generated with Claude Code

https://claude.ai/code/session_01Gh7L9vJhwrpvpBhzYMdjPK

---

Generated by Claude Code

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[d-morrison]

OpenCode adversarial review (deepseek-v4-pro): round 1 findings (diff failure read as not-relevant; diff against merge HEAD) fixed in 80f5415.
Round 2 found one valid issue (the filter step must not run on closed, where the head commit may be gone and a diff failure would block preview removal), fixed in 4793198 with if: github.event.action != 'closed'.
Its second point (regex anchoring) is a false positive: every alternative sits inside the single ^( ... ) group.
latex-macros is a submodule gitlink, so git diff --name-only lists it as the file latex-macros, same as the old paths entry.
No round 3 run; the fix is a one-line guard checked with actionlint.

Posted by Claude Code (AI agent) --- not written by a human.

---

Generated by Claude Code

[github-actions]

PR Preview Action v1.8.1
🚀 View preview at https://Morrison-Lab.github.io/win/pr-preview/pr-117/
Built to branch gh-pages at 2026-10-09 09:36 UTC. Preview will be ready when the GitHub Pages deployment is complete.

[d-morrison]

/review

---

Generated by Claude Code

[d-morrison]

/review

Posted by Claude Code (AI agent) --- not written by a human.

---

Generated by Claude Code

[github-actions]

👀 Claude Review Dispatched — run reviewing PR #117 at 4793198.

[github-actions]

Claude finished review — View run

---

The current PR head commit (4793198... matching 47931986bdb043b1de1d12d905ba7d6e66bdfb10) is identical to the commit already reviewed in the prior round. No new commits have been pushed since that review.

Verdict

Ready for merge — no new findings. The prior review (comment from 2026-10-09T10:00:52Z, verdict: CLEAN) reviewed this exact commit (47931986bdb043b1de1d12d905ba7d6e66bdfb10), which is still the PR's current head — I confirmed via gh pr view that headRefOid matches. That review reconstructed and YAML-validated the patched preview.yml, traced every conditional path (opened/synchronize/labeled with relevant true/false, and closed), verified the relevance regex against sample paths, confirmed bash -eo pipefail semantics for the git diff failure-propagation claim, and found no AI-hallucination concerns (all referenced actions — rossjrw/pr-preview-action, quarto-dev/quarto-actions, r-lib/actions, marocchino/sticky-pull-request-comment — are real and unchanged). I independently re-read the saved diff in this round and confirm it is byte-for-byte the same change already assessed; that prior verdict stands.

No R/Quarto content or LaTeX math is touched by this diff, so the SERG lab-manual coding-style/practices checks and the latex-macros macro-opportunity checks remain not applicable, as previously noted.

Structured Review Data (JSON)

{
  "schema_version": "1.1",
  "reviewer": "claude",
  "commit_sha": "47931986bdb043b1de1d12d905ba7d6e66bdfb10",
  "verdict": "CLEAN",
  "findings": [],
  "detailed_assessment": "No new commits since the prior round; preview.yml diff at this head is byte-identical to the already-validated version, so no new failure modes were introduced.",
  "holistic_assessment": "Scope remains limited to restoring always-reporting required-check behavior for preview builds without regressing deploy or removal integration paths."
}

Reviewed commit: 4793198

[github-actions]

💰 Cost: $0.1772 (review) — run