name	render-check
description	Render-regression check for the resume renderer. Use BEFORE any refactor touching src/components, src/data/adapter.js, src/config, or the build setup (capture a baseline), and AFTER the change (diff against baseline — must be empty for pure refactors). Also use to verify intentional render changes.

render-check

Golden-reference render-regression loop built on scripts/capture.mjs. Each capture writes <prefix>.dom.html, <prefix>.pdf, <prefix>.png.

Authoritative signal: the DOM diff. Captures are deterministic for the same build (verified). PDF bytes ALWAYS differ between runs (embedded timestamps) — never byte-compare PDFs. PNGs are for human eyeballing only.

1. Capture baseline (before the change)

pnpm build   # output: apps/site/build
mkdir -p .render-baseline
node scripts/capture.mjs apps/site/build .render-baseline/resume

There is a single canonical résumé. The static build renders it from the bundled seed when no API is reachable. To check a tailored render, also capture an application=<id> fixture the same way.

2. Make the change, then capture current

Same into .render-current/resume.

3. Diff

diff -q .render-baseline/resume.dom.html .render-current/resume.dom.html \
  && echo "clean" || echo "DRIFT"

Pure refactor (restructure, dedupe, dependency bump): the résumé must be clean. Any drift = the refactor changed behavior — stop and investigate before proceeding.
Intentional render change: inspect the drift with diff .render-baseline/resume.dom.html .render-current/resume.dom.html and the PNGs side by side; confirm the change matches intent. Then re-baseline (step 1) so the next check starts clean.

Notes

.render-baseline/ and .render-current/ are gitignored artifacts.
Vite hashes asset filenames (/assets/index-<hash>.js); these appear in the DOM only via script/link tags outside #root, so they don't pollute the diff. If a hashed path ever shows up inside #root, normalize it before diffing rather than accepting the drift.
Run pnpm validate && pnpm test alongside this check — render-check covers pixels/DOM, not data contracts.

render-check

Golden-reference render-regression loop built on scripts/capture.mjs. Each capture writes <prefix>.dom.html, <prefix>.pdf, <prefix>.png.

1. Capture baseline (before the change)

pnpm build # output: apps/site/build mkdir -p .render-baseline node scripts/capture.mjs apps/site/build .render-baseline/resume

There is a single canonical résumé. The static build renders it from the bundled seed when no API is reachable. To check a tailored render, also capture an application=<id> fixture the same way.

2. Make the change, then capture current

Same into .render-current/resume.

3. Diff

diff -q .render-baseline/resume.dom.html .render-current/resume.dom.html \ && echo "clean" || echo "DRIFT"

Pure refactor (restructure, dedupe, dependency bump): the résumé must be clean. Any drift = the refactor changed behavior — stop and investigate before proceeding.

Intentional render change: inspect the drift with diff .render-baseline/resume.dom.html .render-current/resume.dom.html and the PNGs side by side; confirm the change matches intent. Then re-baseline (step 1) so the next check starts clean.

Notes

.render-baseline/ and .render-current/ are gitignored artifacts.

Vite hashes asset filenames (/assets/index-<hash>.js); these appear in the DOM only via script/link tags outside #root, so they don't pollute the diff. If a hashed path ever shows up inside #root, normalize it before diffing rather than accepting the drift.

Run pnpm validate && pnpm test alongside this check — render-check covers pixels/DOM, not data contracts.