A performance budget that blocks the merge

Most teams treat performance as a discipline problem: everyone agrees the site should be fast, and it degrades anyway, one convenient dependency at a time. I moved the standard out of the review conversation and into the pipeline, where it cannot be forgotten or argued with under deadline.

Every pull request against driftpilot.ca runs Lighthouse CI three times against a production build and asserts the median. A failure is a red check, not a comment.

What the gate caught

The gate earns its keep by catching what review misses. It stopped a WebGL shader path that hit 39 seconds of total blocking time on software renderers, a footer contrast pair below WCAG minimum, and a third-party scheduling embed that quietly blew the script budget.

One budget was re-baselined during the project: the original 110 kB script ceiling was fiction against a framework-plus-shader reality measured at 237 kB, so I moved the number deliberately and wrote down why. Measurement can justify moving a threshold. Silently deleting a failing assertion to get a green check cannot — that distinction is the whole value of having the gate.

The rest of the build

37 statically prerendered routes with no runtime database. Content sits behind typed async accessors, so a headless CMS can replace the source without editing a page or a component.

The lead pipeline runs Zod-validated Server Actions through honeypot and time-to-submit spam gates into a retrying webhook client that replays 5xx and 429 and treats 4xx as permanently failed — verified end to end on the live site. 88 commits across 21 reviewed pull requests.