Performance Release Gate
21 items · source
Thresholds
- Verify the p75 field LCP is under 2.5 seconds on mobile, or that the exception is written down with an owner and a date.
- Verify the p75 field INP is under 200 milliseconds.
- Verify the p75 field CLS is under 0.1.
- Verify the lab Lighthouse performance score on mobile meets the target agreed for this project, measured as a median of several runs.
- Verify the Lighthouse accessibility score is 100, or that every remaining item is a known false positive with a note.
- Verify the Lighthouse best-practices and SEO scores are at target.
- Verify these numbers were measured on the templates that carry traffic, not only the homepage.
No regressions
- Verify the performance budget is not exceeded by this release.
- Verify no new render-blocking resource was added to the critical path.
- Verify no new third-party script was added without an owner and a measured cost.
- Verify bundle size did not increase without a deliberate decision recorded in the pull request.
- Verify the LCP element on each key template is still what it was, and still discoverable in the HTML.
- Verify no image above the fold gained
loading="lazy". - Verify the CI performance check ran on this commit and passed.
When something is out of your hands
- Verify a third party degrading the score has been quantified, so the conversation with the vendor has a number in it.
- Verify a facade, a delayed load or a self-hosted copy has been considered before accepting the cost.
- Verify the decision to ship despite a failing metric is recorded with who made it and when it is revisited.
- Verify field metrics are watched for the week after release, tied to this deploy.
Coverage
- Verify the gate ran against mobile emulation with throttling, not desktop on a fast connection.
- Verify the logged-in experience was measured if it is where users spend their time.
- Verify the numbers in this gate are stored with the release, so the next regression has something to compare against.
# Performance Release Gate ## Thresholds * [ ] Verify the p75 field LCP is under 2.5 seconds on mobile, or that the exception is written down with an owner and a date. * [ ] Verify the p75 field INP is under 200 milliseconds. * [ ] Verify the p75 field CLS is under 0.1. * [ ] Verify the lab Lighthouse performance score on mobile meets the target agreed for this project, measured as a median of several runs. * [ ] Verify the Lighthouse accessibility score is 100, or that every remaining item is a known false positive with a note. * [ ] Verify the Lighthouse best-practices and SEO scores are at target. * [ ] Verify these numbers were measured on the templates that carry traffic, not only the homepage. ## No regressions * [ ] Verify the performance budget is not exceeded by this release. * [ ] Verify no new render-blocking resource was added to the critical path. * [ ] Verify no new third-party script was added without an owner and a measured cost. * [ ] Verify bundle size did not increase without a deliberate decision recorded in the pull request. * [ ] Verify the LCP element on each key template is still what it was, and still discoverable in the HTML. * [ ] Verify no image above the fold gained `loading="lazy"`. * [ ] Verify the CI performance check ran on this commit and passed. ## When something is out of your hands * [ ] Verify a third party degrading the score has been quantified, so the conversation with the vendor has a number in it. * [ ] Verify a facade, a delayed load or a self-hosted copy has been considered before accepting the cost. * [ ] Verify the decision to ship despite a failing metric is recorded with who made it and when it is revisited. * [ ] Verify field metrics are watched for the week after release, tied to this deploy. ## Coverage * [ ] Verify the gate ran against mobile emulation with throttling, not desktop on a fast connection. * [ ] Verify the logged-in experience was measured if it is where users spend their time. * [ ] Verify the numbers in this gate are stored with the release, so the next regression has something to compare against. 21 items · https://github.com/FarzamHabibi/pre-production-checklist · CC BY 4.0