test(visual): stop content and API drift from repainting baselines

[elevatebart]

Three things moved the screenshot baselines, and only one of them was the CSS the suite exists to watch. Blog-Index has the most refreshes in the snapshot history, and none of them were about layout.

Dated content

SNAPSHOT_CUTOFF, a new optional server env field, filters every date-sorted blog list the sampled pages can render: WhatsNew.astro and blogs/[category].astro (blogs, external blogs, tutorial videos), fetchSecondaryData behind the /plugins/core* pages, and the orchestration fallback list. beforeCutoff is a pass-through when the var is unset, which is everywhere except the snapshot workflow, so the site keeps rendering its full content.

The workflow sets it to 2026-09-17. A date-only value resolves to midnight UTC, so the one post dated 2026-09-17T10:00:00 falls outside it and the sample runs current through 2026-09-16T13:00. Pass a full timestamp if you want a day inclusive.

Live API

/plugins, /blueprints and the community metrics are read at request time, so a plugin release repainted them. The snapshot workflow now builds and screenshots against scripts/api-fixture-server.mjs, which Lighthouse already used, with its own FIXTURE_PATHS adding the endpoints the sampled SSR routes need: /plugins/definitions/*, /plugins/metadata*, /plugins/pluginsInformation*, /plugins/artifacts/*, /plugins/*/versions, /versions, /communities/github/*.

isRecorded grew an inner wildcard matching a single path segment, so /v1/plugins/<fqcn>/versions can be recorded without /v1/plugins* swallowing the schema fetches sitting next to it. Trailing-* prefixes behave exactly as before.

Anything recorded during a run is committed with the refreshed baselines. workflow_dispatch gained a refresh_api_fixtures box; ticking it clears the whole directory, including the ~300 blueprint detail responses only Lighthouse reads, which its next run re-records into an artifact to commit. Called out in the input description and the fixtures README.

Incidental content

/docs renders the four newest posts through WhatsNew, where the blog list is not what that baseline is for. A stylesheet injected via stylePath pins the cards to their thumbnail height and blanks the copy, so bumping the cutoff moves Blog-Index and leaves Docs-Landing alone.

Keeping the cutoff honest

The date only moves when someone moves it, so the run reports its age to the job summary, warns past 90 days and fails past 180. Both windows are job env vars. The check runs last, so a neglected cutoff still gets its baselines refreshed and committed before the job goes red.

Verified

Dispatched on this branch (run 35364764388): the fixture server served 360 replays with zero errors, proving the SSR routes under wrangler dev reach it, and recorded the 11 endpoints the widened paths added. 13 baselines moved, and the four Plugin-* pages did not, the recorded data matching what was live. Local builds with the cutoff set come back clean, server bundle included. 422 unit tests pass, oxlint clean.

What to expect

Quiet nightlies until the CSS actually changes or someone bumps the date, and a nag on the run once the date ages. The trade is that the sampled content ages with it, so a card redesign that only triggers on newer frontmatter would not show up until the next bump.

🤖 Generated with Claude Code

https://claude.ai/code/session_01GHRsDvXNS9LLTVhdmjA8or

[github-actions]

☁️ Cloudflare Worker Preview Deployed!

🔗 https://ks-elevate-bot-inspiring-rubin-iz-docs.kestra-io.workers.dev
🔗 https://859373e3-docs.kestra-io.workers.dev

🔦 Lighthouse Benchmark

Tested on 2026-09-23 09:34 UTC  ·  links point to https://ks-elevate-bot-inspiring-rubin-iz-docs.kestra-io.workers.dev
Compared against main baseline from 2026-09-23
Runner CPU index per shard: 4254, 2484, 2625, 2501 (baseline: 2496, deltas hidden on 10 page(s))

Scores (0–100, higher is better)

Page	Performance	Accessibility	Best Practices	SEO
Home	96	84	77	92
Get Started	95	83	77	100
Pricing	96	92	77	100
Enterprise	97	83	77	100
Cloud	93	81	77	100
Features	93	92	77	92
About Us	97	91	77	100
Docs Landing	89	88	77	92
Contribute to Kestra (simple docs)	97	88	77	92
Flow (full featured docs)	93	90	77	92
Blog Index	99	91	77	100
Blog Post (sample)	96	88	77	100
VS Page (sample)	97	88	77	100
Plugins Landing	96	81	77	92
Plugin Page (sample)	95 🟢 +13	88	77	100
Plugin Debug Page (sample)	96	88	77	100
Plugin Debug Return Page (sample)	79	88	77	100
Blueprints Landing	85	91	77	100
Blueprint Audit Logs CSV Export	85 🔻 -10	92	77	100

Core Web Vitals (lower is better)

Page	LCP	FCP	TBT	CLS	Speed Index
Home	1.21 s	0.58 s	15 ms	0.000	1.29 s
Get Started	1.46 s	0.61 s	2 ms	0.000	1.03 s
Pricing	1.04 s	0.64 s	132 ms	0.000	1.04 s
Enterprise	1.21 s	0.57 s	0 ms	0.003	1.02 s
Cloud	1.60 s	0.61 s	99 ms	0.049	0.91 s
Features	1.16 s	0.75 s	151 ms 🔻	0.000	1.17 s
About Us	1.16 s	0.64 s	60 ms	0.000	1.01 s
Docs Landing	1.20 s	0.57 s	215 ms	0.001	1.27 s
Contribute to Kestra (simple docs)	1.10 s	0.57 s	78 ms	0.000	0.98 s
Flow (full featured docs)	1.05 s	0.69 s	173 ms	0.000 🟢	1.24 s
Blog Index	0.95 s	0.54 s	0 ms	0.000	0.80 s
Blog Post (sample)	1.29 s	0.60 s	83 ms	0.004	0.75 s
VS Page (sample)	1.14 s	0.58 s	56 ms	0.000 🟢	1.10 s
Plugins Landing	1.03 s	0.63 s	129 ms	0.000	1.13 s
Plugin Page (sample)	1.01 s	0.69 s	155 ms 🟢	0.000	1.12 s
Plugin Debug Page (sample)	0.88 s	0.57 s	140 ms	0.000	0.97 s
Plugin Debug Return Page (sample)	0.97 s	0.57 s	402 ms	0.000	1.47 s
Blueprints Landing	2.35 s	0.78 s	84 ms	0.000	1.48 s
Blueprint Audit Logs CSV Export	1.02 s	0.70 s	253 ms 🔻	0.000	1.88 s

Legend

🟢 improved  ·  🔻 regressed  ·  (blank) no significant change
Score threshold: ±10 pts  ·  Metric threshold: ±30% of baseline

Median of repeated runs: Home x5, About Us x3, Docs Landing x5, Flow (full featured docs) x3, Plugins Landing x3, Blueprints Landing x3. A single run of these swings 20+ points between runners.
The sample is measured across 4 parallel runners, so each page carries its own CPU index (Lighthouse's benchmarkIndex). Lighthouse does not normalise for host CPU, so a page's deltas are hidden when its runner differs from the baseline's by more than 10%.

View full Lighthouse HTML report for a page

Full per-page Lighthouse Results (LHR) are attached as the lhr-reports-shard-* artifacts on this run. Download and unzip one, then open https://googlechrome.github.io/lighthouse/viewer/ and drop the <page>-lhr.json file into the page to see every audit, opportunity, and diagnostic.

[iitzIrFan]

Nice work. This goes after the actual causes of baseline churn, and it fits the existing setup: it reuses the fixture server and startup steps from lighthouse.yml, declares SNAPSHOT_CUTOFF like the other env settings, and puts snapshotCutoff.ts next to its test in the same style as the other utils. I checked out the branch: the new path matching handles exact, trailing-* and inner-* paths correctly (/v1/plugins/core is still not recorded, /v1/plugins/<fqcn>/versions is), and the new tests pass.

Approving. The inline notes are non-blocking suggestions, fine here or in a follow-up.