Production lab evidence for roadmap item P1.5.
2026-07-28 baseline
- Target: `https://www.metamuseum.org`
- Tool: Chrome DevTools performance trace
- Load mode: cold navigation with reload
- Network: Fast 4G
- Mobile: 412 x 915 at 2.625 DPR, touch/mobile emulation, 4x CPU slowdown
- Desktop: 1440 x 900 at 1 DPR, 1x CPU
- Budgets: LCP <= 2.5 s, CLS <= 0.1, representative INP <= 200 ms
- Field data: CrUX did not return page-level field metrics for the audited routes.
| Journey | Mobile LCP | Mobile CLS | Desktop LCP | Desktop CLS | Result |
|---|---:|---:|---:|---:|---|
| Home | 600 ms | 0.00 | 359 ms | 0.00 | Pass |
| Explore | 2,286 ms | 0.00 | 1,550 ms | 0.00 | Pass |
| Artwork `61658` | 2,154 ms | 0.00 | 1,326 ms | 0.00 | Pass |
| Projects | 771 ms | 0.00 | 360 ms | 0.00 | Pass |
| Pilot | 730 ms | 0.00 | 382 ms | 0.0665 | Pass |
A representative consent-control interaction trace measured 28 ms INP. This is
synthetic lab evidence, not a substitute for a route-level field distribution.
Findings
- Explore mobile is the limiting route at 2,286 ms LCP, leaving 214 ms of
budget headroom. Treat a regression here as the first performance priority.
- Pilot desktop produced the only non-zero CLS, a single 0.0665 shift cluster.
DevTools identified no potential root cause and estimated no savings.
- The Pilot LCP image was discoverable in the initial document and was not
lazy-loaded. DevTools noted that it lacked `fetchpriority=high`, but estimated
no LCP savings, so this is not a recommended change from this run.
- Render-blocking and cache insights that appeared on otherwise fast routes
estimated 0 ms FCP/LCP savings and are not remediation priorities.
- No audited page breached the current P1.5 lab budgets.
Remaining P1.5 evidence
the same profile so results remain comparable.
- Repair and repeat the two failed visual checks documented below.
- Repeat this baseline after a material frontend or deployment change; retain
Step 6 visual and accessibility checkpoint
The July 28 follow-up reviewed the same five production journeys at 412 x 915
and 1440 x 900 after choosing essential-only privacy storage.
critical axe violations. `/agents` and `/automation` correctly enforced their
authentication redirects.
overflow at either viewport.
decoded intrinsic dimensions.
viewports (`405/397` mobile and `1433/1425` desktop), producing horizontal
overflow.
rendered at up to `397 x 240` mobile and `835 x 416` desktop in the hero.
Its repeated source card is also enlarged to `363 x 272` mobile and
`345 x 259` desktop.
- `pnpm a11y:check` passed all 18 routes with zero moderate, serious, or
- Explore, artwork detail, Projects, and Pilot showed no document-width
- Artwork detail and Pilot did not render their principal images above the
- The initial Home review found two visual acceptance failures:
- `.home-source-band` is 8 px wider than the document client width at both
- The selected V&A Chair source is only `77 x 100` decoded pixels but is
The July 28 local remediation moves the source band outside the padded
`.container` and uses normal `width: 100%` layout, removing the
scrollbar-sensitive `100vw` calculation. Image selection now prefers a declared
IIIF Image API service and requests a 1,200 px derivative before considering
thumbnail representations. At a 1280 px local viewport, document and body
scroll widths both equal the 1265 px client width, the band begins at x=0 and
matches the client width, and the active V&A source is
`full/1200,/0/default.jpg`.
Focused page, selection, and layout tests pass, as does the production build.
The local post-fix verification then repeated the complete matrix:
| Journey | Mobile LCP | Mobile CLS | Desktop LCP | Desktop CLS | Result |
|---|---:|---:|---:|---:|---|
| Home | 703 ms | 0.00 | 356 ms | 0.00 | Pass |
| Explore | 1,628 ms | 0.00 | 1,250 ms | 0.00 | Pass |
| Artwork `61658` | 687 ms | 0.00 | 312 ms | 0.00 | Pass |
| Projects | 695 ms | 0.00 | 329 ms | 0.00 | Pass |
| Pilot | 789 ms | 0.00 | 360 ms | 0.00 | Pass |
All five journeys pass mobile and desktop document-width checks. The Home
source band matches the client width, the V&A card uses the 1,200 px IIIF
derivative, and artwork/Pilot principal images remain within their intrinsic
dimensions. The 18-route axe gate passes with zero moderate, serious, or
critical violations. A representative Home carousel interaction measures
29 ms INP with 0.00 CLS.
The retained result is
`artifacts/performance/frontend-p15-local-verification.json`. P1.5 remains open
only for the same capture against the deployed revision; local implementation
and verification are complete.