Measured, not promised.

01 · Method

How the numbers are made.

web-vitals beacon → nightly percentile rollup → p75 over a 28-day window

Every real visit to a gallery or to this site runs Google's open-source web-vitals library and beacons its Core Web Vitals back when the tab closes: LCP, INP, CLS, TTFB and FCP. A nightly job folds the raw beacons into daily percentiles per surface; days with fewer than 20 samples for a metric are discarded as noise rather than published. What renders below is the sample-weighted 75th percentile across the window - the experience the slower quarter of visits got, which is a harder number to flatter than an average.

The beacon is anonymous by construction, not by promise: no cookies, no IP addresses, no user agent strings, no gallery, visitor or session identifiers are collected or stored - each beacon is four fields (metric, value, surface class, mobile-or-desktop) and nothing else, so two visits from the same person are indistinguishable from two strangers. Our own admin traffic is excluded so staff on fast machines can't flatter the numbers, and raw beacons are deleted after 30 days once rolled up.

02 · The numbers

Collecting.

below the sample floor - methodology only, no numbers

The beacon is live and collecting from real visits right now. Below the floor, percentiles are noise dressed up as precision - one slow hotel-wifi page load would swing the whole chart - so we publish nothing rather than something shaky. No extrapolation, no seeded numbers, no lab runs standing in for real traffic while we wait.

Receipts publish once this window holds 500+ real-visit measurements.

When they do, this section shows p75 LCP, INP, CLS and TTFB for galleries and for this site, each against its Core Web Vitals threshold, with a 28-day daily trend - automatically, the hour the floor clears.

03 · What we don't do

The fine print, plainly.

no lab cherry-picks · no synthetic "real" traffic · no per-visitor tracking

No lab-run cherry-picks: a Lighthouse score from a warm datacenter says little about a phone on venue wifi, so we don't publish hand-picked lab runs as if they were your clients' experience. No synthetic numbers presented as real traffic: if the floor isn't met, the page says so instead of inventing a chart. And when a metric misses its threshold, it stays on the page without a badge - the miss is part of the record.

No per-visitor tracking, ever: the beacon cannot be joined to a person, a gallery, or a session, because the identifiers needed to do that are never collected in the first place. Faster pages are not bought with your clients' privacy - that trade isn't on the table.