Research

Every figure here comes from running the open-source pseolint engine over a corpus of live sites, not from a survey or a vendor panel. The rules that produce those numbers are the same ones that run when you audit your own site, at the same thresholds, so a claim in a report can be reproduced by pointing the CLI at the same URL. Where a report cites a percentage, the denominator is stated inline rather than left implied.

Snapshots are dated and never silently revised. Search behaviour moves — a core update lands, a platform changes its default templates — so a re-measured figure ships as a new dated snapshot with the delta called out, and the original stays readable. The scoring model, per-template aggregation, and the severity demotions we apply are documented at /methodology, and the continuously-updated clean-corpus rankings are at /leaderboard.

The honest limit: this corpus is a sample, not a census of the web. It skews toward sites that publish programmatically and toward the platforms our users run, so treat cross-platform comparisons as directional and read the stated sample size before quoting a number. Where a finding rests on too few sites to generalise, the report says so instead of rounding it into a headline.