Docs platforms
One row per docs platform we can identify. Every mean is over the hosts we could measure, and the share we could measure sits beside it.
A cohort we could measure on fewer than 80% of its hosts is shown but never ranked. The reason is never the score: a mean over a tenth of a cohort is not comparable with a mean over a census.
Ranked
| platform | hosts | measured | share | mean | sd |
|---|---|---|---|---|---|
| mintlify | 638 | 619 | 97.0% | 63.6 | 9.8 |
| fern | 75 | 74 | 98.7% | 53.8 | 7.8 |
| gitbook | 548 | 546 | 99.6% | 44.4 | 8.5 |
| docusaurus | 653 | 646 | 98.9% | 38.2 | 10.5 |
Shown, not ranked
These cohorts stay on the page because a cohort that vanishes is indistinguishable from a vendor we never scanned. What their row reports is our own measurement failure, not the vendor's quality.
| platform | hosts | measured | share |
|---|---|---|---|
| readme | 656 | 64 | 9.8% |
Method
11152 hosts carry no platform we recognise. Our fingerprint table has five entries and misses MkDocs, Redocly, Slate and Stoplight, so that number describes the table, not those sites.
Association, not causation. These are the sites we scanned on each platform, not a controlled comparison, and nothing here supports "move to platform X and gain N points".
Measured under rubric v5, among the sites in our corpus — not a sample of the web.