Vendor website benchmark · corrections
Public correction log
Every correction to published benchmark data, method text or a release is recorded here with its date, scope and what changed. An immutable snapshot is never rewritten: a data error gets a superseding snapshot and the old one is marked superseded. Categories are editorial, so a category change is logged but never moves a score.
Classes: data_publication_error (display or export wrong, measurement right), measurement_error (the scanner was wrong; affected rows re-measured), method_text (a description of the method was inaccurate; no values affected), category (editorial placement), identity (vendor name or domain).
| Date | Class | Scope | What was wrong | What changed | Values affected |
|---|---|---|---|---|---|
| 2026-09-18 | measurement_error | 5 directory rows (and 6 registry rows) | Grades were published for sites the scanner never reached - the homepage timed out or redirected in a loop on the apex and on www: fauna.com D/40, hippocratic.ai D/40, profound.dev D/40, robustintelligence.com D/40, milvus.io C/60 (plus blueheronrestaurant.com, registry only, since 2026-09-10). | An unreachable homepage is now recorded as no grade, with the reason (build 570). The six records were removed; the rows are re-measured by the directory job. | yes |
| 2026-09-16 | method_text | Dataset descriptions on /directory, /data and the snapshot metadata | The measurement technique text said each site is fetched by "fifteen named crawler identities and one unnamed client". | The scanner sends twelve named crawler identities plus an unnamed client and two mobile-browser controls; the text now says so in all three places. | none |