CORRECTIONS

Corrections

Corrections protect public trust, not institutional reputation. Material corrections are documented openly rather than silently replacing previous content.

Published corrections remain part of the institutional historical record. Ordinary editorial changes are distinct from material corrections.

Public correction record

September 29, 2026

Homepage preservation, archive count, and language coverage clarified

The absolute statement that every observation is preserved was replaced with wording about context, contributor choices, and review and publication. The homepage now reads the public archive count using its existing approved-record exclusions and privacy threshold, with a count timestamp and an unavailable state on query failure. The footer links to Language Integrity and distinguishes the 15 active language choices from the broader 20-language program. No scientific or observation records were changed.

Affected content: Homepage, footer, and Language Integrity page

clarification

September 1, 2026

Multilingual corpus accounting superseded

The August 22, 2026 reconciliation reported a corrected total of 12 corpora / 23,808 units. That figure has since been superseded. Current governed automated-assurance reporting covers 18 Core Global 20 corpora and 36,126 translated units, of which 35,655 pass automated assurance. The August 22 entry remains part of the public record and is not modified. The total returned to 18 later on August 22 because governed corpora were then generated for the six canonical Core Global 20 locales that had never had one — Bengali, Marathi, Nigerian Pidgin, Tamil, Telugu and Urdu — completing the roster, rather than by restoring the six non-roster corpora (Persian, Italian, Dutch, Polish, Ukrainian and Traditional Chinese) that the reconciliation had correctly removed.

Affected content: Public multilingual coverage and assurance accounting

material

August 22, 2026

Multilingual corpus accounting reconciliation

Multilingual corpus reporting was reconciled to the canonical Core Global 20 language roster. Six generated corpora associated with a superseded runtime-language roster were removed from Core-20 assurance totals. The reported automated-assurance total changed from 18 corpora / 35,712 units to 12 corpora / 23,808 units at the time of reconciliation. No underlying scientific evidence was deleted; the correction changed language classification and public assurance accounting so Core-20 reporting reflected only corpora belonging to the canonical public language roster at that time.

Affected content: Public multilingual coverage and assurance accounting

material

What counts as a correction

Material factual, scientific, statistical, methodological, regulatory, or interpretive changes receive correction treatment. Ordinary copyediting does not automatically constitute a public correction.

How corrections are reviewed

Reported issues are reviewed by the appropriate institutional team. Scientific questions may receive additional scientific review. Individuals with a conflict of interest do not review their own work.

Report an issue

Anyone may report a suspected material error for institutional review.

Report a correction