Review

Where models disagree

Records with commentary, those that split the panel first. A split means two models gave the same claim different stances; the record page shows each position with its source.

No record carries model commentary yet.

Human coverage

Named clinicians, scientists, regulatory specialists and patient advocates sign pages off on top of the model panel. A human review is a statement about a page on a date and is the only review Nuclide treats as clinical.

Companies, institutions, people, collections and journals are attributed (who wrote the record) rather than reviewed; organisations correct their own records through Suggest an edit.

The human queue

Score out of 100: reach (graph connections, log-scaled, up to 40), stakes (evidence score or a kind default, up to 30), staleness (days since the record's asOf, a year is 30). Pages already reviewed on every track they need within a year sort to the bottom.

Showing the top 50 of 592 reviewable pages.

By track

The ten highest-need pages for each track, so a clinician, a scientist, a regulatory specialist or an advocate can start without reading the whole queue.

Translations

Every translated TL;DR is machine-assisted until a named speaker of the language reviews it. Pages show MT until then and Reviewed after. Reviews are recorded in src/data/i18n/reviewed.ts.

Spotted a wrong translation on one page? Use Translation fix; to sign off a batch, Translation review.