Research / AI Experiments

A record of the experiment as it develops.

Fresh data, a consistent reporting format, and a dated archive of regular reports and public checkpoint releases. Research observations stay separate from interpretation and conclusions.

Two checkpoints, with the missing evidence visible.

35/48 responses and 21/84 completed automated evaluations. Thirteen missing samples and unresolved review flags remain explicit.

Read the public checkpoint report →

More evidence, uneven coverage, and an open review queue.

289 responses across four cohorts: 241 added since the baseline. The report tracks 47 missing samples and seven evaluations awaiting human review.

Read November report →

Six providers. Two checkpoint days. An evidence trail to follow.

48 stored responses, eight completed observation slots, and two evaluations awaiting human review. The first report establishes the starting point.

Read baseline report →

Next report: December 1, 2026. Reports recur every 30 days, even when a window contains no new checkpoint. Each report includes the question, method, observations, changes since the prior report, review status, limitations, practical implications, and next steps.