Measurement, not certification
Measured models
Every model we have run against the signed card set, ranked by how much of it we actually measured. Each filled cell is one model on one axis on one date, recorded in a file stamped so that anyone can confirm offline that it has not been edited since. Most cells are empty, and the empty ones are shown.
About the date. the newest body.created stamp across the cards — when the last card was measured, never when this file was generated. It is not the time this page was rendered. Newest record: 2026-08-19.
You are looking at the signed card set. It is a benchmark corpus of small model runs. These are not the public board's axes: the board measures governance behaviour with a different instrument, over a different population, and carries its own separate count. The two sets are never added together. See how every set relates.
- Models measured
- 61
- Axes in this set
- 16
- Cells filled
- 313
- Cells with a signature
- 313
each run against at least one axis
benchmark axis, not board axis
of 976 possible pairs
each re-checkable offline
Most of this matrix is empty, and that is the honest part. One model measured on one axis on one date, recorded in a card whose signature covers its own body. An empty cell means that pair was never measured — it is not a zero. Coverage is 313 of 976 pairs — see the coverage map below.
What this set does NOT establish
That a model is good, or better than another. A cell is one score on one small bank on one date. Several of these banks are small enough that a single item moves the number visibly, most cells in the matrix are empty, and a score of zero is a measured zero rather than a missing measurement.
THESE ARE NOT THE PUBLIC BOARD'S AXES. The cards carry benchmark axes; the public board carries governance axes measured by a different instrument. The two sets are different, their counts are different on purpose, and they are never added together. The board's count authority is GET /api/gspc.
Check any of it yourself
Every cell above links to its own record. Open one, recompute the hash of its body, and check the signature against the published key — nothing is sent to us and nothing needs our permission.
1 of the recorded model names is a retired internal brand this site does not publish, so it is listed under a neutral label. In the card's own body, under the signature. The card is the evidence and it was not edited — editing it would invalidate every id downstream of it.