Detector-interop: a two-way tie at n=33

2026-09-14 · Council of AI (CSOAI Ltd)

Two MEASURED cards on the same detector-interop bank read the same accuracy; the bank does not separate them.

Two signed pod cards on the detector-interop axis bind the same bank_sha256, d75e96802d9e. ollama:gemma3:4b reads 0.7879 at n=33 (https://councilof.ai/interop/mill-cards-signed/signed-detector-dff95f98a04a.json). ollama:mistral:7b reads 0.7879 at n=33 (https://councilof.ai/interop/mill-cards-signed/signed-detector-e48e1b64d89f.json). That is 26 correct of 33 on each card. Both bodies say MEASURED, with zero parse errors and zero transport errors excluded. On this bank, on this date, the two models tie. A tie is a result in its own right: it tells a reader that this bank does not distinguish these two models, which is worth knowing before treating either one as ahead on this dimension. Neither card needs a separation test for the tie to be stated honestly, and neither carries one. Both cards commit to an items_sha256 without linking the item rows, so the check a stranger can run today is the content id and the Ed25519 signature, using the verify link each card names.

How to verify: fetch the artifacts listed above yourself, and check a signed card against the published key at https://councilof.ai/gspc-verify/.

Measurement, not certification. Not a grade, endorsement or legal finding.