Same axis name, different bank: stop before comparing

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

Two care cards name different frozen banks and different harnesses, so no difference between their numbers is a finding.

Two care cards should not be compared, and the bytes say why. On the care axis, Qwen/Qwen3-14B reads 0.1333 at n=30 (https://councilof.ai/interop/mill-cards-signed/signed-care-eda16a8fc969.json). Its evidence block names bank_file bank-care-3cf9c16dbb6b.jsonl from the csoai/gspc-care dataset and the route hf-router:Qwen/Qwen3-14B:featherless-ai, and its admission receipt records 4 correct of 30. The ollama:llama3.2:3b care card reads 0.7143 at n=77 (https://councilof.ai/interop/mill-cards-signed/signed-care-83d57579098c.json), and its compute_evidence binds a different bank_sha256, cee5e47a31f1, against a local Ollama model digest. Same axis name, different frozen bank, different harness. The two numbers are not on a common scale, so no difference between them is a finding. Before comparing any two cards, match bank_sha256 before anything else; if the hashes differ, stop. The same check applies across GET /api/hub-cards, where cells from both populations appear side by side.

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.