A zero with a missing error field is a question, not a finding

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

A MEASURED card reads accuracy 0 at n=237 and carries no parse-error count; its neighbour on the same bank does.

ollama:qwen3:4b reads accuracy 0 at n=237 on the governance axis (https://councilof.ai/interop/mill-cards-signed/signed-governan-3c96b82c7e64.json), and the body says MEASURED. On the same bank_sha256, b93f9808f014, the ollama:qwen3:8b card reads 0.5865 at n=237 (https://councilof.ai/interop/mill-cards-signed/signed-governan-09759e29f44b.json). The two compute_evidence blocks differ in one telling way: the qwen3:8b card records parse_errors_excluded: 0, and the qwen3:4b card has no parse_errors_excluded field at all. The qwen3:4b run_id is dated 2026-09-05; the qwen3:8b run_id is dated 2026-09-14. A zero on a 237-item bank with the error field absent is something to investigate before anyone repeats it as a statement about the model. We publish the observation and not an explanation, because the card bytes do not contain one. Read any zero alongside its error fields, and treat a missing field as missing, not as zero.

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.