FIELD TESTS

Field Tests

Pre-registered checks of other people's AI systems.

Each field test is written down and hashed before the first request is sent. The hash goes to a public ledger, so the plan cannot be changed after the results are seen. This page lists every test, including any that never ran.

TALLY · UPDATED 2026-10-13

1 listed · 1 run · 0 with published results

Tests

  • FIELD TEST 001 · STATUS: RUN

    Jev Calibration

    System under test: TypeSafe AI, Jev model

    Run once, privately, on October 13, 2026, exactly as registered. TypeSafe’s terms ask for its consent before either party publicizes the relationship. Consent was requested in writing on September 30, October 6, and October 10. On October 12 TypeSafe support replied that it does not have the capacity to review or approve specific use cases, which is neither a yes nor a no. The results are sealed: each run’s response file, the analysis script, and the results were hashed to the ledger as they were produced. They are published only if consent or clear legal footing arrives, whatever they show. If neither ever does, they are never published.

    Pre-registered 2026-09-30 · Hedera mainnet, topic 0.0.9376001, sequences 1 through 7

What the statuses mean

pre-registered
The protocol is frozen and its hash is on the public record. No request has been sent to the system under test.
consent requested
The system’s terms require consent before results can be published, and consent has been asked for in writing. No answer yet.
consent refused
The vendor said no. The test is not published beyond this line. It stays on this list with the date and reason.
run
The test has been run under the frozen protocol. Results are not yet published.
results published
Results, item files, responses and analysis code are public, whether they favor the system or not.
withdrawn
The author stopped the test before completion. The reason is stated here.

Standing rules

  • Results are published whether they favor the system under test or not, subject only to any consent the system's terms require.
  • A test that is refused, withdrawn, or never run stays on this list with its status and a one-line reason. Nothing is removed.
  • The protocol file is the artifact of record. A page on this site renders it and does not replace it.
  • Each item set is used once. No later run on the same items is reported as held-out.
State of the WorkHome