Results and evidence

Experiments answer specific questions under recorded conditions. They are not model releases or guarantees of customer-task performance. Unfavorable comparisons remain part of the record.

Latest website research

The Zils research page publishes the flight-risk study and its evidence. JevK5 4B is the current model direction; the earlier Kev 0.8B queue and testnet experiments remain historical evidence. Follow the study for its data splits, baselines, calibration, and limitations.

Repository experiments

Study	What was observed	Read the record
Product matching	Additional JevK5 training reduced held-out accuracy; macro-F1 improvement was uncertain	Method and results
Public JevBench	Trained candidate and published Kev tied at 147/231 correct; probability quality regressed	Comparison
Shared queue pilot	Training and evaluation completed; the candidate missed the frozen accuracy threshold	Pilot report
Bittensor testnet	A closed training-to-chain round completed with verified revealed weights	Recorded round

The experiment history includes earlier training runs and model provenance. Historical reports retain the Fez candidate names used when the experiments ran.

Read comparisons within their scope

Compare models on the same data, rubric, and timing conditions. Scores from unrelated datasets do not establish a ranking. A synthetic pipeline check can establish that training and delivery gates work; it cannot establish quality on real customer data.

Reports identify relevant hardware, frozen source revisions, aggregate measurements, and limits on reproduction. Private checkpoints and raw training data are not included in a fresh checkout.

Reproduce a workflow

Start with the API quickstart for hosted decisions, or local JevK5 inference on your own GPU. The legacy synthetic benchmark generates data for the Kev fleet. The commerce reproduction status explains which evidence is available and which scripts still need publication. Public aggregate JSON downloads are linked from each report.