Stable evaluation
Validate wrapped predictions, pin the protocol, and produce a comparable report.
Evaluate existing predictions
Start with one wrapped predictions artifact, validate the predictions interface contract, and produce a comparable report. Search the generated command and schema references without navigating the full repository or PDF manual.
Choose the lane after the stable artifact works
The four lanes are intentionally separate. Bridge, Benchmark, and Research work extend or qualify the Stable evaluation boundary; they do not silently promote themselves to Stable support.
Bridge
Export from an external framework into the same predictions interface contract.
Benchmark
Qualify runtime parity only after the stable evaluation artifact is working.
Research
Run opt-in experiments over an already evaluated stable artifact and keep the maturity boundary visible.
Short paths, repository-backed details
30-minute tutorial
Install, produce CPU-only proof artifacts, validate both sides, evaluate, and read the report.
Follow the tutorialGenerated command reference
Every entry comes from tools/manifest.json, including
maturity, inputs, examples, and source links.
Schema browser
Inspect the checked-in prediction, evaluation, training, and Research JSON Schema surfaces.
Search schemasExamples and report reading
See report and overlay examples, expected companion files, and the checklist for fair comparison.
Open examplesGlossary
Use the same terms for artifacts, protocol identity, maturity, and skipped lanes.
Open glossaryWhat can go wrong
Start from observed symptoms, verify the report, and avoid attributing drift before inputs and protocol align.
Open the failure guide