Reference

These pages state the exact boundary contracts needed when implementing a training loop or extension. Begin with getting started, then read the inference contract before implementing a real callback; the remaining pages are references for exact shapes, fields, and boundaries.

• Inference contract: callback inputs, outputs, routing, and legality.
• Batch formats: named fields for each learner family.
• Glossary: collection, game-state, network, lifecycle, and search terms.
• Telemetry fields: collection metrics and episode records.
• Python API: module map and source of signature truth.
• Troubleshooting: common callback, stream, build, and learning failures.
• Current boundaries: deliberately unsupported or bounded semantics.
• References: sources for the algorithms, encodings, and validation targets.