Pokémon TCG — Experiments and Anamnese

An experiment is a generic research container and does not imply retraining. It may represent a training run, an existing model tested with a new deck, a matchup, an arena evaluation or replay analysis.

The base experiment has stable identity and lifecycle facts. Type-specific tables store training runs, deck tests, matchups and replay analyses. Training configuration is normalized and immutable rather than stored as generic JSON text.

This remains planned scope. The current rl/results_db.py schema does not yet include generic experiments, experiment observations, normalized training configs, training runs, deck tests, matchup experiments, or replay analyses.

Anamnese is an append-only temporal observation stream for models and experiments. It records evolving hypotheses, results, regressions and context without replacing history with a mutable note field.

Related

pokemon_tcg_local_overhaul · pokemon_tcg_local_platform_status · pokemon_tcg_implementation_specification · pokemon_tcg_schema_evolution · pokemon_tcg_physical_schema · pokemon_tcg_submissions_and_elo