Kaun_boardSourceTraining run logging and monitoring.
Kaun_board records training metrics to disk and reads them back for dashboards and analysis.
Log is the writer: it creates a run directory and appends scalar events as JSONL. Run and Store are readers: they load run directories and aggregate metrics incrementally.
A run directory contains a run.json manifest and an append-only events.jsonl log. Run IDs follow the format YYYY-MM-DD_HH-MM-SS_XXXX with an optional experiment suffix.