Right now, running calibrate.py pulls in every single correction ever recorded in the local database. You often want to tune weights using only recent matches, a specific tournament, or ignore early games from when you were still getting used to the app. The database already stores match names and timestamps, so we just need to hook up filter options.
Suggested fix:
Add filter arguments (e.g., --since, --match, --competition) to feedback.training_pairs() and expose them as command-line flags on calibrate.py. Definition of Done:
Running python -m tracer.calibrate --since 2026-08-01 only runs calibration on corrections logged on or after that date.
Right now, running calibrate.py pulls in every single correction ever recorded in the local database. You often want to tune weights using only recent matches, a specific tournament, or ignore early games from when you were still getting used to the app. The database already stores match names and timestamps, so we just need to hook up filter options.
Suggested fix:
Add filter arguments (e.g., --since, --match, --competition) to feedback.training_pairs() and expose them as command-line flags on calibrate.py. Definition of Done:
Running python -m tracer.calibrate --since 2026-08-01 only runs calibration on corrections logged on or after that date.