THALis: trained-model artifact bundle

save the whole model bundle, not just the classifier!! CSP + normalizer + channel order need to match. future me is not going to remember which set of settings went with a random pkl

Saved bundle (c059325): data/models/tip-classifier-base/ contains csp_model.pkl, lda_classifier.pkl, normalizer.pkl and training_metadata.json. These components belong together: the classifier expects features produced and normalized in the same way used during training.

Related code: TrainingPipeline / TrainingConfig in src/control/lda_train.py; CSP extraction and the window buffer. The archive records serialization and cross-validation-reporting changes alongside the bundle.

CSP/LDA training console and model files

Review questions: which sessions and classes were used, how were folds split, and was every fitted transformation isolated from validation data? Check channel order, band definitions, feature dimension and normalization before loading the bundle for a later recording.

Evidence boundary: saving a trained model proves the pipeline produced an artifact; it does not by itself establish a particular held-out score, cross-session robustness or usable continuous control. Recover the actual metadata/metrics before reporting those outcomes.

Archived code references: c059325.