v0.2.0

Model-quality release: better generalization and weak classes, calibrated confidence, and a retrain on all 15 subjects. No API contract change (512-sample window, 8 classes).

Added
Train-only data augmentation (jitter / scaling / time-shift / magnitude-warp), pure numpy in preprocessing.py, applied to the train split only (val/test untouched).
Config-driven regularization: encoder Dropout1d + head dropout via ModelConfig.
Cosine LR schedule + early-stopping (patience) in training.
Temperature-scaling confidence calibration: fit on validation (fit_temperature, torch-free), stored in the checkpoint, applied at inference in predict.py (softmax(logits / T); backward-compatible, defaults to T=1.0).

Changed
Retrained on all 15 subjects (S6 included); regenerated model/metrics/*.json.
Multimodal test accuracy 0.650 → 0.776; walking F1 0.36 → 0.81. The old overfitting gap closed — best validation now lands at epoch ~10 and test ≈ validation.
ECG-only baseline test 0.371 → 0.609 on the same pipeline.

Honest note
working is now the weakest class (test F1 0.47, confused with lunch_break); with only 2 validation + 2 test subjects the metrics are noisy (val/test ordering can flip).

v0.1.0

First tagged release: an end-to-end, multimodal biosignal activity classifier delivered as a product slice (model → API → LLM report → dashboard), with honest, subject-wise metrics and the educational disclaimer travelling with every output.

Added

Phase 0 — Scaffolding. Repo structure, disclaimer, dataset selection (PPG-DaLiA, CC BY 4.0), design docs, and the FastAPI skeleton (GET /health, GET /).

Phase 1 — ECG-only classifier. PPG-DaLiA loader, a 1-D CNN, a subject-wise training CLI with honest metrics, and POST /predict.

Phase 2 — Multimodal fusion. ECG + PPG + 3-axis accelerometer: cross-device resampling to a shared 64 Hz base, per-modality normalization, one encoder per modality with late fusion. Test accuracy 0.371 (ECG-only) → 0.650 (multimodal) on the same held-out subjects.

Phase 3 — Claude report layer. POST /report turns the numeric prediction into a plain-language, disclaimer-prefixed report via the Claude API.
Phase 4 — Next.js dashboard. Signal plots (ECG/PPG/ACC) with the relevant segment shaded, prediction + report panels, an always-on disclaimer, and a backend-less demo mode (badged synthetic/canned data). Talks to the API through a server-side proxy (no CORS).

Phase 5 — Release. GitHub Actions CI (Python light/full tiers + dashboard), Docker Compose deployment (api + dashboard), a MODEL_CARD.md, and an in-app "About this model" panel surfacing the honest metrics.

Fixed
Corrected the Phase 2 validation Weighted-F1 in the README (0.744 → 0.705) to match the canonical model/metrics/phase2_multimodal.json.