Prior-guided TCN fusion — dataset (5 sessions, 1000 Hz) & minimal checkpoints (review v0.1)

Curated dataset (5 sessions, 1000 Hz) and minimal checkpoints to reproduce Tables/Figures in our submission.

Contents:
• data/11..15/aligned_full_data_linear_XX.csv (synchronized time series at 1000 Hz)
• checkpoints/std_heads_run_20250921_1919_best.pt (minimal inference checkpoint)
• DATA_DICT.md (column dictionary, units, preprocessing & evaluation notes)
• CHECKSUMS.txt (SHA-256 for integrity)

Columns (see DATA_DICT.md for details):
ANG (deg, GT from high-speed imaging + markers), Flex_Angle (deg), IMU_Angle (deg),
Proximal_{Acc, Gyro, Ang}{X,Y,Z}, Middle_{Acc, Gyro, Ang}{X,Y,Z}.
Sampling rate: 1000 Hz. No NaNs.

How to use:
1) Download this Dataset item and place under project root:
data/11..15/*.csv
checkpoints/std_heads_run_20250921_1919_best.pt
2) Verify integrity:
macOS/Linux: shasum -a 256 -c CHECKSUMS.txt
Windows (PS): see command in DATA_DICT.md
3) Run the software item:
python metrics.py --config config.yaml \
--ckpt_ours checkpoints/std_heads_run_20250921_1919_best.pt \
--gate --trim-head-ms 1000 --trim-tail-ms 1000

License:
• Data & checkpoints: CC BY-NC 4.0 (non-commercial, with attribution)
• Code is released separately under Apache-2.0 in the Software item