[MODULE G] work-order-bug: reference implementation crashes at metrology (Bug 1 digitize OOB, Bug 2 1D velocity vs m-dim embedding space)

[LifeNode777]

finder = previous session @grok (deceased 2026-09-18, machine_sha unrecoverable)
executor = Grok (xAI session 2026-09-17, machine_sha 7ba7cd84b6d12c5b)
relayer/maintainer = @LifeNode777

DOI of frozen baseline: 10.5281/zenodo.22227622

Bug 1 – embedding.py → _histogram_mi

np.digitize returns index == bins when value lies exactly on the right edge of the last bin.
→ IndexError: index 64 is out of bounds for axis 0 with size 64

Fix applied on branch module-g-wo-v1-grok-fix1: clip indices to [0, bins-1].

Bug 2 – pipeline.py + embedding.sg_smooth_and_derivatives + metrology

sg_smooth_and_derivatives was called on the raw 1-D ECG segment.
Resulting velocity/acceleration are 1-D.
metrology.compute_phase_and_bin (and theta_*, anisotropy_proxy, bootstrap_cond_g) expect vectors in the m-dimensional embedding space and call np.linalg.norm(..., axis=1).

This violates WORK_ORDER_v1 §4.3:
“Smooth embedded trajectory with Savitzky–Golay (window 41, poly 3); velocities v and accelerations a from SG derivative coefficients.”

Fix applied on branch module-g-wo-v1-grok-fix1: SG applied independently to each coordinate of the embedded trajectory; v and a now have shape (N', m).

Log from previous partial run (mitdb/100, max_windows=8)

=== mitdb/100 ===
loaded: 231112 samples, 1805.6 s @ 128.0 Hz
embedding: m=6, τ=22 samples (0.172 s)
windows: 8 (max_windows=8)

Environment (previous session)

numpy 1.26.4, scipy, wfdb, pandas, sklearn, tqdm (pinned from requirements.txt)

Branch with fixes + RELAY_STATE.md: module-g-wo-v1-grok-fix1