Experiment YYYY-MM-DD — toy next-step prediction

Goal

Sanity-check HelicalCell v1 on a periodic signal.

Config

• dim=32, epochs=3, Adam lr=3e-3

Results

• Loss decreased across epochs on periodic synthetic data.

Observations

• Training runs without errors; helical rotation + triangle-fusion is numerically stable at this scale.

Next

• Add copy/reverse/parity benchmarks.
• Try learned-phase Δφ and compare vs fixed.