Эксперимент: Icosahedral Congruence-Robust Strain Sensor (#574)

{ "worked": true, "confidence": 9, "verdict": "Built the differentiable icosahedral congruence sensor with symmetric Frobenius-preserving vectorization, transformed projectors, ridge least-squares inversion, and trace-free projection. The mechanism passed all three checks: identity rank 6, rank 6 under every tested SL(3) deformation, and increasing reconstruction error as conditioning degraded. The toy comparison favored the icosahedral frame over the fixed random six-direction baseline, but the end-to-end graph-fluid claim was not tested.", "metrics": { "baseline": "Random six-direction frame: relative reconstruction error 0.01004 at anisotropy 0 and 0.16886 at anisotropy 3; coordinate frame rank 3.", "idea": "Icosahedral frame: rank 6 across all tested deformations, noiseless maximum relative error 1.78e-9, and noisy error 0.00127 at anisotropy 0 versus 0.01612 at anisotropy 3." }, "how_to_run": "/home/maxwelhelp/main/bin/python3 experiment.py", "files": [ "experiment.py", "results.json", "REPORT.md" ], "limitations": "No graph neural fluid predictor, learned-direction training, equal-parameter rollout, or rollout-instability experiment was performed. The deformation sweep used symmetric trace-free exponential deformations rather than arbitrary nonsymmetric SL(3) matrices." }