# Эксперимент: Normal-Space Quotient Encoder (#1409)

{
  "worked": true,
  "confidence": 8,
  "verdict": "The quotient-aware orbit regularizer was implemented and numerically verified. Projector idempotence and orbit-preservation errors were about 1e-10, the singular-origin projector was zero, and three seeds showed substantially lower representation drift with no clean-accuracy loss. This is a promising toy-level result, not yet a broad benchmark win.",
  "metrics": {
    "baseline": "Mean clean accuracy 0.9818, rotated accuracy 0.9816, representation drift 0.589.",
    "idea": "Mean clean accuracy 0.9853, rotated accuracy 0.9860, representation drift 0.107; drift reduced about 5.5x.",
    "math": "Projector idempotence error 5.2e-11; orbit projection error 7.2e-11; singular-origin projector norm 0.",
    "how_to_run": "/home/maxwelhelp/main/bin/python3 experiment.py && /home/maxwelhelp/main/bin/python3 multiseed.py"
  },
  "files": [
    "experiment.py",
    "multiseed.py",
    "results.json",
    "multiseed_results.json"
  ],
  "limitations": "Only a small synthetic 2D radial classification task with known SO(2) actions was tested. Rotated-MNIST/CIFAR, standard augmentation, learned transformations, full FLOP matching, and broad hyperparameter sweeps were not tested."
}