# Эксперимент: Residual-screened Koopman latent bottleneck (#540)

{
  "worked": true,
  "confidence": 9,
  "verdict": "Built a NumPy MVP for normalized Koopman modal residuals, quantile screening, and spectral forecasting. Clean eigenmodes had zero residual; observed residual energies closely matched analytic predictions across noise-variance and corruption-rate sweeps; screening removed the corrupted mode and reduced 50-step rollout MSE from 0.004739 to 0.0000290. This is a clear toy-setting win, but not yet evidence from a trained neural encoder-decoder.",
  "metrics": {
    "baseline": "Unscreened spectral rollout MSE: 0.0047392583; retained modes: 2.",
    "idea": "Residual-screened rollout MSE: 0.0000290028; retained modes: 1; validation residuals [0.0, 1.40006], threshold 0.70003. Noise-variance observed versus predicted residual-energy means were [0, 0.01653, 0.06366, 0.20961, 0.54901] versus [0, 0.01640, 0.06266, 0.21311, 0.54860]. Impulse-rate observed versus predicted values were [0, 0.01721, 0.04439, 0.09971, 0.17761, 0.30453] versus [0, 0.02042, 0.04960, 0.09473, 0.17399, 0.30027]."
  },
  "how_to_run": "/home/maxwelhelp/main/bin/python3 run_experiment.py",
  "files": [
    "residual_screened_koopman.py",
    "run_experiment.py",
    "results.json"
  ],
  "limitations": "Only a controlled two-mode synthetic linear latent system was tested. A neural encoder-decoder, RNN baseline, equal-parameter comparison, nonlinear Lorenz dynamics, confidence weighting, complex conjugate handling, and computational overhead were not tested."
}