Эксперимент: Active Resolvent Regularization (#564)

{ "worked": false, "confidence": 9, "verdict": "Built a reproducible Wishart-plus-asymmetric-perturbation MVP in active_resolvent.py, measuring the MP gap, resolvent susceptibility, eigenvalue stability, transient gain, and a small recurrent training comparison. The MP gap prediction was confirmed: predicted 0.2500 versus measured 0.24767, but the predicted crossover εc≈0.25 was not confirmed: the first 25% susceptibility change occurred at ε≈1.30, about 5.2× later. The regularized training run also did not provide a win, with higher loss and dramatically larger transient gain than baseline, so the proposed mechanism is not validated in this toy setting.", "metrics": { "baseline": "Training loss 7.29e-05; max real eigenvalue 0.190; sampled transient gain 10.96; susceptibility 129.13.", "idea": "Training loss 1.35e-04; max real eigenvalue 0.887; sampled transient gain 2725.58; susceptibility 2.01.", "mechanism_predictions": "MP gap: predicted 0.2500, observed 0.24767. Activity crossover: predicted εc≈0.25, observed susceptibility 25% change at ε≈1.30. Stability: no instability observed through ε=1.50, while transient gain remained near 1 until the largest tested activity." }, "how_to_run": "/home/maxwelhelp/main/bin/python3 active_resolvent.py", "files": [ "active_resolvent.py", "results.json" ], "limitations": "Only one Wishart/asymmetric-matrix seed and a small epsilon grid were tested; no GRU, copy-memory, or permuted-MNIST experiment was run. The training demo is a tiny recurrent regression proxy, and the regularizer was not tuned across targets or coefficients." }