Эксперимент: Consensus-Corrected Topology-Invariant GNN (#1054)

{ "worked": false, "confidence": 9, "verdict": "Built and numerically verified the consensus correction and the separate Laplacian diffusion stability condition. The mechanism checks passed: mean-consensus disagreement matched (1-alpha)^2, Fiedler-mode decay matched (1-alpha*lambda2)^2, and the Laplacian instability boundary matched 2/lambda_max. However, the learned consensus model reduced representation disagreement while worsening prediction MSE at every tested outage level, and the stated Laplacian bound does not directly apply to the proposed global-mean correction.", "metrics": { "baseline": "MSE at edge retention [0.75, 0.55, 0.35, 0.20]: [0.000133, 0.000270, 0.000883, 0.003274]; representation disagreement: [0.2093, 0.2183, 0.2386, 0.2780]", "idea": "MSE at edge retention [0.75, 0.55, 0.35, 0.20]: [0.001707, 0.002694, 0.006369, 0.015859]; representation disagreement: [0.1411, 0.1431, 0.1470, 0.1572]" }, "how_to_run": "/home/maxwelhelp/main/bin/python3 experiment.py", "files": [ "experiment.py", "results.json", "REPORT.md" ], "limitations": "Only a tiny synthetic node-regression task was tested, not the proposed 14-bus or 118-bus power-flow benchmark. No multi-seed statistics, learned-alpha sweep, FLOP/runtime comparison, or larger-graph evaluation was performed." }