PhaseProbe evidence report · schema 2.0

Finding

{}

Search and refinement history

Configuration

{
  "adapter": {
    "factory": "predator_prey_adapter",
    "kind": "python",
    "module": "phaseprobe.examples.scipy_models",
    "options": {
      "atol": [
        1e-12,
        1e-12
      ],
      "dense_output": false,
      "identity": "phaseprobe-predator-prey-scipy-v1",
      "initial_state": [
        10.0,
        5.0
      ],
      "max_step": 0.05,
      "method": "DOP853",
      "rtol": 1e-10,
      "state_names": [
        "prey",
        "predator"
      ],
      "t_eval": {
        "kind": "linspace",
        "points": 1201
      },
      "t_span": [
        0.0,
        30.0
      ],
      "vectorized": false
    }
  },
  "check": {
    "analysis": "invariants"
  },
  "classification_rule": "Require both populations to remain positive over the declared retained grid.",
  "invalid_state_policy": "Abort on solver failure, invalid shape, NaN, infinity, or non-finite observables.",
  "model": "predator-prey-scipy",
  "parameters": {
    "alpha": 1.1,
    "beta": 0.4,
    "delta": 0.1,
    "gamma": 0.4
  },
  "policy": {
    "forbid_findings": true,
    "require_finding": false,
    "require_invariants": true
  },
  "refinement_rule": "DOP853 with tight tolerances and a bounded maximum step; compare with the committed coarse negative control.",
  "replay": {
    "endpoint_time_atol": 1e-10,
    "event_time_atol": 1e-08,
    "expected_solver_success": true,
    "invariant_measure_atol": 1e-08,
    "max_unmatched_points": 0,
    "mode": "tolerance",
    "observable_atol": {
      "first_integral": 1e-08,
      "predator": 1e-07,
      "prey": 1e-07
    },
    "require_classifier": true,
    "require_invariants": true,
    "retained_grid_time_atol": 1e-12,
    "state_atol": 1e-07,
    "state_rtol": 1e-07
  },
  "schema_version": "2.0",
  "seed": 31,
  "simulation": {
    "hard_state_limit": 1000.0,
    "trace_cap": 1201
  },
  "tolerances": {
    "invariant_drift": 1e-08
  }
}

Replay and generated test

The run directory contains a versioned replay.json fixture with declared tolerance comparison. Artifact SHA-256 integrity is retained in both modes. The generate-test command validates and copies that fixture into a fixed pytest template.

Scientific limitations

This report contains bounded numerical evidence. It does not prove global minimality, an exact bifurcation point, chaos, a formal Lyapunov exponent, or scientific validity outside the declared model, integration settings, search space, and tolerances.