{
  "status": "pre-execution audit plan; no model run performed",
  "date_local": "2026-10-08",
  "software_regression_target": {
    "model": "LSV1M_tiny_stepcurrentmodule",
    "framework_commit": "f5c09a0eccc5786daf78881d883bbce586a00e6d",
    "test": "tests/full_model/test_models_stepcurrentmodule.py::TestLSV1MTinyStepCurrentModule",
    "proposed_command_from_framework_root": "pytest tests/full_model/test_models_stepcurrentmodule.py -v",
    "upstream_acceptance": "numpy.testing.assert_equal on saved spike-time vectors for six populations and membrane-potential vectors for four cortical populations, with up to 25 stored voltage neurons per population",
    "additional_reporting": ["model, framework and dependency commits", "simulator, Python and compiler versions", "seeds, thread and MPI settings", "individual neuron and segment identity, units and time bounds", "spike count, first divergent sample and maximum voltage deviation on mismatch", "wall time and peak resident memory"],
    "failure_policy": "Keep upstream exact-equality criterion. If a different environment changes floating-point output, report a failed exact match before proposing a separately versioned tolerance. Do not silently loosen the test.",
    "claim_boundary": "Passing establishes a software regression match for the small test configuration; it does not reproduce the full 2024 paper or independently validate cat biology."
  },
  "biological_benchmark_candidates": [
    {
      "source": "10.7554/eLife.41925",
      "target": "Figure 7E, especially reliable versus unreliable firing rate at 50 recorded LGN inputs",
      "published_mean_rates_hz": {"reliable": 1.8, "unreliable": 7.8},
      "published_trials": 10,
      "duration_seconds_methods": 89,
      "time_step_ms": 0.1,
      "required_before_execution": ["identify original simulation entry point or explicitly choose independent reimplementation", "recover the exact selected 50 LGN spike trains and stimulus alignment", "recover joint synaptic amplitude, variability and failure-rate samples", "resolve methods reset -60 mV versus appendix reset -65 mV", "resolve original randomization and reference uncertainty"],
      "tolerance_status": "not frozen; reference outputs and ambiguity resolution required before a quantitative pass/fail tolerance is justified"
    },
    {
      "source": "10.1371/journal.pcbi.1012342",
      "target": "Figure 4 spontaneous-activity population statistics",
      "model_release_commit": "04bf11b72ff5f78e66f28e037016e8e33b09c12e",
      "entry_point": "LSV1M/run_spont.py with param_spont/defaults",
      "protocol_duration_ms_from_source": 40320,
      "metrics": ["mean population firing rates", "inter-spike-interval coefficient of variation", "pairwise binned spike-count correlation", "membrane-potential and excitatory/inhibitory conductance summaries"],
      "tolerance_status": "not frozen; recover full reference exports or record a digitization method and uncertainty before running the scientific comparison",
      "claim_boundary": "A smaller or downsampled configuration cannot be labelled reproduction of the full reference model without its own comparison."
    }
  ],
  "future_controller_evidence": [
    "explicit, versioned input encoder and output decoder",
    "use of replayable synthetic inputs before external events",
    "intact versus frozen/disconnected model comparisons and simpler matched baselines",
    "separate simulation output, decoded request and executor decision",
    "bounded sandbox actions, recovery and duplicate-event tests before any live token operation"
  ]
}