{
  "status_at_freeze":"prepared-not-run",
  "run_id":"2026-10-28_t4-channel-specificity_v01",
  "host":"linuxpc",
  "source_commit":"91bdd1e7dcf193f3e7ca5a8933497fcef63b7960",
  "visual_commit":"0a69fa2fc8e1bfe1d837d722c252ab5f5d6fcbfa",
  "seeds":[20261111,20261112,20261113],
  "conditions":["quiet","T4a_left","T4b_left","T4a_right","T4b_right"],
  "duration_ms":1000,"dt_ms":0.1,"stimulus_start_ms":200,"stimulus_end_ms":700,"readout_start_ms":200,
  "aggregate_candidate_hz":50000,
  "gain_basis":"Retain the previous visual-interface per-eye ceiling as a single fixed diagnostic budget, not a gain fit or a biological rate calibration. Each active condition uses exactly the same event times and total number for a seed.",
  "input_identity":"All 2701 eligible original T4a/b IDs from the previous frozen bridge membership, sorted by numeric root ID per group. Later subtype annotation transfer remains unvalidated biological identity. Original graph and weight rows unchanged.",
  "candidate_schedule":"For each 0.1ms step in 200-700ms, draw Binomial(min_group_size, aggregate_hz*dt_ms/1000/min_group_size) counts. Reuse these counts across all five conditions. Shared per-step uniform random keys rank each group's cells; take count distinct cells per step. This prevents duplicate pulses to a cell within a time step and equalizes aggregate event times, not per-cell rates. Quiet uses T4a_left mapping and gate0.",
  "native_inputs":"All four potential T4 input groups have refractory0ms in every condition, following original author stimulated-input convention. Other neurons retain original refractory2.2ms. Pulse amplitude w_syn*f_poi unchanged. No direct HS/H2 or descending stimulation.",
  "primary":"Within each anatomical side and seed, baseline-corrected ipsilateral HS total (HSN+HSE+HSS) under T4a exceeds T4b by at least one spike AND preferred HS response exceeds quiet by at least one. Baseline-corrected ipsilateral H2 under T4b exceeds T4a by at least one spike AND preferred H2 response exceeds quiet by at least one. All12 side/seed/channel comparisons must pass. Do not compare raw three-HS totals against one H2 cell.",
  "secondary":"Report contralateral HS/H2, both anatomical DNp15/DNa02 counts, native T4 spike counts and all failures. Descending outputs are not optimization targets or a steering success criterion.",
  "source_transform_audit":"Read saved visual arrays only. Compare mean(ReLU(activity)) with ReLU(mean(activity)) and the tested positive change from pre-motion mean. These are population representation summaries before synaptic weights, not the complete source weighted-current computation. Assert actual recovered source activation type and learned parameter hash before any interpretive claim.",
  "scope":"A channel-specificity diagnostic and saved-data representation audit, not a validated new visual interface, causal localization, learning, physical turn or driving test. No retinotopic mapping, body or external policy. Do not tune signs, identities, gain, margins or seeds after reading outcomes. Preserve any failure and do not overwrite attempts."
}