{
  "run_id":"2026-10-28_visual-neural-bridge_v01",
  "status_at_freeze":"prepared-not-run",
  "source_commit":"91bdd1e7dcf193f3e7ca5a8933497fcef63b7960",
  "flyvis_commit":"0a69fa2fc8e1bfe1d837d722c252ab5f5d6fcbfa",
  "annotation_later_commit":"ebd66db2596fcc39c6950fb54ea3efa00f7fe8a0",
  "host":"linuxpc",
  "authorization":"User: 이어서 쭉죽, following the completed Flyvis visual baseline and proposed source-backed interface feasibility test. Existing brain-only behavior contract remains open.",
  "hypothesis":"A prespecified population-mean T4a/b interface can propagate opposite binocular motion patterns through the unchanged full spiking circuit and produce opposite side preferences in DNp15. This is an engineered hybrid-neural interface test, not a biological or retinotopic reconstruction.",
  "seeds":[20261091,20261092,20261093],
  "duration_ms":1000,
  "dt_ms":0.1,
  "input_motion_onset_ms":200,
  "readout_start_ms":200,
  "flyvis_stimulus":{"dataset":"MovingEdge","offsets":[-10,11],"intensities":[1],"speeds":[19],"height":80,"post_pad_mode":"continue","t_pre":1.0,"t_post":1.0,"dt":0.005,"angles":[0,180]},
  "upstream":"The same author-specified pretrained flow/0000/000 visual model; exact source/checkpoint reused without training. Save every T4a/b model-column trace and input movie for 0 degrees,180 degrees,and uniform grey. Compute positive baseline-subtracted population means; no motor data used to choose or normalize them.",
  "encoding":"For each eye, each T4a/b group receives rate=(50000 / number of eligible original neurons in that group)*positive_population_mean/denominator Hz, with denominator the maximum over time of the sum of T4a/b positive means across both moving movies. Max total injected candidate rate is 50000 Hz/eye. Zero values below 1e-6 model units. Piecewise-constant 5ms visual samples, source time=simulation time-0.2s. This conversion is an uncalibrated engineering assumption, not measured firing rates. No body heading, target, route or desired motor output enters the encoding.",
  "membership":"Use later annotated T4a/b root IDs only when exactly present in the original 127400-neuron index and uniquely present in the pinned v630 annotation with matching left/right side. No root lifting, replaced IDs or graph modifications. Later labels are an explicit annotation-transfer assumption; root membership alone does not independently validate biological identity or spatial mapping. Homogeneous population broadcast discards individual retinotopy.",
  "input_convention":"Official source input pulse weight w_syn*f_poi, stimulated native T4a/b refractory0ms per authors' sensory stimulation helper; all other model parameters unchanged. These biological graded cells are represented as spiking units by this model. Common per-cell Bernoulli/Poisson-discretized uniforms per seed across inputs, candidate schedules fixed across interventions.",
  "patterns":{"A":"left eye:180deg/front-to-back-favored; right eye:0deg/back-to-front-favored","B":"left eye:0deg; right eye:180deg","translation":"both eyes:180deg","grey":"both eyes:uniform0.5 grey"},
  "conditions":["grey","A","B","translation","A-input-block","B-input-block","A-route-cut","A-route-restored","B-route-cut","B-route-restored"],
  "route_cut":"Zero only original chemical rows from all 6 HSN/HSE/HSS cells and both H2 cells to the two DNp15 outputs. Preserve all other weights. Restore original weights before a fresh paired simulation; no in-run healing.",
  "outputs":{"DNp15":{"left":"720575940610301413","right":"720575940623431351"},"DNa02":{"left":"720575940629327659","right":"720575940604737708"}},
  "criteria":{"primary":"DNp15 side selectivity: countLeft-countRight >0 for A and <0 for B on every prespecified seed. Anatomical channel criterion, not an animal turn-angle calibration.","secondary":"Report DNa02 left-minus-right counts under the same A/B criteria without selecting or flipping output signs.","input_block":"Both DNp15 and DNa02 counts must be0 under blocked external drive; actual baseline activity is reported.","route_effect":"Report driven-side DNp15 paired intact-minus-cut counts. Only if positive in all6 input/seed pairs call the route effect consistently positive in this setup; no guaranteed silencing assumption.","restore":"All6 fresh restored runs must exactly match whole-network spike-event sequences, native weight hashes and fixed driver events of their intact pairs."},
  "limits":"Only two source visual directions, one speed, three simulated input-noise seeds, one fixed upstream fitted model, one prespecified gain. No spatially resolved native projection, physical firing-rate calibration, body, closed-loop environment, navigation, learning, or driving. No proof of superiority or preservation of an imaged animal's behavior. No fallback controller or permanent worker. Preserve failures and stop instead of tuning gain, selecting a new output, flipping signs or changing criteria after results."
}