{
 "status_at_freeze":"prepared-not-run",
 "run_id":"2026-10-28_paired-t4-route_v01",
 "host":"linuxpc",
 "source_commit":"91bdd1e7dcf193f3e7ca5a8933497fcef63b7960",
 "seeds":[20261121,20261122,20261123],
 "conditions":["quiet","A","B","A-route-cut","B-route-cut","A-route-restored","B-route-restored"],
 "groups":["T4a_left","T4b_left","T4a_right","T4b_right"],
 "pair_mapping":{"A":["T4a_left","T4b_right"],"B":["T4a_right","T4b_left"]},
 "duration_ms":1000,"dt_ms":0.1,"stimulus_start_ms":200,"stimulus_end_ms":700,"readout_start_ms":200,
 "aggregate_candidate_hz_per_template":50000,
 "candidate_schedule":"Two independent numpy SeedSequence(seed).spawn(2) streams, one per template. Each draws 5000 Binomial(global_min_group_size,50000*0.0001/global_min_group_size) counts, then5000×global_max_group_size uniform keys. Reuse both streams across A/B and all interventions. Within each0.1ms step, choose distinct native cells by shared-key rank restricted to the selected group. Template1 driver channels start at global_max_group_size. Exact aggregate event times/counts match between A/B; per-cell rates differ with group size.",
 "gain_basis":"Retain the prior single-group50000Hz fixed diagnostic budget without fitting to descending outcomes; two groups total100000Hz expected. Artificial model drive, not biologically calibrated rates.",
 "input_identity":"Unchanged2701eligible original-model IDs from t4-specificity-v01/results-v01/input-membership.csv, numeric-root sorted per group. Later subtype transfer constrained by original-index and unique-v630-side membership is an unvalidated biological identity assumption. All4potential native inputs rfc0ms in every condition; others retain original2.2ms; pulse w_syn*f_poi unchanged68.75mV.",
 "route_pairs":[["HSN_left","DNp15_left",16],["HSE_left","DNp15_left",26],["HSS_left","DNp15_left",2],["H2_right","DNp15_left",85],["HSN_right","DNp15_right",65],["HSE_right","DNp15_right",72],["HSS_right","DNp15_right",6],["H2_left","DNp15_right",74]],
 "route_intervention":"Exactly the8specified original chemical rows/346positive contacts. Assert row uniqueness and original contact-scaled weights. Cut only those weights. Restore by zeroing and reinstating their original values in a fresh full-network trial before running. Save row identities and original weights. All other weights must equal the original array exactly.",
 "primary_direction":"In intact A, DNp15_left-DNp15_right>=1spike; in intact B, DNp15_right-DNp15_left>=1spike, on every new seed. Count200–1000ms. No output sign flip, margin/gain/seed adjustment or alternate output selection.",
 "distinct_controls":"Quiet retains both candidate streams and A mapping with gate0; DNp15/DNa02 must be0. All6cut trials should have both DNp15 counts0 and a positive decrease from the corresponding intact driven-side count. All6freshrestores must reproduce the intact entire-network neuron/time event sequence, actualdriver events and fullweight hash exactly. Report each criterion separately even if one fails.",
 "secondary":"HS/H2 and DNa02 counts, allnativeinput population counts, quietwhole-network activity and input-noise variability are descriptive. No prespecified DNa02 direction-success criterion.",
 "scope":"Direct synthetic paired T4 input into the original full spiking circuit; no food signal or direct HS/H2/DN stimulation. No live image projection, retinotopy, physical-turn calibration, new learning, body, closed-loop world or driving. A pass supports anatomical paired-channel access and joint route necessity only under this frozen operating point. No necessity claim for individual edges or unique localization of the earlier visual-interface failure. Preserve all outcomes without adaptive retuning."
}