Gap 3 — the axonal connectome and the O×W double dissociation

Gap 3 adds the second physical coupling channel — the axonal connectome — that Sim 1 omitted. On the surviving Gap-2 micro-model, routing (W) is separable from cell-count capacity (O): a W-fault gives a sociality-selective deficit, an O-fault a capacity-selective one. ST-3 confirms the crossover is sign-stable across nine cells — two separable axes.

Sim 1 is ephaptic/proximity-only, so it omitted the long-axon connectome. Gap 3 adds it on the surviving Gap-2 micro-model: a grounded long-range connectome of ten cited cortical edges, plus an OWTCortex that frees cell count N while holding K = 7 hubs fixed. Two orthogonal axes form a relative-selectivity crossover — W-axis (routing, the hub long-range edges) and O-axis (capacity, cell count). A W-fault is sociality-selective (W_sel = +0.128), an O-fault capacity-selective (O_sel = +0.032). ST-3 holds the crossover sign-stable across 9 cells (W_sel +0.112→+0.128, O_sel +0.032→+0.094), so autism reads as a W-fault and intellectual disability as an O-fault — held at arm's length from clinical use.

The gap: a missing coupling channel

Sim 1 couples organs by the measured ephaptic near-field and by proximity only. Two organs wired by long axons but not spatial neighbours — cortex and thalamus, cortex and hypothalamus — therefore had negligible coupling in the kernel (W0[cortex,thalamus] = 4e-4, W0[cortex,hypothalamus] = 2e-5). Gap 3 adds the axonal connectome as the second physical coupling channel, on the surviving Gap-2 micro-model.

The fill, in two parts

Part A — a grounded long-range connectome. Ten cited cortical edges, each carrying an explicit grade. cortex↔thalamus weight 1.00 [L cited] (Klein 2010 NeuroImage 51:555; J Neurosci 43:7780, 2023); cortex↔hypothalamus light [L cited] (Öngür & Price 1998 JCN 401:480; Radley 2006 J Neurosci 26:12967); the rest carry [L ordering; O magnitude]. An organ-scale integration check is recorded as an honest negative (a single-tract organ-scale lesion is not sign-stable).

Part B — the O×W sweep. An OWTCortex parameterizes the Gap-2 model to free cell count N while holding K = 7 hubs fixed, and runs the orthogonality sweep. The harness is non-circular at the code level: at N = 120 it reproduces the Gap-2 far_binding value exactly (< 1e-12) — the same surviving model, re-parameterized. new_tuned_constants = 0; the engine is read-only.

The two orthogonal axes

The honest form is a relative-selectivity crossover. The W-axis (routing) is the hub-borne long-range edges; the W-fault cuts them (f_W up to 0.9 at N = 120), and the read-out sociality is far_binding (cue a near block, recover a distant block — never the cut edge). The O-axis (capacity) is cell count N (Hopfield capacity ≈ 0.14 N), with the wiring backbone intact; the O-fault reduces N (120 → 44), and the read-out capacity is scattered partial-cue completion. The sweep grids (not tuned) are f_W_grid = (0, 0.3, 0.6, 0.9), N_grid = (120, 90, 64, 44), margin 0.02.

What the build found

The W-fault (f_W = 0.9) gives ΔS = −0.196, ΔC = −0.067, so W_sel = ΔC − ΔS = +0.128 (sociality-selective). The O-fault (N = 44) gives ΔS = −0.028, ΔC = −0.060, so O_sel = ΔS − ΔC = +0.032 (capacity-selective; with the tracts intact, reducing cells sometimes even helps sociality). The non-circular control holds: hub-cut sociality 0.665 vs mass-matched random-local-cut 0.872 — long-range-specific, not generic mass.

ST-3: are the axes really separable?

The break ST-3 was designed to cause is that the axes are not separable — one factor, not two. Had it broken, the autism/ID double dissociation would collapse to a single factor and the clinical mapping would be rewritten. The test imports the exact build model and tests three signed features across 9 cells (3 seed cohorts × settle depths 20/40/60): W targets sociality (W_sel > margin), O targets capacity (O_sel > margin), and the double dissociation (both positive ⇒ crossover; −1 if either axis crosses the wrong way ⇒ single factor).

What survived, and what stays open

The dissociation survived: W_sel ranged +0.112 → +0.128 and O_sel +0.032 → +0.094, both strictly positive on every cell, so the crossover never flips (dissociation_sign_stable = True, broke_on = []). Gap 3 is graded [L] in-silico: two separable axes — autism as a W-fault (routing), intellectual disability as an O-fault (capacity). This is held at arm's length as an in-silico phase / autoassociator result, not a clinical measure. The dissociation is relative (each fault has a small off-target effect), gene-blind, and a micro-model property (the organ-scale single-tract lesion is not sign-stable); magnitudes and gene-specificity stay [O].

Firewall (Axis-A). Every quantity on this page is a structural quantity of an in-silico model — a sign, a margin, an ordering — and is not a felt state, an experienced quality, or a level of consciousness (consciousness_claim = 0; the hard problem stays open, hard_problem_open = 1). These are model results, not validated neuroscience and not clinical guidance. efficacy = 0; not medical advice.