Maya-Vaidya P6: From Synapse to Sentence — Modelling Formal Thought Disorder via NMDAR-Perturbed Antahkarana Ring-Attractor Spiking Neural Network

Version 2 (2026-08-04). Revised to the NLL Universal Paper Format v6. The original text is retained in full; nothing has been deleted. Corrections appear as marked blocks placed at the section that carries the claim, and each one states what the paper said, what the data show, the corrected claim, why it happened, and what still stands. This version withdraws a confirmed prediction and a strawman ablation.Temperature is a deterministic function of bump width and coherence is monotone in temperature, so a cliff in one must appear as a cliff in the other — the "matched cliff at g ≈ 0.77" is the chain rule, not a prediction; and EXP-D compares a nonlinear map against a comparator defined as linear, which any nonlinear map would win.The measured coherence decay (0.397 ± 0.010 → 0.166 ± 0.008, ρ = −0.664, p ≈ 5.8e-103) stands, as does the Path B structural negative on the rotationally symmetric ring. Everything below this line is the original description from version 1. It is retained unchanged for the record. Where it conflicts with the corrections above, the corrections stand. Two framings it repeats have since been withdrawn in full — the Bhaya Quiescence Law and the Buddhi S-Curve. Both are addressed in Maya-Meta P1, now superseded, and in the self-audit of Maya-Meta P2. Maya-Vaidya P6: From Synapse to Sentence builds the first direct computational bridge from a synaptic perturbation to degraded language-model output in a spiking neuromorphic framework, grounding Formal Thought Disorder (FTD) — the disorganised-speech symptom of schizophrenia — in a biologically motivated synaptic lever. The construct is Viveka (the discriminative faculty that holds the real apart from the unreal): a healthy network holds a sharp, localised internal representation; weaken the inhibitory governor and that representation spreads into noise. The substrate is a two-population excitatory/inhibitory spiking ring attractor (80 Bhaya excitatory, 20 Vairagya inhibitory; Murray-lineage bump), derived by mean-field analysis and validated before any language generation (pre-registered Failure Gate #1, 5/5 seeds monotonic). A graded NMDAR conductance g on the Bhaya→Vairagya drive is the sole lever; bump width (the circular standard deviation of the persistent-phase activity profile) is the readout; and a single fixed zero-free-parameter map, T = 0.7 × W(g)/W(1.0), routes that width to the sampling temperature of a stock GPT-2 124M, whose output coherence is measured as inter-sentence cosine similarity (all-MiniLM-L6-v2). No new free constant is introduced. Across four NMDAR levels (g = 1.0 → 0.5, n = 800), coherence falls monotonically from 0.397 to 0.166 (Spearman ρ = −0.664, p ≈ 5.8e-103); a fine nine-level dose-response (n = 1440) confirms the SNN width cliff and the GPT-2 coherence cliff coincide at g ≈ 0.77 — a pre-registered matched-location prediction, not a smooth gradient — and all 20/20 prompts show the effect (per-prompt ρ mean −0.674). An honest ablation shows the SNN is not redundant: at g = 0.9 it holds a non-linear "grace zone" (coherence 0.339 vs 0.264 for a Fradkin-style manual linear dial) that a hand-set dial cannot reproduce. A semantic-coupling route (Path B) is tested and reported as a structural negative — a rotationally-symmetric ring attractor is content-invariant by its geometry, precisely defining the feature-tuned, symmetry-broken substrate a future semantic-modulation paper (P7) would require. This is the sixth paper in the Maya-Vaidya Series and treats the Buddhi S-curve as a Formula Property (R² = 1.0 is a construction artifact; measured R² = 0.07–0.15 across substrates) and Bhaya Quiescence as a Reproducible Property within Antahkarana-governed architectures, per the series meta-protocol Maya-Meta P2; the driven bump-centre rate (~60.1%) is explicitly noted as NOT a resting Bhaya-Quiescence re-test. Honest caveats: the simulation is fully synthetic (no EEG, speech, or clinical data; researcher-generated prompts), width and temperature are collinear by the fixed map so mediation is not identifiable, and the coherence metric captures loose inter-sentence association only — not poverty of speech or tangentiality. Scope-lock per Maya-Meta P2: all findings are reproducible signatures within an Antahkarana architecture, NOT a clinical model of schizophrenia; this work is a research prototype only — it is not a licensed medical device and makes no diagnostic or treatment claim. Series: Part of the Maya-Vaidya Series — neuromorphic AI for clinical neuroscience. India-first. Research prototype only — NOT a licensed medical device. For medical emergencies in India: 112. Links: GitHub Repository (private — to request access: email research@nexuslearninglabs.in with subject Code Access Request — Maya-Vaidya-P6 and your research context) | Interactive Dashboard | FAQ | Full Series Index — venky2099.github.io Nexus Learning Labs, Bengaluru · UDYAM-KR-02-0122422 · BHASKAR IN-0526-9452JS ORCID: 0000-0002-3315-7907 · VAIRAGYA_DECAY_RATE = 0.002315 — an ORCID-derived provenance mark, not an experimental parameter. Each paper's disclosure block states whether it reached a result in that paper. Canary: MayaNexusVS2026NLL_Bengaluru