Maya-Sutra P4: Listening on Silicon — Reliable Rejection and Unreliable Report: An Analog Neuron as a Sound-Event Detector Across Three Nightly Calibrations of BrainScaleS-2

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 adds one scoping disclosure; no claim is retracted.The paper's most valuable output is a self-correction produced by a pre-registered prediction that failed: neurons called "always-on" in P02/P03 read exactly 0.00 under a true silence null while carrying the highest gain on the chip, so they are hypersensitive rather than spontaneously active.The service finding stands — across three nightly calibrations minimum instrument gain fell 1.00 → 0.85 → 0.60 while within-session disagreement rose 0/40 → 17/40 → 32/40. One caveat is extended: silence encodes to exactly zero input spikes, so zero false fires across 219 pooled readings is close to the trivial expectation; injecting a noise floor would convert it into a measurement. 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 Two framings it repeats have since been withdrawn in full — the Bhaya Quiescence Law and the Buddhi S-Curve. Both are addressed in Maya-Meta P1, now superseded, and in the self-audit of Maya-Meta P2.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-Sutra P4: Listening on Silicon turns the chip outward. Papers 1 to 3 characterized a BrainScaleS-2 neuron acting on its own synaptic weights; here, forty clips recorded in the author’s own voice — a loud and a soft sustained vowel, five spoken phrases, and silence — are reduced to an RMS loudness envelope and replayed to one calibrated analog neuron as a rate-coded spike train. The construct is Śravaṇa (श्रवण; hearing, the first response to the world). No novelty is claimed for audio on BrainScaleS-2: microphones have been wired directly into the chip’s analog inputs for real-time sound localization (arXiv:2602.04582), speech benchmarks have been trained on it (10.1073/pnas.2109194119), and the SHD dataset with its Lauscher cochlea model originates from the same laboratory (10.1109/TNNLS.2020.3044364). The contribution is what three independent nightly calibrations reveal about reliability, and the answer is an asymmetry. Transmission degrades: the minimum instrument gain over known spike counts falls 1.00 → 0.85 → 0.60 across the three calibrations, and within-session reproducibility collapses from 0 of 40 clips varying across repeats to 32 of 40, with a maximum spread of 32 spikes on a single clip. Rejection does not: across 219 silence readings spanning all three calibrations the maximum response was 0.0 — not one false fire. A selectivity window in which only the loud event passes exists on every calibration, with a robust core at synaptic weights 44 and 48. A falsification test guards the core claim: replaying an identical 146-spike clip while varying only the synaptic weight moves the response from 0 to 146 spikes, so the chip is deciding rather than echoing. Per-neuron identity persists (pairwise gain correlations r = 0.961, 0.962, 0.994), the same cell least sensitive and the same cell most sensitive on all three days. This corrects our own prior work: the neurons reported as “always-on” in Papers 2 and 3 read exactly 0.00 under silence on every calibration while carrying the highest gain on the chip — they are hypersensitive, not spontaneously active. Papers 2 and 3 could not have discovered this, because their protocol always delivered a burst and a true null stimulus was impossible; P4’s silence clips carry exactly zero input spikes by construction. The honest negative was predicted in advance and confirmed: “hi” and “did you have breakfast?” deliver 20.8 and 21.0 input spikes and are the same event to this chip — it hears loudness over time, not language. Six of seven pre-registered predictions held; the one that failed produced the correction. Scope-lock per Maya-Meta P2: reproducible signatures within THIS architecture under THIS protocol — Property, not Law; one chip, one encoding, one corpus, three calibrations; a characterization of detection reliability, not new physics. Series: Part of the Maya-Sutra Series — Nexus Learning Labs’ hardware-native track, characterizing neuromorphic substrates directly (one substrate, one honest characterization at a time). The arc: let go (P1) → forget (P2) → remember (P3) → listen (P4). Research characterization only. Links: GitHub Repository (private — to request access: email research@nexuslearninglabs.in with subject "Code Access Request — Maya-Sutra-P4" and your research context) | Interactive Dashboard | FAQ | Full Series Index — nexuslearninglabs.in 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