Maya-Sutra-P6: Karma on Silicon Inherited Synaptic State Across Nightly Calibrations of BrainScaleS-2: Expression, Composition, Depletion, and the Drift a Fixed Ledger Suffers from the Body Alone

Maya-Sutra P6: Karma on Silicon carries synaptic state across the boundary every other BrainScaleS-2 experiment respects. The platform regenerates its analog calibration nightly, and every published experiment on it begins from fresh weights — the field runs the no-carryover control by default. The construct is Karma, used as the factor structure rather than as decoration: an append-only ledger (Sanchita), weights carried in at session start (Prarabdha), and the plasticity applied today (Kriyamana), crossed 2×2 across five consecutive sessions on five independently regenerated calibrations. Synaptic weights on this substrate are 6-bit digital SRAM, so the ledger round-trip is exact by construction; what changes overnight is the analog body that expresses it. Four results. (1) Inherited state expresses, and always weaker: across four full nights, every neuron that fired when fresh fired less when it inherited — 20 of 20 paired comparisons, 4 of 4 nights (exact sign test across nights, one-sided p = 0.0625). The caveat is stated in Results rather than buried: on nights three and four the ledger had decayed below the response cliff, so the carried arms read zero trivially; the graded evidence is nights one and two. (2) The substrate drift floor — the measurement this paper exists for: holding the digital weights exactly constant and changing only the nightly calibration moves per-neuron responses by up to 1.33 population spikes, about 19% of the largest response, with zero weight change. Two of six neurons drifted not at all across four nights; two flickered between silence and firing. This is a floor no learning rule creates and none can remove — any continual-learning scheme deployed on analog hardware inherits it whether or not it is measured, and weight anchoring cannot suppress it because the weights are not what moved. (3) Composition holds: decaying a carried ledger on-chip gives the same response as writing the already-decayed ledger directly, on all four nights, largest deviation 0.67 population spikes — extending Paper 5's hardware-equals-arithmetic result across a session boundary, a recalibration and a full write-read cycle. (4) Inherited state depletes: the carried lineage extinguishes entirely by the third full night. The pre-registered window for that extinction was ledger entries four to six and it arrived at entry three — mechanism confirmed, count one entry early, reported as a miss. Predictions were locked on 23 August 2026, before any hypothesis-bearing datum existed. The campaign ran five sessions with five gate passes, five instrument-check passes and no platform incidents. Nearest prior art, cited and distinguished: Billaudelle et al. 2021 transferred learned weights between circuit instances within one calibration (1.2% deviation); Weis et al. 2020 retrained after deliberate decalibration. Neither carries state across nightly recalibrations, and neither crosses carryover with plasticity. Scope-lock per Maya-Meta P2: reproducible signatures within THIS architecture under THIS protocol — Property, not Law. 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) → the floor and the price of forgetting (P5) → carry (P6). Research characterization only. Links: GitHub Repository (private — to request access: email research@nexuslearninglabs.in with subject "Code Access Request — Maya-Sutra-P6" and your research context) | 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 (embedded in all canonical hyperparameters) · Canary: MayaNexusVS2026NLL_Bengaluru