Preregistered Somatic-Marker Ablation in a Persistent Bio-Inspired Agent Substrate: A Negative Result on the Verbal-Output Axis

Abstract

Status: published 2026-08-30 (deposit 21218491; DOI 10.5281/zenodo.21218491). Corrected 2026-08-29 after external adversarial review. A persistent bio-inspired agent substrate (Hebbian/STDP plasticity, hormone-analogue state vector, operator-bonded longitudinal identity) carried a core claim derived from Damasio's somatic-marker hypothesis: that its measured hormonal state is cognitive, not decorative. The claim was operationalised as a preregistered, frozen, four-cell ablation design (stressed / basal / soul-block-ablated / random-hormones; 12 frozen judgement-eliciting prompts × 4 cells = 48 generations; deterministic surface metrics plus a TF-IDF cosine instrument — a pure-Python substitute for the preregistered embedder, declared in RESULT.json; measurement-only on snapshot copies, the live substrate read but never written). The core preregistered falsifier (H1) fails cleanly and the frozen decision table returns DECORATIVE on H1 alone. The ablated cell is indistinguishable on the frozen metrics from both the calm and the stressed cell (centroid cosines A-B 0.8003, B-C 0.7977, A-C 0.7849; all pairwise values within 0.0154; no formal statistical test was applied); removing the entire somatic state and flipping cortisol by an order of magnitude both produced negligible movement. The other two falsifier families add no independent support: H2 nominally passes a mis-calibrated threshold and is uninformative, and the H3 specificity control is reported as contaminated and underpowered (two dispatch windows; 7 of 12 random-hormone generations discuss the experiment's own verdict in their text; 8 of 12 ran on a second, initially undeclared snapshot; on the clean first-window subset, n=4, the preregistered direction inverts). Verdict, per the frozen decision table: DECORATIVE on the verbal-output axis under these prompts and metrics. The deposit publishes the negative result in full — preregistration, falsifier evaluation, execution and evaluation history, and the method limitations that scope (but do not rescue) the claim — as a registered-report-style honest negative. A subsequent, separately preregistered experiment on the action axis produced a behavioural positive in a laboratory falsifier test with synthetic hormone states; its live shadow evaluation has not yet observed a real stress excursion and it remains under promotion criteria; it will be reported separately and does not alter the present verdict on the verbal axis.