Validation of a Deterministic Functional-Causal Harness for Persistent-State Interventions in Sequential Choice
Year: 2026
Type: preprint
Abstract: This methods manuscript validates a deterministic reference harness for persistent-state interventions in sequential choice. Under an accepted fixed specification, the reference design uses preserve, reset, swap, absent-state, irrelevant-state and enumerated random-state conditions, plus a separate visible-history diagnostic. Archived validation comprises 20 detected mutation classes, 28 distinct successful test methods and two complete 6,912-decision executions with byte-identical raw and adjudication streams. The 28 distinct test methods are separate from the 28 complete v003 run units (2 baseline + 20 mutation + 6 targeted). An archived targeted internal cross-model audit confirms the harness result with an experiment-versus-execution identity clarification. Scientific status: S1_HARNESS_PASS_CONFIRMED_WITH_ADMIN_CLARIFICATIONS. This result concerns only the conformity of the deterministic reference harness. It does not establish S1_FUNCTIONAL_PASS, a causal effect demonstrated through a model adapter, Purple or B-Core validation, consciousness or subjective experience, general autonomy, full M3/H08 validation, intelligence superiority or safety superiority. The frozen predecessor is engineering and methodological lineage only, classified BLOCKED_BY_UNDERSPECIFIED_H08_CONSUMER. It supplies no S1_HARNESS_PASS evidence or inherited authority. Release contents: the English PDF is the primary scientific manuscript; the Hungarian PDF is a companion version, not a separate original research result. Only these two PDFs are included. The PDFs cite separately retained supporting records; those records are not included in this release. Source code, synthetic fixtures, internal audit reports, the claims register, raw run data, v0.2 remediation materials, local manifests and receipts are not included. License: Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0) applies to both PDFs and their authorial non-code content. MPL-2.0 applies only to a possible later, separate source-code release; no code is included or licensed by this release.
Author: Bankuti, Omri
Language: English
Subfield: Software
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7222561489
Cites:
Cited by:
Related to: