GraphOS Labyrinth Proof-Carrying Decision Continuity with BAPT and BAPT-F over Evolving Evidence Graphs
Year: 2026
Type: article
Abstract: Decision-support systems increasingly operate over evidence that changes after a decision is made. A decision may be supported at time t, yet its cited certificate can expire, a supplier can change ownership, a source can be retracted, or a regulation can be amended. Point-in-time proofs answer whether a decision was supported when it was produced; they do not answer whether it remains recoverable, which future changes invalidate it, or whether one executable policy covers the declared future. This paper presents BAPT-F, a finite, future-bounded decision-continuity framework that composes four layers: an evidence graph (GraphOS), a constrained decision and planning layer (Labyrinth, including CP-SAT), a proof-carrying decision contract (BAPT, Belief-Aware Proof Transitions), and a future-policy certificate (BAPT-F). For a declared finite set of optional future events, BAPT-F produces a portable bundle that establishes, for the declared model, four properties: 1. at every modeled checkpoint, the decision is recoverable by an authorized add-only action set within declared cost/time/resource limits; 2. a finite adaptive policy over event histories is executable; 3. recovery, approval, execution, revalidation and future policy refer to the same decision, graph, action set, cost and state evolution; 4. an offline verifier can check these properties without trusting the producer. The contribution is a scoped decision-continuity protocol and its cross-artifact handoff, not a new optimizer. CP-SAT and Labyrinth propose candidates; BAPT and BAPT-F certify the claimed decision/recovery objects; the verifier checks the bundle. The evaluated class is finite, discrete and add-only. We report synthetic two-domain evidence, a 14-case mutation matrix, a bounded state-merging probe and an end-to-end API/UI flow. We explicitly do not claim general optimality, unmodeled-event prediction, real-domain validation or production certification.
Author: Skouras, Panos
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7225574922