Evaluates counterfactual verifier conformance, suggesting stability in autonomous-agent actions.
ReceiptOS is a recomputable evidence substrate for autonomous-agent actions. It defines an Evidence Capsule whose canonicalized, anchor-stripped contents determine a receipt root; admissibility is a gate rather than a score, and anchoring remains a pluggable commitment layer. The v0.8 artifact retains the v0.6 capsule model, threat analysis, and cross-specification composition while adding a formal, executable account of counterfactual verifier behavior.Counterfactual Conformance v0 evaluates ten authenticated decision-boundary witnesses in an exact ten-member Deterministic Counterfactual Neighborhood: eight semantic-neighbor members, one boundary-rejection member, and one composite-relational member. Counterfactual Traversal Stability v1 evaluates the same neighborhood under twelve frozen schedules, yielding 120/120 stable scheduled evaluations, including 110 non-canonical comparisons, with first-position profile coverage of 10/10 members. A separate runner-invoked fresh-worker repeat of the authenticated canonical order matched 10/10 member records and is reported separately from the frozen 120-evaluation profile.In parallel, Recursive Singleton Fold provides a co-equal mechanically closed result for terminal-eligibility semantics. The paper also reports a bounded TEE experimental evidence class and a worked ReceiptOS × WYRIWE composition with byte-exact recomputation across three separately written implementations. The claims are intentionally bounded: ReceiptOS does not establish source truth, universal verifier correctness, legal admissibility, or downstream policy correctness.
No takes yet. Share an insight, caveat, or question.
Pavlo Tvardovskyi (2026) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: