Monotonic Proof-State Advancement for Distributed Workflow Verification
Sterling Morrison · GDF Research · DOI 10.5281/zenodo.19102106
Abstract
The substrate the agent-memory work is built on. A method for verifying distributed workflows when no single observer can witness every step: a monotonic confirmation-level hierarchy that prevents proof-state regression, threshold-triggered evaluator dispatch that suppresses premature policy evaluation, and boundary-capped proof resolution that distinguishes architecturally unobservable evidence from true policy failure. Illustrated with a payment-settlement case study.
The question
A payment passes through a gateway, a processor, a bank and a ledger. No single system can see every step, yet someone has to say whether the payment settled. How do you verify a workflow like that without treating missing evidence as failure, or a single acknowledgment as proof?
What it proposes
Three mechanisms. A confirmation ladder runs from observed intent to attempted action, acknowledged action, independently confirmed state and end-to-end completion, and a workflow's proof state can only move up it. Expensive policy checks are dispatched only once enough evidence exists. And when evidence sits beyond what any observer the system controls can reach, the check is marked boundary-limited instead of failed.
The paper illustrates the method with a payment-settlement case study. It's the substrate the later agent-memory and verification papers are built on.
What later work found
Building it in Evidence-Graded Memory exposed a gap: dispatching on confidence level alone can evaluate a check before the evidence it reads has arrived. Under out-of-order delivery that turned a perfect judge into a 91% one. Gating dispatch on the evidence frontier restored 100%. That paper also replaced the single boundary per workflow with per-check caps.
Limitations
- A method paper, illustrated with a case study. The quantitative evaluation of the design is in Evidence-Graded Memory.
Terms
Questions this paper answers
Cite
Morrison, S. (2026). Monotonic Proof-State Advancement for Distributed Workflow Verification. Zenodo. https://doi.org/10.5281/zenodo.19102106
@misc{morrison2026monotonic,
author = {Morrison, Sterling},
title = {Monotonic Proof-State Advancement for Distributed Workflow Verification},
year = {2026},
publisher = {Zenodo},
doi = {10.5281/zenodo.19102106},
url = {https://doi.org/10.5281/zenodo.19102106}
}