ExperimentalGraph ReasoningEvidence, Audit and Replay

Graph Reasoning Runtime

An experimental reasoning runtime separating a planner, a transaction log and an evidence ledger, with constraint enforcement over structured domain knowledge.

Abstract

We are prototyping a reasoning runtime in which planning is separated from execution: a planner proposes steps, a transaction log records them, and an evidence ledger keeps supporting material alongside intermediate state. Domain packs supply structured knowledge and constraints. The aim is reproducible, inspectable reasoning; this is an early experimental system, not a product.

Problem & motivation

Free-form LLM reasoning is hard to reproduce or audit: intermediate steps are not recorded in a structured, replayable way, and domain constraints are applied inconsistently.

Research questions

Does separating planning from a transaction log improve reproducibility of multi-step reasoning?
Can domain constraints be enforced without collapsing the reasoning space unhelpfully?

Methods

Implement a planner / transaction-log / evidence-ledger separation.
Encode domain packs with explicit constraints.
Replay logged runs to test determinism.

Limitations

Experimental; interfaces and results are unstable and may change substantially.
No performance claims; no external benchmark has been run.

Disclosures

Funding: Infrastructure support provided by Octopus Core Pty Ltd.
Conflicts of interest: No specific commercial conflict identified for this experimental line.