Notes on the project's research posture: how the benchmark is being prepared for preregistration, and whose shoulders the engine stands on. Everything here cites the claims registry and the canonical repository documents.
How IrrevonBench uses a two-stage freeze before live and confirmatory evidence, while disclosing synthetic development pilots and a falsification criterion that would declare the engine unnecessary.
Irrevon stands on decades of work — outboxes, idempotency, sagas, durable execution, agent-effect staging, replay, and reconciliation. This is the dated survey and its narrow inference.
No preprint exists yet. The publication path — arXiv preprint, an agents/reliability workshop, the NeurIPS Datasets & Benchmarks track — is a plan recorded in the master document, not an achievement. When a preprint exists, it will be listed here with its identifier.