We build evidence-bound infrastructure for autonomous execution. ProofRail is our flagship: a tested protocol pilot for single-use authority, replay and drift denial, with explicit failure boundaries.
Start with the ProofRail v0.6.0 pilot release: installable wheel, source distribution, checksums and exact-commit workflow metadata. Run the 10-minute break test and inspect the limitations before adoption.
We propose one bounded evaluation using synthetic data: install the tagged artifact, demonstrate one authorized action and its replay/drift failures, interrupt and restart the selected integration, then rehearse an upgrade. Agree on acceptance criteria and record the exact commit, results, latency and recovery behavior. Contact prettybusysolutions@gmail.com to scope a pilot or independent review. No engagement or customer result is implied by this offer.
ProofRail's source tests and installed-wheel reproduction run in author-controlled CI. Independent semantic review and external reproduction remain pending. A release, a signature and green CI do not establish independent assurance, production readiness or market adoption. ESRM source-conformance gates are separate from legacy ProofRail semantics.
| Repository | Current evidence | Maturity and remaining gate |
|---|---|---|
| ProofRail | Tagged pilot artifacts and repeatable break tests | Protocol pilot; independent review and operational key/checkpoint hardening pending |
| VerifiAgent | Truthful tri-state verdicts, authenticated durable jobs and constrained-worker configuration tests | Operator pilot; real isolated-worker and provider integration pending |
| Adversarial Coder | Known-bad target and missing-evidence tests | Framework alpha; adapters and external deployment validation pending |
| IMOS | Owner-scoped SQLite decisions, restart and schema-boundary tests | Service pilot; graph/ingest and legacy migration unvalidated |
| LeakLock | Flask scanner, access/payment tests and PostgreSQL/Redis CI | Application pilot; real checkout, migration and deployed fulfillment pending |
| DenialNet | Account-bound pattern access and atomic settlement tests with PostgreSQL/Redis CI | Application pilot; real provider flow, clinical validity and federation unproven |
| Thermodynamic Substrate | Numerical, concurrent/restart invariants and reproducible SQLite overhead comparison | Research prototype; no demonstrated useful-work advantage |
| Residue Classifier | Versioned filename-heuristic CLI and explicit report limits | Utility pilot; real-workspace accuracy unmeasured |
| x402 MCP Server | Discovery and synthetic payment-boundary tests | Package prototype; real settlement/delivery/replay and synchronized distribution pending |
Evidence before claims. Required behavior checks on the exact reviewed head. Missing evidence returns PARTIAL; demonstrated failure returns FAIL. Source access and Git cloning do not establish operational restoration. We publish measured limits alongside results.