NC

ncrofford.com

Current focus

Parity

Describe the work. Parity runs it.

Most operational work is a person reading a document and deciding what happens next. Invoices, claims, submissions, filings, remittances, authorizations. The tools built for that work have mostly made the reading faster, which helps, and leaves the cost structure exactly where it was.

Parity executes the work instead. You describe what needs to happen in plain language, or hand it a file and say what to do with it. That becomes a repeatable flow that runs unattended, returns the same result every time, and produces a tamper-evident record of what it did that anyone can verify without an account. No code, no templates, no configuration. Anything the system is not confident about gets flagged for a person rather than guessed at, which is the part that makes the rest of it safe to use.

Parity launched recently. The systems underneath it were built and tested well before that, and the research behind them is public.

parity.work

The systems underneath it

Two layers, built separately and for different reasons. Everything above is composed from them.

Probity

A runtime governance control plane for enforcing, recording, and reconstructing AI decision behavior.

Probity is a runtime governance infrastructure layer for AI systems. It evaluates every action against policy during execution, enforces constraints in real time, and produces tamper-evident decision evidence that can be reconstructed independently of the original run. Across the Probity experiments, existing architectures fail to produce reliable constrained execution, and external guardrails render systems non-functional. Probity addresses this by making governance a first-class system property rather than an external wrapper.

  • Per-action policy enforcement (ALLOW / REQUIRE / DENY) during execution
  • Tamper-evident Policy Restriction Evidence (PRE) with full decision trace reconstruction
  • Domain-invariant governance infrastructure across models, workflows, and policy types
Open source SDK

Opus

A compiled execution layer that turns human intent into deterministic, governed infrastructure.

Opus is a compiled execution system that separates reasoning from execution in AI-driven workflows. It transforms natural-language operational intent into deterministic execution artifacts that run with zero inference, full replayability, and cryptographic governance. Across the Opus experiments, compiled execution achieves perfect determinism, complete policy-change visibility, and scale-invariant guarantees. As workflows accumulate, they compose into governed pipelines that evolve under change and form a persistent operational substrate.

  • Intent → IR → compiled artifact pipeline with deterministic execution (no runtime inference)
  • Structural governance: diff, replay, and causal impact analysis by construction
  • Composable and accumulating workflows that evolve under governed recompilation