Epistemic Engine verifies AI code and predicts what will break

Epistemic Engine is a deterministic Python tool that models a codebase as a set of justified beliefs — claims like "this function authenticates the caller" — reconstructed from git history. Its first tool, ee guard, scans staged or full repos for weak crypto, injection flaws, unsafe sinks, hardcoded secrets, and misconfigurations, blocking risky commits with SARIF output and ready-made CI workflows. It's built specifically to catch issues in AI-generated code before it merges.

The second tool, ee predict-chain, forecasts which beliefs are about to collapse, why, roughly when, and the cost of fixing now versus later. A companion command, ee calibrate, back-tests the forecasting model against the repo's own history and reports whether it actually beats guessing — refusing to present unreliable predictions as fact. The whole system runs offline with no LLM calls, is fully deterministic, and ships with 167 tests.

For engineering teams, the key value is honesty about uncertainty: gates are evaluated on the lower bound of a Wilson confidence interval rather than point estimates, and cost/ROI figures are explicitly labeled as assumptions. Installable via pip under Apache-2.0, it's been field-tested on 1,500+ production files and includes a baseline mode for adopting it incrementally on existing repos.