Harder to Fool

Active

A practical framework for human-machine collaboration that makes the joint process harder to fool than either participant alone.

Visit Project → Source Code
agents framework methodology

A practical framework for human-machine collaboration. Its aim is not to eliminate bias, no process can, but to make the joint process harder to fool than either participant alone.

Reality is the reference. Models are instruments. Confidence must be earned. Correction is progress.

The framework has three layers, in order of authority:

  1. CODE.md: the portable normative core, with kernels, maxims, thresholds, a conformance test, and a self-test.
  2. CHARTER.md: the normative operating procedure for substantial collaborations, covering evidence, authority, responsibility, the decision cycle, risk, restraint, and review.
  3. COMMENTARY.md: the non-normative rationale, design choices, rejected alternatives, and open limitations.

It ships as drop-in files for a repository (.ai/harder-to-fool/ plus an AGENTS.md block), with a CI check that fails if the inlined block drifts from the Code.