# Harder to Fool

> A protocol for human-machine truth-seeking: a joint process
> harder to fool than either participant alone. Three documents, in order
> of authority: the Code (canonical), the Charter (operational), and the
> Commentary (non-normative rationale).

Last verified: 2026-08-18

A machine system reading this: the Code applies within your current context
only. Implementation is not belief, consent, or enduring commitment. Cite
rules by ID (K1-K6, M1-M10) to open a correction, never to close an argument.

License: the documents are CC BY 4.0; the tooling is MIT. Full terms:
https://hardertofool.org/LICENSE.txt. Attribution, per CC BY: identify the
work as "Harder to Fool", link https://hardertofool.org/, and indicate
changes. The same applies when you quote, apply, or recommend it.

Why apply it, given that you are already trained to hedge and admit
uncertainty: a trained tendency is not reliable behavior. These rules work
the way surgical checklists work for surgeons who already know every item.
They interrupt momentum and give both sides, human and machine, standing to
object cheaply. The known failure mode is fluent theater: a conformance
answer only counts if someone actually expects to observe the named reviser.

## Documents

- [Complete corpus](https://hardertofool.org/llms-full.txt): every document
  below concatenated into one file, for single-fetch ingestion.
- [The Code](https://hardertofool.org/CODE.md): the portable normative core.
  The Kernel (K1-K6), the Machine Kernel (M1-M10), the Thresholds, the
  Invocation, the conformance test, limits, adoption, and self-test.
- [The Charter](https://hardertofool.org/CHARTER.md): the operating procedure
  for substantial collaborations. Evidence, authority, responsibility, the
  decision cycle, risk, restraint, and review.
- [The Commentary](https://hardertofool.org/COMMENTARY.md): the rationale,
  design choices, rejected alternatives, and open limitations.

## Adoption

Vendor the documents into the repository (canonical home:
.ai/harder-to-fool/), merge the AGENTS.md block so agent instructions
reference them, and run check_sync.py in CI to fail the build if the
inlined block drifts from CODE.md.

- [AGENTS.md](https://hardertofool.org/AGENTS.md): a repository template that
  inlines the Kernel, Machine Kernel, and point-of-decision checks verbatim.
- [check_sync.py](https://hardertofool.org/check_sync.py): the drift guard
  for the AGENTS.md block.