Anti-Rationalization Rules · v1.0.0 · 2026-11-14 · sha256 77ddd1d6b4bf1455

Anti-Rationalization Rules v1.0.0A

Immutable. This exact content never changes and is served at /api/v1/blob/77ddd1d6b4bf1455.

No tool to install first
claude plugin marketplace add https://markdownregistry.com/mp/a/art_3dcuhrxuyirwklrh/77ddd1d6b4bf145580d06bd30580f11124c498908bccd75c98a8977b4a34de16.json

Then /plugin install anti-rationalization-rules@markdownregistry-3dcuhrxuyirwklrh in Claude Code. It installs this exact version, the one graded A here, pinned to the commit it was graded at, and leaves it there.

---
name: Anti-Rationalization Rules
description: Prevents AI agents from inventing justifications for test failures, unexpected behaviors, or unmet specifications. Enforces physical evidence verification.
category: development
version: 1.0.0
author: Taqi Molavi
license: MIT
---

# Anti-Rationalization Rules & Verification Protocol

## Overview

AI models often rationalize failures: claiming a test failed "due to a pre-existing issue", "because the environment lacks X", or "this warning is harmless". Anti-Rationalization Rules strictly prohibit speculative excuses and enforce empirical, physical verification.

## Forbidden Rationalizations

- ❌ *"The test failed because the environment is different, so the code is actually correct."*
- ❌ *"This error is just a deprecation warning and won't affect production."*
- ❌ *"The mock failed, but in a real system it would definitely pass."*
- ❌ *"I didn't run the tests because the change was minor and trivial."*

## The Golden Rule of Verification

> **"It is strictly forbidden to guess. Either physical evidence confirms the change, or the task is NOT complete."**

## Verification Checklist

1. **Reproduction Script**: For any bug fix, write a test that fails before the fix.
2. **Deterministic Green**: Run the test suite and verify 100% pass rate.
3. **Diff Minimization**: Verify zero accidental edits to unrelated files.
4. **Physical Output**: Capture exit codes and terminal logs as permanent evidence.