#!/usr/bin/env npx tsx
/**
 * MR-GSM8K Benchmark — Real LLM A/B Test
 *
 * Uses 21 problems from MR-GSM8K where the <<expr=result>> arithmetic is
 * provably wrong. Tests: does the LLM rubber-stamp the error, and does
 * the formal verifier catch it?
 *
 * Usage: npx tsx src/lib/__tests__/mr-gsm8k-benchmark.ts
 */
export {};