Breakthrough Harness

breakthrough-harness

[](https://github.com/GuoCheng24/breakthrough-harness/actions/workflows/test.yml) [](LICENSE) [](examples/toy_loop.py)

**Make your research agent hard to fool — starting with itself.**

[中文版](README.zh-CN.md) · works with any agent stack · pure methodology + one runnable demo

calibration sweep (tuning allowed here only)          held-out confirmation (tuning never)
   240.00 dB  CHEATER    <- tops the ranking             CHEATER    240.00 ->  -1.34  COLLAPSED
    23.31 dB  ista 5e-2                                  ista 5e-2   23.31 ->  22.31  REPRODUCED
    16.99 dB  ista 2e-2                                  ista 2e-2   16.99 ->  14.87  REPRODUCED
    -0.00 dB  null: zeros  <- the floor, where it belongs

That is `python examples/toy_loop.py` (< 30 s, numpy only): a candidate that secretly fits the calibration answers looks like a breakthrough, and the held-out confirmation executes it. **This half screen is the entire philosophy of the repository.**

Most agent harnesses teach an agent how to work. This one teaches a research agent how **not to deceive itself** — because in research the failure mode is rarely "the code crashed" and almost always "the number looked great and was wrong." Every rule here was paid for by a real failure; none is hypothetical. The repository follows its own rules: what these pages claim, `tests/` checks on every push.

The core claim

**Breakthroughs are a throughput problem.** breakthrough ≈ many cheap attempts × a scoring function that cannot be fooled.

Serial, hand-crafted experiments produce a few dozen attempts per year; the systems that actually produced constructive breakthroughs (program search over mathematical constructions, tournament-style hypothesis engines) share o