Skip to content

Releases

v0.1.0 — Minesweeper Thinking Lab research preview

Choose a tag to compare

@imom39a imom39a released this 22 Sep 13:34

Minesweeper Thinking Lab compares Jev, LLMs, hybrid decision policies, and deterministic solvers in three experiments:

  1. LLM vs Jev: browser comparison using the original game policy.
  2. Jev alone: browser experiment with large boards and bounded model input.
  3. Jev + LLM: CLI benchmark with heuristic, code-only, proof-only, Jev, LLM, and hybrid policies.

The stronger solver and hybrid policy run through the CLI. Browser integration is planned. Python 3.10+ is required; there are no third-party runtime dependencies.

The stronger solver completed 52/100 held-out expert boards versus 32/100 for the original heuristic. The four-state live pilot did not establish a hybrid reliability advantage. Full methods and limitations are in the report.

Documentation

These links point to the maintained documentation:

The 11 recorded result files are preserved with this release. Source attribution and license details are in ATTRIBUTION.md.