Skip to content

tsai-sc

Harness where Jev plays the first StarCraft shareware mission from structured state, with recorded per-decision action probabilities.

BenchmarkresearchMIT

Overview

tsai-sc is a harness in which Jev plays Strongarm, the first combat mission in the original StarCraft shareware campaign. The 1998 Windows executable runs inside BottleShip; the harness reads structured game state, asks jev-latest to choose a command, and executes it through ordinary mouse and keyboard input. The game is paused during state reads and inference. It was inspired by TypeSafe's Doom demo but uses independent code.

The README reports that a development attempt completed the mission, with the original victory screen captured and visually reviewed and an evidence verifier passing. The verified run made 421 model decisions over about 17.5 minutes, with a median API latency of about 383 ms. Videos and a reproducible evidence bundle are published. The author stresses it is a bounded mission experiment, not a benchmark of real-time competitive play.

What the decision model decides

TypeQuestionAnswers
ChoiceWhich command should be issued next (separate Economy and Army decisions)?—

How it works

  1. Pause the game and read structured state.
  2. Ask Jev to choose an Economy or Army command.
  3. Execute it through mouse and keyboard input.
  4. Record probabilities, frames and traces for verification.

Strengths

  • Verified run with published evidence bundle and videos.
  • Per-decision probabilities are recorded and overlaid.
  • Measured latency and token accounting are reported.

Caveats

  • Game pauses during inference; not real-time play.
  • A single mission; not a competitive benchmark.
  • Earlier attempts had interrupted runs and runtime faults.

At a glance

Type
Benchmark
Status
research
Runtime
harness around BottleShip-emulated Windows game
Has code
yes
License
MIT

Use cases and integrations

How curated lists describe it

  • Gaming: drives original StarCraft shareware through keyboard and mouse with Jev action probabilities recorded per decision.

    yibie/awesome-jev
  • Structured-state harness for the original StarCraft shareware campaign, with verified run and probability traces.

    AnotiaWang/awesome-decision-models
  • A Jev harness for StarCraft shareware’s Strongarm mission, pausing the game during state reads and inference.

    logicrw/awesome-jev-projects
  • Original StarCraft shareware controlled with recorded Jev action probabilities.

    cobanov/awesome-jev
  • TypeSafe Jev controls original StarCraft shareware through keyboard and mouse with recorded action probabilities.

    hellogumbo/awesome-jev
  • Jev plays the first StarCraft shareware mission, with its recorded action probabilities.

    MrJev/awesome-jev

Frequently asked questions

Did Jev win?

Yes, the README reports a verified completion of the Strongarm mission.

Does it play in real time?

No, the game is paused while state is read and Jev decides.

Is the run reproducible?

An evidence bundle with decisions, traces and frames is published for checking.

Related projects

  • Browser 2D driving sim where Jev answers lane, speed, hazard and pedestrian-yield questions every 200 ms, with an LLM comparison mode.

    ProjectprototypeJavaScript
  • Experimental Conway's Life game where Jev makes each move with one boolean per legal cell, plus a broker that wraps any LLM as Jev.

    ProjectprototypeTypeScript
  • Evaluates Jev as a fixed policy in Gymnasium and Atari games, picking a legal action each step via one Choice question.

    ProjectresearchPython
  • Model Kombat: a 12,000-neuron fly connectome fights Jev in mk.js, with Jev picking one of seven moves about five times a second.

    ProjectresearchPython
  • Fifteen live arcade games where Jev and LLMs play each other, led by WikiRace in which Jev picks only real links.

    ProjectbetaTypeScript
  • Real-time bot battles where you write a fighter's strategy in plain English and a System One model pilots it several times a second.

    ProjectbetaTypeScript