{
  "title": "Token Tide \u00b7 Batch Dispatch",
  "repository": "https://github.com/mourner/flatqueue",
  "source_reference": "https://github.com/mourner/flatqueue/tree/eee96a6df7705e57f8dbb3702aa3efc40bb6e4b5",
  "revision": "eee96a6df7705e57f8dbb3702aa3efc40bb6e4b5",
  "stars_observed": 144,
  "license": "MIT (original game); ISC (FlatQueue)",
  "pushed_at": "2026-07-21T11:18:57Z",
  "tags": [
    "github-source",
    "game"
  ],
  "description": "Original inference dispatch puzzle: batch same-model requests under an eight-block memory cap, meet arrivals and finish-by deadlines, pay cold-start time and share energy costs. Eight verified solvable shifts, a forecast timeline, exact feasible hints, undo, medals and device saves. Uses the actual FlatQueue event engine; no model or backend runs. Fictional integer rules, not hardware measurements.",
  "provenance": "Original fallback game inspired by current 2026 inference demand. GitHub repository is the bundled scheduling dependency, not an upstream game. FlatQueue by Volodymyr Agafonkin is used to order request arrivals, deadline expiration and batch completion events."
}

Original game files: index.html, style.css, app.js, engine.mjs (MIT). flatqueue.mjs is the upstream index.js, byte-for-byte, at the pinned revision (ISC). Event priorities order arrivals, then completion at its deadline, then expiration after that tick. The original solver explores valid same-model batches plus idle actions.

Topic references inspected November 3, 2026:
https://www.mckinsey.com/industries/semiconductors/our-insights/scaling-ai-inference (October 5, 2026)
https://vllm.ai/blog (October 2026 inference scheduling updates)
These are thematic references, not numerical sources for the fictional game rules. No text, art or model weights copied.