This folder holds all of GPTD's documentation. New here? GPTD is a tower-defense game that simulates running an LLM-inference data center — ▶ play it now, or read the project overview first.
Two files live at the repo root, not here: ../README.md (the project landing page — what GPTD is, how to run it) and ../AGENTS.md (the working rules for contributors and AI coding agents, kept at the root so tools discover it).
| Doc | What it is |
|---|---|
| manual/en.md | The player manual — what the game is and how to play it, no prior knowledge needed. English (primary). |
| manual/zh-TW.md | The player manual, 正體中文（台灣）. |
| manual/README.md | Language picker for the manual. |
| AGENT-BRIDGE.md | Let a local terminal agent (Claude Code / Codex) play the game in a browser tab you already have open — the protocol, action vocabulary, and /state snapshot shape. |
| Doc | What it is |
|---|---|
| DESIGN.md | The canonical Game Design Document — the readable design of record. Start here to understand the design. |
| BLUEPRINT.md | The authoritative design spec (§0–§7): the unified data model, the serving spine, the dual-clock / SLO physics, and every subsystem — the deepest implementation layer beneath the GDD. (正體中文.) |
| MATH.md | Quick reference: every formula, constant, and variable the simulation uses, extracted from the code (glossary → constants → formulas by subsystem). |
| SYSTEM-MENU.md | Design spec for the implemented system/pause menu, settings & options (audio volume, language, accessibility, gameplay), content browsers (How to Play / Codex / About, with Achievements reserved), and the persistence layer. |
GPTD's slogan is "the board is the metaphor, the numbers are real." These docs are where the numbers come from and how they map to reality.
| Doc | What it is |
|---|---|
| REALISM.md | The realism ledger: each mechanic → its real-world basis, the literature, what GPTD ships, and where it knowingly deviates. |
| REALISM-MODELS.md | The real-model grounding: how the model roster, capability calibration, and post-training tie to real open-weight models and public benchmarks. |
| MODEL-CATALOG.md | The fact-checked model-roster reference — real developer, release, license, link, the five benchmarks, and lineage for every model the in-game roster is drawn from. |
| PARETO.md | Roster Pareto analysis — the five size↔capability frontiers (mermaid charts), per-size-bucket ceilings, and a frontier-coverage review of the curated models. |
| REFERENCE-DOSSIER.md | Deep background: the real-world 2024–2026 facts about LLM inference, post-training, safety, serving systems, and hardware the simulation is calibrated against. The blueprint's §ref citations resolve here. (正體中文; not required to play.) |