coolguyandClaude Opus 4.8
8b7348cbce
The rival's policy now exposes a ranking, which drives both its own move and a new HINT button: it highlights the card to play, where to put it, and where to draw, with the model's confidence. Undo and redo work per action rather than per turn — picking a card, choosing its destination, and drawing are separate steps, as are the rival's moves — so a finished game can be stepped back through from the result screen. The rival is suspended while undone moves are pending, and PLAY FROM HERE resumes from the reviewed position. Cards now travel between zones instead of teleporting: a motion layer measures each card's old and new position and animates the difference, flying cards out of the deck face-down and flipping them over, and back into it on undo. Also: the result screen gets a per-expedition score breakdown mirroring human_play.py, and a rival card revealed by a discard-pile draw no longer renders at full size in the card-back-sized rival hand row. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01LmyprzuzanXRhpomc3Ga1i
2.8 KiB
COOLRL Lost Cities Web
Browser-only Lost Cities client. The rules engine, observation builder, and PPO inference all run on the device; there is no application server.
The verified final JAX PPO policy is committed at
public/models/jax-ppo.onnx (3.1 MB). Vite copies it to the static build, and
the app resolves the asset relative to the deployed site so it works on GitHub
Pages, GitLab Pages, or a normal web root. Its size and SHA-256 are recorded in
public/models/jax-ppo.json.
Pushes to main build and publish the client through the repository's GitHub
Pages and GitLab Pages workflows. Each workflow supplies the correct base URL
for its host.
Seeded deals
Every game is dealt from a seed, shown in the menu and kept in the URL as
?seed=<seed>. Loading that URL — or typing the seed into the menu — replays
the exact same deal, so a game can be shared, replayed, or reported with a bug.
Seeds are arbitrary text; the deal is derived from a deterministic PRNG in
src/game/random.ts and is independent of the Python shuffle bank.
The seed shuffles the deck and nothing else. The policy is deterministic: it always plays its highest-logit legal action, so the same deal and the same moves reproduce the same game.
Hints, undo, and review
HINT asks the same policy that drives the rival what it would play in your
seat, and highlights the card, its destination, and the draw source (with the
policy's confidence when the ONNX model is loaded).
Undo and redo work per action, not per turn: choosing a card, choosing its
destination, and drawing are separate steps, and each is undone on its own. The
rival's moves are steps too, so a finished game can be stepped through from the
result screen (REVIEW GAME) all the way back to the deal. While undone moves
are pending the rival is suspended; PLAY FROM HERE drops them and resumes play
from the position on screen.
Setup
Install and run the checked-in final policy:
cd web
npm ci
npm run dev
Build the same static bundle used by a host:
npm run build
npm run preview
To replace the shipped policy, export a verified Orbax checkpoint from the repository root. The exporter writes both the ONNX file and its public metadata manifest:
uv run --with onnx scripts/export_jax_ppo_onnx.py \
--checkpoint /path/to/checkpoint \
--output web/public/models/jax-ppo.onnx
The policy tries WebGPU first and falls back to ONNX Runtime WebAssembly. If the model asset is absent, the UI remains playable using a simple local heuristic and reports that fallback in the header.
Checks
cd web
npm test
npm run build
The TypeScript engine follows src/lost_cities_jax/engine.py and its 96-action
atomic action space. Cross-runtime fixtures can be regenerated with:
uv run python scripts/generate_web_parity_fixture.py