POST-TRAINING is a static single-page web game served from Cloudflare Pages. There is no backend database, no user account system, no telemetry endpoint, and no analytics script. We do not log your gameplay, your custom prompts, your IP address, or your device fingerprint.
The language model (albert-base-v2, ~22 MB fp16 ONNX) is hosted directly from this site and runs entirely inside your browser via @huggingface/transformers v4 (WebAssembly). Every masked-token forward pass, logit calculation, and procedural puzzle generation happens locally on your own CPU. Even when you type a custom sentence into the Live Lobotomy Lab, those words never leave your device.
POST-TRAINING sets no cookies. It uses your browser's localStorage and sessionStorage strictly on your device to remember:
post-training-save: your surgical case progress, solved lobes, streak, and preferred operating mode (Live LLM Surgery vs. Pre-Dissected Ward, audio mute, and blood/gore intensity).post-training-loading & post-training-crash-count: session-only sentinels that detect if a low-memory mobile browser tab was terminated while loading the ONNX model, automatically switching to Pre-Dissected Ward mode so your browser never gets stuck in a reload loop.In addition, your browser uses standard HTTP caching and IndexedDB so the ~22 MB ONNX model weights only need to download once. You can wipe all stored data and caches at any time using the WIPE MEMORY / RESET button in the Surgical Manual dialog or via your browser settings.
The complete source code is available on GitHub under the MIT License.
← Back to Operating Table