Privacy Policy
How account, save, and leaderboard data is handled.
Data we process
Account email, password-derived authentication records, profile identity, an optional profile image, sessions, cloud game states, recovery snapshots, automatically maintained leaderboard scores, and security audit events.
Optional AI learning comparisons
The Tokenizer Workbench runs its exact local tokenizers in your browser. Only if you request provider totals, the current workbench text is sent through Cloudflare AI Gateway to Anthropic and Google to obtain provider-reported input usage. If you request the Context Window Laboratory's optional live illustration, only fixed synthetic records and your bounded laboratory controls are sent to Cloudflare Workers AI; no typed learner text, account data, or game state is submitted. The Context Evidence Use Laboratory likewise submits only four categorical controls for a fixed synthetic record set; its response is not stored and cannot affect progress. If you request the Runway Laboratory's optional Assumption Finder, only allowlisted strategy and scenario IDs plus the visible bounded numeric plan are submitted. If you request the Data Quality Laboratory's optional Data Card Auditor or the Model Specialization Laboratory's optional Regression Diff, only the selected allowlisted fictional card or output-pair ID is submitted. If you request the Customer Lifecycle Laboratory's optional Synthetic Customer Panel, only an allowlisted fictional case, segment, and action ID is submitted. If you request the Prompt Experiment Laboratory's optional live comparison, only one allowlisted technique ID is submitted; five fixed fictional cases, both prompt arms, call order, model, settings, and rubric remain server-owned. No learner prose is accepted by these bounded instruments, no account or game data is submitted, and deterministic local code remains the sole calculator and grader. AI Lab CEO does not store these optional requests or responses. The bounded open-model instruments call Workers AI directly and do not create an AI Gateway record; where AI Gateway is used for provider totals, payload logging is disabled per request. Provider terms still apply, and optional model output cannot affect learning progress.
Security reports
Private vulnerability reports contain the details and optional contact email supplied by the reporter, a request ID, and a keyed IP hash used for abuse prevention. Query strings and raw IP addresses are not retained with these reports.
Account-required play
The game opens only for a signed-in account with a verified email address. The active game state loads from the account cloud save and updates there automatically while you play.
Public information
Public score publication is currently disabled, regardless of the saved visibility preference. Personal bests remain private to the account. Email addresses and profile images are never published by leaderboard APIs.
Control and deletion
The account area provides full export and password-confirmed hard deletion. Exports include the optional profile image; deletion removes it together with account-owned saves, snapshots, and leaderboard scores.