Real, deterministic assignment generator — no LLM judgment anywhere in the grading. A problem is generated, a numeric answer is computed by code, and your answer is compared to it directly. Data is stored only in your own browser (localStorage) — nothing is sent anywhere.

Known calibration limitation, stated directly: a real forward-walk test found that the difficulty-selection fallback used before enough history accumulates is a fixed function of the target alone — it never samples a second tier once one is being used, so it can't discover a better-fitting difficulty for students far from average ability. Convergence toward the stated 75/70/65/60/55% targets is measurably real for average-ability students; it is not yet real for students at ability extremes, and shouldn't be treated as calibrated for them. Full detail: Backtest-Only Audit.