Quality
Our accuracy standard
Every question a student can be shown must first pass automated validation: a machine checks its structure, its mathematical notation, the form of its accepted answers, the integrity of its distractors, and that it is not a duplicate. An item that fails validation is never delivered.
Automated validation catches structural and notational faults. It cannot, by itself, guarantee every fact and final answer is correct — that is what human subject review and reader-reported errata are for. We record which gates each lesson and item passed, when they were verified, and whether a human reviewer has signed off (its provenance), and we show a “last reviewed” date on each lesson driven by that record.
Machine-verified items
23,326
pass automated validation before delivery (as of 2026-09-03)
Error rate on audited sample
Not yet measured
no items have completed human/errata review yet (n = 0)
Median time, report → fix
No fixes yet
no reported errors have been fixed yet
Automated bank-quality audit (duplication and difficulty-calibration checks — not factual errors): 15,535 blocking and 9,262 advisory findings across the bank; release gate not yet passed. These are quality signals we are working down, not measures of factual correctness.