Mastermind turns evidence you create inside the app into an explicit study model. Here is what that model does, where its boundaries are, and how you can inspect it.
It keeps durable evidence — and respects resets
Supported study modes write attempt-level evidence to your account's study history instead of keeping only a single average. That history is durable, but it is not beyond your control: resetting progress starts a new reset epoch, so earlier evidence no longer drives the active model, and account deletion follows the deletion policy. From active-epoch evidence, the brain maintains concept estimates and schedules around modeled forgetting. Those are estimates from work done inside Mastermind, not permanent verdicts on what you know.
It adapts within each study type — and labels the boundary
Adaptive Session first chooses a study type. Within each supported type, Mastermind then selects an item using that type's current policy: review timing, model estimates, prerequisites, or a focus request. It does not yet score every question, vocabulary card, kanji card, listening item, reading item, and speaking item in one shared contest. On scorer-served adaptive question cards, tap "Why this now?" to see a serving reason. A "Why this?" label elsewhere explains the answer, not the selection.
It shows how a prediction compared
Before a mock exam starts, Mastermind locks in its expectations. After you finish, it shows the receipt: "we expected about N correct; you got M," section by section, including where the model was overconfident. That compares one pre-exam prediction with one mock result; it does not prove that the readiness band predicts the real JLPT. We measure these misses so the model can be evaluated instead of asking you to take it on faith.
What it deliberately does not do
Missed days do not trigger a study penalty or loss-shaming copy. Daily-completion and in-session streaks can still appear, but after time away Mastermind recalculates the recommended work from current evidence. And honest limits: this beta is built first for serious N5 learners with some prior Japanese exposure, not absolute beginners or a complete course. Higher levels are roadmap work. Adaptation uses evidence produced inside Mastermind, and uncertainty can widen as well as narrow when new answers arrive.
Your data is yours
Speaking recognition runs on your device, and microphone audio does not leave it. Recognized text and coarse attempt telemetry may be sent to Mastermind. Your learning history is downloadable from Settings as a single file, and account deletion follows the deletion policy, including its payment-record limits. No third-party analytics, no ad trackers.
Questions this page didn't answer? Read the FAQ