Telemetry & Performance Metrics

The dashboard rejects vanity metrics. Raw accuracy is visible but not the primary performance signal. Every interaction you have with the platform is instrumented at the event level, and these events feed a set of composite metrics designed to reflect actual learning progress rather than comfortable repetition.

Every question interaction generates up to five event types: option_selected (which option you chose and when), hint_used (which tier and at what elapsed time), confidence_rating (your 1–5 star self-assessment, if provided), report_error (if you flagged the question for review), and time_spent (total and per-step where available). These events are immutable once logged.

Learning Velocity is the primary metric: a composite score measuring how efficiently you absorb and retain new concepts over a 7-day rolling window. It is calculated from the slope of your accuracy curve on questions encountered for the first time at each difficulty level. A student whose accuracy on Level 4 questions improves from 45% to 72% over 7 days has high velocity. A student whose accuracy on Level 2 questions holds steady at 88% over 7 days has near-zero velocity.

High raw accuracy paired with low learning velocity triggers a system alert. This is the 'comfort zone' warning: you are successfully practicing mastered material but generating no new learning. The engine will force-route you toward higher-difficulty or lower-familiarity topics to restore productive friction.

Weighted Accuracy applies a recency weight to your accuracy history — recent sessions count more than older ones. This prevents a strong early performance from masking a recent decline. The weight decay function has a half-life of approximately 14 days.

Speed Score measures time efficiency relative to question difficulty. A Level 3 question solved in 45 seconds scores higher than the same question solved in 3 minutes. Speed Score is not rewarded at the expense of accuracy — it is only a positive signal when accuracy is above the 70% threshold.

Trend Detection is a classifier that evaluates your Learning Velocity trajectory and labels it as Improving, Stable, or Declining. A Declining trend triggers a session alert and prompts the engine to restructure the next Daily Module around your weakest recent topics. An Improving trend is acknowledged in the Reflection phase. A Stable trend — particularly at high accuracy — triggers the comfort-zone alert described above.