Confidence Insights
How the confidence scoring system has been behaving across your students over the last 30 days
Access Denied
Teachers only.
Every time one of your students answers a question, the system updates their confidence score for each grammar tag attached to that question. This page rolls up that activity so you can see at a glance that the system is healthy, how much your students are practising, and which individual answers caused unusually large score swings (worth a quick look to spot lucky guesses, confusing questions, or genuine breakthroughs).
Notable recent updates
Answers where the actual outcome had ≤ 15% predicted probability — strong enough to be worth a look. A correct answer the model gave little chance to may be a real breakthrough or a lucky guess; a wrong answer the model expected to be right may point at a confused student or a problem with the question itself (wrong key, ambiguous wording). Each row shows the most striking answer in the session — click the row to see every answer for context.
| When | Student | Grammar tag | What happened | Source |
|---|
Every time a student answers a question, the system runs a Bayesian-Elo update that nudges their confidence for the tagged grammar point. The numbers below are rolled up daily so you can spot anomalies — e.g. a day with unusually large updates, or a jump in profile-save failures. Anomaly thresholds:
per-answer surprise > 0.85, |difficulty| > 3, evidence > 1000.
If Anomalies stays near zero and Avg surprise drifts below 0.5,
the system is learning as intended.
Daily rollups
| Date | Source | Updates | Avg surprise | Max δ | Max α+β | Anomalies | Save fail | Diff fail |
|---|
Top outliers (largest per-answer surprise)
Filtered by maxAnswerSurprise (∈ [0, 1], source-independent — the model gave the actual outcome at most 1 − maxSurprise probability). maxΔ is the same answer's per-answer delta (= K × surprise, varies by source). Σδ is the summed delta across every answer for that tag in this event. n* = how many per-answer surprises in the session crossed the threshold.
| When | Student | Tag | maxSurprise | maxΔ | Σδ | n* | Direction | Source | α / β |
|---|