How ThinkTrace works
The innovation is not that it contains a tutor, a quiz generator and a polling tool. It is that classroom-level confusion and individual-level diagnosis are the same loop, and the loop does not close until the misconception is demonstrably gone.
- 1LecturePulse
How the class is confused
Every written answer is grouped by the kind of reasoning behind it, not by which option was ticked.
- 2ConceptLens
Why this student is confused
The specific belief that produced the answer, with the counterexample that breaks it.
- 3Prerequisite repair
What is actually missing
The shortest path back through the concept graph — five minutes, not a whole chapter.
- 4ErrorTwin
Does the error repeat?
Practice built around the reasoning pattern, in contexts the student has never seen.
- 5Teach-back
Has it actually gone?
The loop closes only when the student can explain it — not when a video finishes.
1 · LecturePulse — how the class is confused
The teacher asks one question and, crucially, asks for the reasoning as well as the answer. The analysis groups students by the kind of reasoning behind their answer — two students who picked different options for the same underlying reason land in the same group. The teacher gets the dominant misconception, the prerequisite gaps ranked by how many students need them, a two-minute intervention, a counterexample and a follow-up question that would separate the groups.
2 · ConceptLens — why this student is confused
A student opens their own answer, or types what they think without knowing what the confusion is. The diagnosis names the specific belief, shows why it fails against concrete numbers, gives a counterexample and states its own confidence. The student can reject the diagnosis — rejecting it stops the cycle rather than practising against the wrong premise.
3 · Prerequisite Detective — what's actually missing
Concepts form a graph. Rather than sending the student back through a chapter, the system walks backwards from the misconception to the two or three concepts underneath it and orders them deepest-first. Each node is a one-minute explanation, a worked example and a single check question.
4 · ErrorTwin — does the error repeat?
Practice is built around the reasoning pattern, not the original question: a near-identical case, the same trap moved to a new domain, then a transfer question somewhere unrelated. Every item carries a distractor built from the student's own misconception, so a relapse is distinguishable from a random miss.
5 · Explain My Way and PerspectiveLab
The same concept in simple language, technical form, step-by-step, low-text visual, or as a sustained analogy — and in another language, optionally keeping technical terms in English. PerspectiveLab adds the voices of people whose work depends on the idea: a physician, a security analyst, an examiner, a data scientist, and a plain everyday analogy.
6 · Teach-back — has it actually gone?
The cycle closes only here. The student explains the concept in their own words, and the evaluation asks whether the original misconception is still driving the explanation, whether the phrasing is merely recalled, and whether the idea transfers beyond the first example. Mastery moves red → yellow → green → blue on that evidence, and the teacher's dashboard updates.
What this build is running on
Supabaseanthropic modelThinkTrace runs with or without external services. With Supabase configured, accounts, classrooms and responses persist and stream over Realtime; with a model key configured, every piece of analysis is generated live and validated against a strict schema before it reaches the screen.
Without them, the application is still complete: an in-memory store backs the same interface, and a deterministic analyzer reads each student’s actual words against a misconception catalogue for the sample lesson. Every screen states which of the two produced what you are reading.