Our method

A better record
of what broke.

SolveTrace is designed around a simple distinction: understanding an explanation and independently finding the idea are different achievements.

Correctness is the beginning of the review.

An incorrect answer doesn't say whether a student lacked a concept, missed a cue or made an arithmetic slip. A useful postmortem looks for the first meaningful failure, rather than labelling every downstream symptom.

Keep the student's first account of the miss.

Seeing a solution can change the way a student remembers their approach. SolveTrace asks for self-diagnosis before reference-solution exposure, then keeps the initial and confirmed categories distinct.

AI proposes. The student reviews.

The diagnosis is a fallible interpretation of the evidence provided. The student confirms or corrects it. Retry timing and learning-state updates follow explicit rules; model confidence is not a mastery score.

Test independently, then test elsewhere.

A cold retry checks the original later with earlier work hidden. Unseen transfer probes the same idea in a changed setting. Passing one level supports evidence at that level; failing does not advance it.

The failure taxonomy

Eight ways reasoning can break.

These categories organize review. They describe an attempt, not a student's ability or identity.

Knowledge: Did not know a required concept, theorem, technique, or fact.
Recognition: Knew the relevant idea but did not recognize that it applied here.
Representation: Represented the problem poorly: wrong sample space, poor variables, unhelpful formulation, or failure to reframe.
Strategy: Understood the mathematics but chose an ineffective or incorrect high-level approach.
Execution: Approach was appropriate but could not be carried through correctly.
Precision: Arithmetic, algebra, or case-bookkeeping slips despite an essentially correct approach.
Assumption: Made an unjustified assumption (independence, equiprobability, ordering, replacement, conditioning, hidden constraints).
Time management: The dominant issue was time allocation: dead path held too long, abandoned too early, or an approach too slow for contest conditions.

What this beta can claim

SolveTrace records attempts, organizes reasoning reviews and distinguishes cold retries from unseen transfer. It is not a validated predictor of contest scores, and its rules are not claimed to be scientifically optimal. No improvement guarantee is implied.

Transfer coverage depends on the problem library. The current seed set is small and concentrated in counting and probability. It is not a complete contest curriculum.

Read the beta FAQ →

Start with one problem

Make your next miss useful.

Solve independently. Review your reasoning. Know what to test next.

Create your beta account