Reasoning diagnosis is useful only when uncertainty remains visible. This page separates published evidence, product hypotheses, and the boundaries of the current prototype.
A guarded tutor can improve performance while the tool is present without producing a positive unaided-exam effect. MySquire therefore treats delayed, unaided transfer as the intended outcome—not chat completion.
The same mathematical trace can reflect a misconception, a forgotten condition, notation trouble, or a simple lapse. A credible system must maintain alternatives, probe, and abstain.
The cost of the error is visible, the original attempt is recent, and faculty need to decide what to reteach. Paid partner discovery must still confirm that this is a high-priority workflow.
The website demonstrates one algebra scenario. It does not claim arbitrary handwriting recognition, full JEE coverage, autonomous grading, or production efficacy.
MySquire is designed to make bounded claims about supported work and to test possible causes—not convert a probabilistic signal into a permanent learner label.
These sources inform the product thesis. None independently proves that MySquire works; that requires its own paid pilots and controlled outcome studies.
The guarded condition improved assisted practice but did not produce a positive unaided-exam effect over control.
A contemporary meta-analysis of randomized tutoring studies; useful as a benchmark, not a product claim.
Supports focused self-explanation as one component of learning, not unrestricted conversational friction.
Reports 1,538,468 unique candidates appearing across the two 2026 Paper 1 sessions.