Abstract
Abstract Many digital mathematics tools validate only final answers, whereas teachers often need students to produce specific intermediate forms and to receive immediate guidance on where a transformation went wrong. This need is particularly important in inclusive classrooms, where low-stakes, non-judgmental, step-based interaction can support differentiated teacher intervention. This paper presents a lightweight interactive platform for step-based formative assessment in algebra covering two canonical tasks: solving single-variable equations of the form $$ax+b=c$$ and solving matrix-form linear systems $$AX=B$$ under fixed scaffolds. Across both tasks, the platform separates formula verification from deterministic numeric checking, locks progression step by step, provides explicit error labels, and records process-oriented traces for teacher-facing analytics. Unlike equivalence-oriented or CAS-dependent algebra tools, the proposed platform is designed for teacher-specified canonical intermediate forms, lightweight local deployment, and teacher-facing process analytics across two algebra tasks. The technical core is a CPU-only pipeline based on character n-gram TF–IDF, targeted numeric cues, and a compact multilayer perceptron. For the single-variable equation task, the classifier reaches 99.62% accuracy and 0.9962 macro-F1 on held-out test data. For the $$AX=B$$ task, the corresponding classifier reaches 99.87% accuracy and 0.9987 macro-F1. Beyond synthetic evaluation, the paper analyzes authentic classroom interaction logs, including timing, attempts, step-level error concentration, and cohort-aware usage patterns. These results are interpreted as exploratory classroom evidence rather than as causal proof of learning effectiveness. The contribution is a unified, locally deployable platform that combines reliable step-level diagnosis with teacher-facing analytics in an inclusion-oriented setting.