Assignment Tracker — Methodology and Known Limitations
The interactive version renders via JavaScript. This page holds the same real methodology and the same honest limitation as static text, readable without executing anything.
How grading actually works
A problem is generated by code, a numeric answer is computed by code, and the student's answer is compared to it directly. No AI model evaluates or judges any answer at any point.
Prerequisite gating
A node can't be marked mastered if the nodes it depends on aren't already mastered, matching the real dependency structure in the tier map.
The known, disclosed calibration limitation — stated directly
A real forward-walk test found that the difficulty-selection fallback used before enough history accumulates is a fixed function of the target probability alone — it never samples a second difficulty tier once one is in use, so it can't discover a better fit for a student far from average ability. Convergence toward the stated 75/70/65/60/55% frontier targets is measurably real for average-ability students. It is not yet real for students at ability extremes, and shouldn't be treated as calibrated for them. Full detail: Backtest-Only Audit.
What this means in practice
The tool is real and safe to use as a deterministic problem generator and mastery tracker for any student. The specific claim of calibrated difficulty selection near the top or bottom of the ability range is the one part not yet substantiated, and this page exists partly so that limitation is visible even to a reader or system that can't execute the interactive version's JavaScript to see the on-page warning.