How Educate grades honestly.
Educate asks families to rely on its grades and its alarms. You shouldn't have to take that on faith — so here is the machinery, plainly.
The student can't reach the answer key
Correct answers and scoring notes live in a separate server-side store that the student-facing app has no way to read — enforced by database permissions, not politeness. There is nothing to peek at, on any screen, in any request.
The AI reports evidence. Code decides.
A misconception streak advances, closes, or reopens only through fixed transitions written in code. The model can grade an answer; it cannot be sweet-talked into declaring something mastered.
Alarms fire from rules — never from vibes
A slump threshold crossed, a missed day, a target grade that stops being mathematically feasible: every alarm traces to a rule and the numbers that tripped it.
Think a grade is wrong? Audit it.
A dispute triggers a full audit with a per-item receipt — a partial audit is a failed audit. Regrades apply the most lenient defensible reading, so the benefit of the doubt goes to the student.
Every report says which rubric graded it
Grading rubrics are versioned, and each report is stamped with the exact version that graded it — so a dispute is judged under the rubric that actually did the grading, not whatever it says today.
Classmates share materials — never state
Your reports, plans, and error log are invisible to classmates and teachers alike; only the shared class layer is shared. The fences are enforced in the database and adversarially tested.
Guardian access exists only by consent
A guardian joins through an invited link, accepted while signed in. From then on, every guardian action on the account is attributed to the person who took it.
The history is never hostage
Everything your student has earned stays readable and exportable — markdown, HTML, CSV — free, forever, subscription or not.
Honest grades aren't a promise. They're a design constraint.