Grade Coach — a calibrated grade prediction and a weekly plan, per course

If you ask a student "what grade are you tracking for in Chemistry?" the answer is almost always a guess. They have a rough feeling — they did well on the last quiz, they bombed the mock exam, the teacher said something vague — but no actual number. The hardest part of revising in Y11 and Y13 isn't doing the work; it's deciding what to do next, and that decision is impossible without an honest read on where you're currently at.

We're shipping Grade Coach today. It's a per-course grade prediction that's calibrated against your actual quiz and exam history on LearningBro, plus an optional AI-generated weekly study plan focused on the single highest-leverage topic for you to work on. Live now in your dashboard at /grade-coach.

What you get

Open Grade Coach and you'll see one card per course you've worked on, sorted weakest first — because that's where the advice matters most. Each card has three things:

A predicted grade. A real GCSE 9–1 grade or A-Level letter grade (depending on the course), based on a recency-weighted blend of your quiz scores and any mock exams you've taken. The number reflects what you'd score today, not what you scored six months ago when you started the course.

A confidence band. Low, medium, or high. A prediction off three quiz attempts is honestly less reliable than one off twenty, and we say so out loud. If the confidence is low, the headline message is: do more practice before you trust the number.

Your three weakest lessons. Computed straight from your quiz attempts on each lesson. These are the topics where the coaching question — "if you had four hours this week, where should you spend it?" — has the clearest answer.

For subscribers, there's also a Generate study plan button. Click it and a Haiku model writes a personalised three-paragraph plan: one short sentence acknowledging where you are, one paragraph on the single highest-leverage topic to focus on this week (and why), and a concrete list of about four hours of study — specific lesson titles to revisit, quizzes to retake, flashcards to run through. No padding, no generic motivation, no "Dear Student" preamble.

The plan is cached for 24 hours so you can't burn tokens by refreshing constantly, and so when you come back tomorrow morning the advice is the same advice you read last night.

How the prediction is calculated

Honest about the algorithm — it's intentionally simple for v1:

• We collect every AssessmentAttempt (the quiz at the end of each lesson) and every ExamAttempt (full timed mock exams) you've completed on the course.
• Each attempt contributes its accuracy (score over total). Attempts completed in the last 28 days count double — the model cares more about where you are now than where you were when you started.
• Quiz accuracy and exam accuracy are blended 60/40 into a 0–100 performance score, then mapped to a grade band: GCSE 9–1 (≥90 = Grade 9, ≥80 = Grade 8, etc.) or A-Level A*–E (≥80 = A*, ≥70 = A, etc.).
• Confidence is high if you've done 20+ attempts including some recent ones, medium if 8+, low otherwise.

That's it. There's no forgetting-curve model, no fancy item-response theory, no per-question difficulty calibration. The honest answer for v1 is: more data, weighted toward recent data, mapped to bands that match how UK exam boards actually scale raw marks to grades. Anything more complex would feel more authoritative without actually being more accurate.

Why "weakest first" matters

The card order is the most important UX decision in this feature. If we sorted by course title alphabetically, the dashboard becomes a list. Sorted by predicted grade ascending, it becomes a tool: the first thing you see is the course where focused effort this week will move your overall A-Level / GCSE outcome the most.

The same logic applies inside each card — the AI plan focuses on the single highest-leverage weak topic, not the long list. Doing a little bit of everything is the most common revision mistake. Doing a lot of one thing that's actually holding you back is what moves the grade band.

What it's not (yet)

Some things we deliberately left out for v1:

• No forgetting-curve modelling. We treat all attempts within the recency window equally. A real predictive model would weight by how long ago you saw the material — and our flashcard data could fuel that. Coming in v2 if usage justifies it.
• No multi-course summary. "You're tracking for Grade 6 in Maths, A in History, B in Chemistry" — that's a real GCSE/A-Level dashboard, and we know students and parents want it. Today's release is per-course; the summary view is next.
• No teacher view. Teachers can already see class-wide weak topics in the weekly AI class reports we ship every Monday. A teacher-facing version of Grade Coach (per-pupil predictions for the whole class) is a natural extension but isn't in this release.
• No "what if you spent N hours" simulator. The spec calls for "spend 4 hours on quadratics and that moves you from Grade 6 to 7 with 70% confidence". That requires modelling the marginal effect of practice on each topic — interesting, hard, and a v2 feature.

These are all on the roadmap. v1 ships the calibrated prediction + the weekly plan, which is the half of the spec that pays for itself immediately.

How to try it

In your dashboard sidebar there's a new Grade Coach link, under Progress. Open it, and any course you've taken at least three quizzes on will show a prediction.

If you've just started a course and don't see a prediction yet, that's working as intended — the model needs a minimum of three attempts before it'll say anything (because confidence below that is genuinely too low to be useful).

The grade prediction itself is free. The AI study plan is part of the subscriber AI suite alongside the AI Tutor chat, Examiner Mode, Synoptic Tutor, and voice mode. Seven-day free trial unlocks all of them.

Where this fits

Grade Coach completes the picture we've been building over the last month:

• AI Tutor chat answers the question you ask.
• Examiner Mode marks the answer you wrote.
• Synoptic Tutor flags the prerequisite from another subject you might have forgotten.
• Grade Coach tells you what to work on next.

Each one answers a different revision question. None of them write the work for you. All of them politely decline if you ask them to.

The next thing UK students need from an AI tutor isn't a smarter chatbot — it's direction. That's what Grade Coach is for.