Your cup on exam morning
It fills to what you'd still hold on the day, if you did nothing between now and then. The dashed line is your target; the arc beneath it is how much of your window is left. Click it to pour it again.
What to do tonight
Every subject, weakest first
Hover one to see the exact figure and the range around it.
Start with your actual syllabus
Paste what your department handed you. Cutoff splits it into subjects and into the individual facts you'd have to recall — because "eight subjects" and "eight subjects, four hundred separate things to recall" are very different amounts of work, and only one of them can be scheduled.
How well do you know each one?
Tap a number. 1 if it's completely gone, 4 if you could teach it. That's the whole setup — everything else is worked out from your syllabus, these taps, and today's date.
1 gone · 2 shaky · 3 solid · 4 could teach it. Tracking 600 separate facts — every one forecast on its own, then added up.
Everything slipping away, subject by subject
Each line is one subject, starting where you are now and falling towards exam day. Click a name in the legend to isolate it.
Which subjects will hurt most
Bars are ranges, not points — the width is the model's measured error at this horizon. The short ones are where an hour is worth the most.
The last day you can still start
Pick any day and imagine you start then and work flat out every night until the exam. How much would you know on the day? That's the line. Drag it.
Why a deadline exists at all
Revision done close to the exam is worth more — that's why cramming works, and we're not going to pretend otherwise. But you can't sit through six hundred facts the night before. You get through maybe 40. Once the nights are the scarce thing, there's a last day you can start and still finish. Almost nobody knows theirs.
The least work that still gets you there
Cutoff picks sessions one at a time, always the one that adds most per minute spent, and stops the moment you hit your target. Each bar is one evening — hover it.
Tonight
Where the effort lands
Sessions per subject across the whole plan.
You don't have one exam. You have two.
Mid-sems in September, end-sems in November, and only so many free evenings. Every dot is a real revision plan using the same nights. There is no plan that wins both.
Pick your trade
Space starts and pauses it. It keeps running in a background tab — the title bar counts down.
Tonight's brew
Straight off your plan: the first session it wants from you.
Cups today
One cup per finished session. Kept in this browser only — nothing is sent anywhere.
How any of this actually works
No maths required. Three ideas, and everything else follows from them.
Forgetting isn't random. It has a shape.
Learn something today and you'll be shaky in a week, worse in a month. That fade follows a curve that's been measured since the 1880s. Because it has a shape, it can be predicted — that's the whole reason Cutoff can exist.
Every fact you know carries three numbers.
This is the model in full. There is nothing else in it.
Some facts stick after one look. Others you re-learn five times. Tracked per fact, and it barely changes.
Days until you're down to a coin flip on it. A successful review pushes it up; forgetting knocks it back.
Given how stuck it is and how long since you looked — could you produce it this second? This is what falls daily.
Revising late is worth more — until you run out of nights.
For one fixed exam, later revision is worth more, simply because there's less time left for it to fade. But you can't revise six hundred facts the night before. The nights are the scarce thing — and once you count them, there's a last day you can start.
Watch it work on one fact
Move the sliders. Every line here is drawn by the same fitted model that draws your forecast — there is no separate demo path, and no language model is anywhere near it.
The model, in sixteen numbers
This is the whole thing. Sixteen floats, fitted by us on 40 real collections,
committed to the repo as artifacts/dsr.json. Nobody typed a rule in, and there
is nothing else in the box.
We tested it on half a million real answers
We took the study records of 40 real people, showed Cutoff only the early part of each history, and asked it to predict what they'd get right later — answers it had never seen.
When it says "8 out of 10", do 8 in 10 happen?
This is the test that matters, because you plan around the number. If it's honest, the roast line sits on the green one.
And how far off, how far ahead?
This table is where the ranges on every other screen come from.