We publish how often we are right.
Onbridger predicts the grade you will get on your exam. After the exam, we record what you actually got. The delta is the only honest measure of whether the system works. This page is the live record.
Predictions made
0
Snapshots taken before exams across every active course.
Outcomes recorded
0
Students who shared their actual grade after the exam.
Within ±2 points
-
Unlocked at 25 recorded outcomes. 25 to go.
How the prediction works
Step 1
We measure your real performance, not your confidence.
Every question you answer updates per-topic mastery. We weight by how recently you saw the topic and how stable your recall has been across re-tests. There is no self-report, no “rate your understanding,” nothing you can game.
Step 2
The prediction is anchored to the weakest topic.
Exam grades track the floor harder than the average. A high mean across nine topics will not save you from one cold topic in the question paper. The prediction is a deterministic function of your weakest topic, lifted toward the average only when the spread is tight.
Step 3
We refuse to predict when we should not.
If any topic is still untouched, no prediction. The system shows you which topic is blocking it instead. We would rather show nothing than show a number we cannot defend.
Step 4
The snapshot is taken before the exam, not after.
Every prediction is timestamped at the moment it was made. After your exam, we record your actual grade. The delta is the calibration. Nothing is recomputed in hindsight, because hindsight is how systems lie about themselves.
Step 5
There is no AI in the prediction.
We use AI to grade open-ended answers and to surface weak topics. The grade prediction itself is a deterministic formula over your mastery state. Run it on the same inputs twice and you get the same number twice. This is a property, not an accident.
Step 6
The forecast is empirical at three layers.
Your responses, your cohort’s responses, and the cross-cohort calibration of every student studying the same course. As the cohort grows, the prediction tightens because item difficulty refits to real responses, not synthetic estimates. We will publish per-cohort calibration deltas alongside the global figure.
Our commitment
We commit to publishing the calibration of every prediction Onbridger makes, including the ones where we were wrong. The counters above are live. The percentage updates as outcomes arrive.
When the system is off, we will say so. When the model improves, we will version it on this page so you can see the lift. When a cohort throws calibration off, we will name the cohort.
No general-purpose AI tool will tie itself to a falsifiable outcome metric, because they have nothing to anchor it to. The Onbridger metric is anchored to your own course, your own mastery state, and your own exam.
See your prediction.
Upload your material. Set your exam date. We will tell you where you stand.
Start for free