Reps and grading
A rep is one problem, answered or skipped. Every rep is scored the same way, and the score is never a matter of opinion: your number against the true one.
Error, with a sign
Section titled “Error, with a sign”Your error is
error = (answer − true) ÷ |true| × 100%It keeps its sign. +25% means you were a quarter too high; −25% means a
quarter too low. Two engineers with the same typical error can be making
opposite mistakes, and only the sign tells them apart. The Stats page reports
your median signed error so you can see your own lean.
The bands
Section titled “The bands”| Band | Your answer is… | What it means |
|---|---|---|
| Excellent | within 10% either way |
A number you could act on. |
| Acceptable | within 40% either way |
Close enough to catch a real mistake. |
| Miss | further out than 40% | The instinct is not there yet. |
| Way off | a tenth of the answer or less, or ten times it or more | Something structural was missed. |
The edges belong to the better band: exactly 10% out is excellent, exactly 40% is acceptable.
“Way off” is measured in ratios, not percentages, because “ten times out” is the same mistake in either direction: a tenth of the answer is 90% low, while ten times it is 900% high.
Time counts — a little
Section titled “Time counts — a little”An excellent answer that took 30 seconds or more is graded as acceptable for scheduling purposes. Your band on the reveal does not change — you were still within 10% — but the type comes back on the same spacing until you are fluent as well as accurate.
Time never turns an accurate answer into a miss, and the very first rep of a new problem type is exempt: that one is spent learning to read a new kind of problem.
Skipping is allowed and costs nothing in your figures — a skip is not counted as an answer. But the reveal after a skip still shows you the answer and the method, so you have seen the type: its clock moves, and a skip spends a cold read without making one. See the schedule.
Why the same problem never repeats
Section titled “Why the same problem never repeats”Each problem type has many generated versions with different dimensions and loads. The schedule tracks the type, never the individual problem, so you are never rewarded for remembering a specific answer — only for the judgement that produces it.