# First Slip

> Find where your math working first went wrong. Type the problem and your working, one step per line - solving an equation, or simplifying, expanding or working out a value - and the page checks every line against the one before it, free, on a squared exercise-book page. One metered run marks it like a teacher's red pen: a mark on every line, a verdict, what went wrong and why with a nudge instead of the answer, the corrected first slipped line, and one practice problem like it.

URL: https://first-slip.skillsafe.ai/
API tutorial: https://first-slip.skillsafe.ai/api.html

## Free, in the browser
- Reading the working: brackets, powers (x^2 or a superscript square), square roots, fractions, mixed numbers (1 1/2), percent, implied multiplication (2x, 3(x - 1)) and one letter as the unknown. Lines with words are notes; a line may start with "=" to carry on from the line before. A .txt file can be dropped in.
- The line-by-line check, solve mode: two equations match when they have the same solutions, found numerically between -10000 and 10000, so a line that loses, gains or changes a solution is caught. The problem's own equation counts as the line before line 1.
- The line-by-line check, simplify mode: two expressions match when they agree at ten values of the letter; plain sums are compared by value.
- After a wrong line, the next line is compared with the wrong one, so correct work that carries a mistake forward reads as following on. It is a check of equality, not of method.
- The page: each line with a pencil note in the margin (= follows, a not-equal sign for does not follow, ? not checked), the page's first slip, whether an answer is reached and whether it fits the problem.
- Practice: load the practice problem from a reply, work it, and the page solves it itself and tells you whether your answer is right - no run.
- Exports: Markdown, a .csv of the lines (page check, mark, comment), JSON, print; kept pages (in the browser, or in your account when signed in).

## Metered (gpt-terra), one task: mark
The model gets the page's check of every line and does the judging the page cannot. Marking rules, in order: a word line is a note; a line the page found does not follow is a slip; a line the page found follows is "follows on" after an earlier slip, else right; a line the page could not read is judged by the model. The verdict rule, checked in order: rework (two or more slips), one_slip (exactly one), unfinished (no slip and no answer reached), full_marks (everything else). Each slip carries what went wrong, why, and a nudge that must not give the fix away; the corrected first slipped line sits behind a button. The page re-checks every reply: the verdict against its rule, a mark on every line, the marks against its own check, the corrected line (solved or evaluated again), the practice answer (solved again), and no number, he/she or internal id the input does not hold.

## Sources
Built on the SkillSafe skill @jmsktm/education-tutor - https://skillsafe.ai/skill/@jmsktm/education-tutor - (personalized tutoring, study strategies and educational support across subjects and learning styles), narrowed to one sitting: marking your own math working line by line.
The trend note on the page refers to a Show HN on Hacker News in October 2026 for an AI that marks handwritten math the way a teacher does.