A right answer isn’t always a known answer

Practice tests grade right or wrong. A lucky guess counts the same as real knowledge, and a confident mistake hides in a decent score until test day. StudySignal asks one more thing after every answer, how sure were you?, and grades both.

One tap, before you see the result

The whole product is one interaction. Pick an answer, then tap Know it, Maybe or Don’t know. That tap locks it in. No submit button, no peeking at the result first, so the confidence is honest. Most of the design work went into making that second tap feel like part of answering, not a survey stapled on after.

Six grades instead of one percentage

Answer times confidence gives six outcomes. Two of them are why the product exists: a Misconception, sure and wrong, the most valuable thing on the page because the student didn’t know they had it, and Lucky, right but unsure, which won’t hold up on test day.

A 7 out of 10, scored by StudySignal

Review only what needs it

Sure and correct doesn’t need another look, so review skips it. It walks through every lucky guess and confident mistake: your answer, the right one, a worked explanation, why the wrong choices tempt, then a fresh question on the same topic to check it stuck.

Where the AI lives

StudySignal is an AI product with no chat window. The model does two jobs: it writes questions for the bank, and a tutor approves every one before a student sees it, and it decides what goes into the next test by reading the student’s radar. Students see one button. No prompt box, nothing streaming in. What the AI makes is a test, which is what they came for, and the rainbow pill only appears on buttons that make one.

Build from what you missed

One sheet replaced four paywalls scattered around the app, and it tells you what it’s going to build before it builds it: eight weak spots from three tests, shown as a bar of difficulty tiers. Then three choices. On-screen test or printable worksheet with an answer key. Drawn from the last test, the last 30 days or all time, each with its count. And how long. Every control is free to play with; the lock, when there is one, sits on the one button that makes something. Guests can set it all up and are told once that signing in comes first, and their tests come with them.

The mix is arithmetic you can read

Any topic under 80% mastery goes in. Its share grows with the square of how far below it sits, so a topic at 30% gets far more than one at 70%, and a floor keeps every weak topic in the draw. A little jitter means no two builds match. Mastery means right and green, so a run of lucky guesses can’t talk a topic out of the mix.

A control that says what it decided

Difficulty is a multi-select from Easy to Insane, plus a fifth tile, Tailored, for pick for me. It aims at the student’s edge: a tier counts as mastered at 80% right-and-green over at least four questions, and Tailored takes the lowest tier they haven’t mastered plus one above it. If a control is going to decide for you, it should say what it decided, so the hint names the tiers and why. Think you’re past Medium? You can see the product disagrees, and the reason. With no results yet it starts at the bottom two and says so: the first test is the measurement.

Written by a model, approved by a tutor

Claude writes new questions to the same schema as the hand-written ones: section, topic, a difficulty on the 400 to 800 scale, the question, four choices or a typed answer, wrong choices built from real slips (a dropped sign, an off-by-one, the wrong setup) and a worked explanation. It re-solves every problem before committing the answer. Each one lands tagged with its model and batch, marked needs review. A couple in every hundred are wrong, so nothing reaches a student until Michael approves it. Here’s one.

Same score, different students

The teacher pitch in one picture. Two students, both 7 out of 10. The gradebook says they’re the same student. The signal doesn’t.

For teachers: a worksheet in a minute, and who guessed

Two things. A worksheet builder on the live bank, topics down the side and difficulty across the top, every cell showing how many questions it holds so a sheet never comes up short. And a roster where every attempt carries the student’s signal: share one link, students join when they open it, and each row shows who knew it and who guessed. Tap the weak cells in a student’s grid and send them a test built from exactly those.

The same signal, in your pocket

The iOS app runs the same questions and the same signal, with history synced to the web. Start a set on a laptop, finish it on the bus.

A brand with three rules

The wordmark is the product in miniature: the name in Geist and a stoplight’s three dots. Green, yellow and red mean three things in the interface (the confidence signal, the difficulty ramp and plain status) and never a fourth, so they read the same on every screen. A selected control changes color and fill, never border weight. Every token comes from one set of variables, which is why light mode is the same product, not a second one. The style guide is public at studysignal.app/design/styleguide and runs on the shipping stylesheet, so what’s there is real.

Under the hood

No framework. One hand-rolled stylesheet and one script, served static, with serverless functions in front of Postgres for accounts, attempts, the bank and rosters. Google or magic-link sign-in, KaTeX for math, Chart.js for charts, Desmos for the calculator, Stripe for billing. Close to 8,000 live questions across SAT, PSAT, ACT, SSAT, HSPT and ISEE, every one calibrated by a tutor, and nothing ships until the test suite passes.

Takeaways

The product is one extra tap. Everything else, the six grades, the review loop, the radar, the roster, the targeted tests, comes from asking a question nobody else asks before showing the result. The analysis wasn’t the hard part. Making that tap feel like part of answering, on a phone, on the clock, so students actually tell the truth, was. The AI stays where it’s useful, writing the bank and reading the radar, and students never type a word to it.