How your reviews are used

You are one of a small group of native speakers who decide what learners read and hear. This page explains exactly what happens to each answer you give: what it can change, what it can't, and what we keep.

What you're asked

Every card is one sentence. It asks at most four kinds of question, and only the ones that are still open for you:

Is this sentence correct, and does the English say the same thing?Yes or no. If no, you can say which half is wrong and offer a correction.

Does this sentence carry this rule?Yes or no, once for each grammar rule linked to the sentence. You can also point out a rule that's missing.

Is the audio easy to understand?A score from 1 to 5. A 4 or a 5 counts as yes.

Is this rule explained well?Yes or no, asked once per rule, the first time it comes up for you. If no, you can say how you would explain it.

You answer blind

No reviewer ever sees how anyone else answered: not before you answer, not after. You aren't shown a running tally, and nobody is shown yours. Even when your answer is the one that breaks a tie, the card doesn't tell you that it is.

This is the whole point. Two people who agree without seeing each other's answer are strong evidence. Two people who agree after seeing each other's answer are not.

How a question is settled

Two reviewers agree. It's settled.A third reviewer is never asked, because a third answer couldn't change the result.

Two reviewers disagree. A third is asked.They answer blind, like the first two. Only the question that split is put to them, not the whole card.

The majority decides.Two out of three. For audio, the middle of the three scores decides.

Most languages need two agreeing reviewers. A few ask for three before anything is settled. Nothing goes live on a split: a sentence that divided the first two reviewers waits until a third has answered.

Nobody overrides the majority. Not the person who runs this project, and not an AI model. Most of the languages being reviewed here are languages the operator doesn't speak, so the system is built to decide without them. The operator's job is the process: who reviews what, and carrying your decisions into the course. The judgement about your language is yours.

What your answer changes

Question	If the answer is yes	If the answer is no
The sentence	It can be shown to learners.	It's dropped. It stays in the history of the project and is never shown.
A rule link	Learners see the sentence as an example of that rule.	The link is removed. The sentence itself isn't affected.
The audio	Learners hear this recording.	The recording is withdrawn and redone, then judged again. The sentence itself isn't affected.
A rule's explanation	It stays as it is.	It's rewritten with the wording reviewers suggested in hand, and the new version is judged again.

Grammar rules themselves are shown to learners without being reviewed. What reviewers decide is whether each one is explained well, and which sentences really demonstrate it. A rule is only ever shown with sentences reviewers have approved.

When you suggest something

A correction, a rule that's missing from a sentence, or a translation into another language you speak natively: none of these is a vote. Each becomes new material that other reviewers judge in exactly the same way, blind, with the same two-out-of-three rule.

You are never asked to judge something you wrote.

If a sentence changes after you answered

Your answer is tied to the exact version of the text you saw. If the sentence is edited later, your answer stops counting for the new version, and the new version is asked about again. Your original answer stays in the record as what you said about the old one.

Nothing you said can ever be attached to text you didn't see. If a sentence changes while you're looking at it, your answer is refused and the card reloads.

What we keep

For every answer: your reviewer handle, the answer, the time, which version of the text it was about, and which versions of the question list and this app you were using.
Answers are never edited or deleted, by anyone, including us. The database itself refuses to.
Answers are kept under your handle rather than a number, so your work stays credited to you even if your account is closed.
Your browser keeps your sign-in and your language and theme choices. This site runs no analytics and no tracking.

The one number we look at

For every question that has been settled, we can see whether you ended up on the side that won. Across many answers, that gives an agreement rate.

It exists because the operator can't read most of these languages. It's the only way to notice that something has gone wrong: a card that misleads everyone, instructions that aren't clear, a question that's been misunderstood.

It is not a score. Other reviewers never see it. Disagreeing is not a mistake: a single "no" that turns out to be right is exactly the case a third reviewer exists for. Nothing happens automatically because of it. A low rate over a large number of answers is a reason to talk, and nothing more.