Writing
Why we count unknown words.
Language shipped first because known versus unknown is a count we can stand behind. Other subjects will need different sensors.
We started with Korean because the signal is unusually clean. A running word is known, or it is not.
That sounds small. It is the whole reason the first product is a reader and not a general tutor.
A math problem can look easy and still be the wrong shape. A cockpit drill can look short and still be the wrong hour. Those will need other sensors. Korean gives us a count we can defend.
In the feed, each text gets a score from the words you still do not know. We aim for a page with a handful of them. Three new words in a short piece is a gift. Thirty is a page you will close.
The chip on the card is that score, written as a quiet unknown rate. Readable is the default band. You can feel the difference in a minute: you are still in the story, and you met something new.
People ask why we did not start with a course. A course picks one sequence for everyone. The count lets the next page move when your known set moves.
That is also why reviews sit to the side. Due words come back on a date, as a number on home, because the job of launch is still to hand you a text.
If we cannot count, we cannot rank. If we cannot rank, we are a library with audio. Libraries are useful. They do not choose the next page.
So the first shipping work was: get an honest known set, score real texts, and put the readable ones first. Korean was the material that let us see whether that loop was true.