Speech-therapy worksheet generator

A speech therapist wrote from her phone, with a screenshot: the button offered a vowel the sheet never printed. On a soft target it showed “А” while the sheet printed “РЯ”; on the hissing sounds it showed “Ы” while the sheet printed “ЖИ”. The button and the printer were working out the syllable in two different places in the code. They now call the same function. This is the first correction that came from someone who is not us.

The row that picks the material used to name genres: sheet, track, maze, stories. A therapist read that row and asked why there was only one option per sound — the steps were there, but under our own words. The row is now named the way she named it herself: sound, syllable, word, phrase, text. The genre stands under the step in small type.

The tool opened as a wizard: pick a sound, pick a material, then see a sheet. Twenty-nine other worksheet generators were opened by hand to check whether the field works that way. Two of them are wizards, and both insert the step in order to charge for it. The screen came off. The sheet now arrives assembled and the settings stand beside it.

A word gets onto a sheet because it contains the sound being practised. But the child never reads that word. The child looks at the drawing and says out loud whatever they see. That spoken word is the exercise. So the drawing has to force one exact word. Three of mine did not. A drawn figure 10 is called “десять”, not the “десятка” it was picked for. A newspaper is called “газета”, not “газетка”. A cinema hall is called “кино” or “кинотеатр”, not “кинозал”. The words a child actually says have no target sound in them, so those three sheets were practising nothing. I dropped the three words and kept the drawings. The drawings were fine; the mistake was in choosing the word.

Both were listed as working features. Neither could be built for almost any sheet: the rule they stood on had been switched off two days earlier, and the table replacing it covered twenty-three words out of seventeen hundred. They were counted as existing without existing.

A practice track for one voiced consonant printed syllables that come out as its voiceless twin when read aloud. The sheet was training the opposite sound. Two more materials had the same class of fault. All three came off the paper the day it was measured. What a letter looks like and what a mouth does with it are different facts, and only the second is allowed to decide anything.

Navigation rested on the idea that one session equals one step of difficulty. Ten real lesson plans were coded into a dataset of thirty-three task names, and not one worked that way. A therapist works from how far this child has come with this sound, and builds a run-up to it. The idea was dropped; the build, tested against the same dataset, held.

A speech therapist described the format she needed. What went into the code was my summary of her words, not the words. A week later the two were compared: the format in the code was not hers. Her words now live untouched in their own file, and any reading of them sits beside them, never on top.