Every game in this series so far had one rule: no "right" and no "wrong". The mistake showed itself, through leftovers in the depot, a folded screen, or a story read back in a strange order. I think the rule was a good one for those games. But it's only fair to challenge it, and to do so with the arguments of the person who would challenge it hardest.

The grandfather of every learning app

In November 1953 the Harvard psychologist B. F. Skinner visited his daughter's fourth-grade maths class. The teacher was good, but she had thirty children. A child would solve a problem and find out whether it was right only days later, when the exercise book came back. Some children waited for the others; others fell behind. Skinner went home and built a machine.

His "teaching machine" showed one small question at a time. The student wrote an answer, pulled a lever, and immediately saw the correct one. The steps were so small that mistakes were rare, and everyone moved at their own pace. In his 1958 article in Science, Skinner laid out the principles that today sit underneath almost every learning app on your phone.

His argument against our rule is simple. If a child doesn't find out straight away whether they were right, they may be practising the wrong thing. Silence isn't neutral.

What the research says

Here the answer is more interesting than "Skinner is right" or "Skinner is wrong".

Feedback is one of the most powerful things in learning. But in 1996 Avraham Kluger and Angelo DeNisi gathered hundreds of studies and found something uncomfortable: in over a third of cases, feedback made performance worse. That happened most often when it shifted attention from the task to the person. "That's wrong" is about the task. "You're not good at this" is about the person.

The 2007 review by John Hattie and Helen Timperley reaches a similar conclusion. The most useful feedback is information about the task and how to do it. The least useful is praise and rewards. And Edward Deci and colleagues showed in 1999 that expected tangible rewards often weaken intrinsic motivation. Points, stars and daily streaks are exactly that kind of reward.

The distinction this research draws is this. Skinner was right that the child needs to know immediately. We were right that the child shouldn't get a reward or a judgement about themselves. The two can be separated.

The engineering side: the fast test

Every programmer knows the difference between a test that fails in two seconds and a bug a customer finds a month later. The faster the feedback, the faster you learn. That's why we run tests on every change.

But a good test says "expected 5, got 4 on line 12". It doesn't say "you're a bad programmer", and it doesn't hand out badges for a streak of green builds. It reports on the code, not the person. That's what feedback in the game should be like too.

The game: "Small Steps"

The learning goal is the first sound of a word and the letter for it: B as in banana, M as in moon. It's a task with a clear right answer, and that's exactly the kind of task Skinner's approach works for.

1. Show. The game shows the letter B, a picture of a banana, and says "banana" aloud. Nothing is asked.
2. Two choices. A picture of a banana and the word spoken aloud: "Which letter does it start with?" Two large letters that look very different, for example B and O.
3. Straight away. If the child picks B, the word sounds again and the letter glows gently. If they pick O, the game calmly shows B next to the banana and says the word once more. No error sound, no "well done", no "wrong". The same task comes back after two others.
4. Little by little. A new letter arrives only when the earlier ones are going easily. If the child misses more than one in five, the step is too big and the game goes back to two choices.

Letters that look alike, such as b and d, come last. Skinner wanted his programmes to be so smooth that the learner almost never made a mistake. The reason is that every mistake is practice of the wrong thing.

The prompt

Build a small game as a single self-contained HTML file, based on
B. F. Skinner's programmed instruction: small steps, active response,
immediate feedback about the task, self-pacing. Language: English.

Goal: the first sound of a word and the letter for it.
A list of 10 letters, each with one word and an emoji (for example
B: banana, M: moon, S: sun, D: dog, C: cat, F: fish, T: tree,
P: pig, R: rabbit, L: lion). Speak the words with speechSynthesis.
Don't speak the letters themselves, because voices say the letter
names, not the sounds.

Steps:
1. Show: the letter, the emoji and the word spoken aloud.
   A "Next" button.
2. Choose from 2, later from 3 large letters: "Which letter does
   [word] start with?"
3. Feedback straight after the choice:
   - for the matching letter, say the word again and let the letter
     glow gently for a second;
   - for another letter, show the matching one next to the picture
     and say the word. The same task returns after 2 others.
   No words like "well done", "right" or "wrong", no error sounds.
4. Add a new letter only when at least 4 of the last 5 answers
   matched. If more than 1 of the last 5 didn't, go back to
   choosing from 2.
   Letters that look alike (b and d, p and q, m and n) come last.

A session is 10 tasks, then "See you tomorrow". No points, stars,
badges, streaks or timer. Nothing is stored after closing.

Design: calm colours, large letters and buttons, works on a phone,
no ads and no external requests.

One technical note. Phone voices pronounce a letter's name, so "bee" instead of the sound "b". That's why the game speaks the whole word, and the sound is heard at its start. If you want the individual sounds, it's best to say them yourself.

Where Skinner doesn't help

Programmed instruction works well where there's a clear right answer: letters, times tables, words in a foreign language. It doesn't work for the story in the Reggio Emilia article, Froebel's symmetric patterns, or Dewey's question "what would we do differently". There's no single right answer there for a machine to show. That's why the earlier games in this series had no "right" and "wrong", and it wasn't a mistake that they didn't.

And one more thing. Knowing the letters isn't the same as reading. Ten minutes of "Small Steps" doesn't replace a bedtime story read together, with a finger moving under the lines.

Skinner isn't one of the six in the book from the AI Learning Game Kit, but the kit does the same job for the part after the prompt: from a working game to an icon on the phone.

A child needs to know straight away whether they got it. They don't need to hear that they're clever or win a star. Good feedback talks about the task, not the child.