The standard homework set is thirty problems, and it's the same thirty for everyone in the room. That made sense when one teacher had to serve thirty students with one photocopier. It's worth being honest about what it costs.

For any given student, most of those problems are things they can already do. Working them is not nothing — repetition has some value — but it's the least valuable hour available. A few of the thirty target the exact thing that's broken. The student has no way to know which few, so they grind through in order, run out of time or patience around problem twenty-two, and never reach the ones that mattered.

Practice you don't need is the most expensive kind, because it doesn't feel like waste. It feels like diligence. You did the work; you just did the wrong work, and there was no signal to tell you.

The same wrong answer, two different problems

Two students hand in the same integral, both with the same wrong result.

The first chose the right method, set it up correctly, executed the substitution, and then dropped a negative sign in the last line of algebra. The second didn't recognize that substitution applied at all, tried to integrate term by term, and produced something that happens to be equally wrong.

Identical score. Opposite remedies. The first student needs nothing about integration — they need to slow down on arithmetic, and thirty more integrals will not help. The second needs to work on recognizing when a substitution is available, which is a conceptual gap that more careless-error drilling will never touch. Give them both the same remediation packet and you have helped neither.

This is the case against grading by outcome. The answer tells you that something went wrong. Only the work tells you what.

Two students can miss the same question for opposite reasons. Only one of them needs more practice on that topic.

Errors are data, if you keep them

Once feedback lands on specific steps rather than whole problems, something useful accumulates. A single dropped sign is noise. The same dropped sign in four problems across two weeks is a diagnosis. A student who justifies every step in algebra but silently skips justification in proofs has a pattern that no single assignment would reveal.

Those patterns are what practice should be built from. Not the chapter you're on, not the topic the syllabus says is next — the specific operations that have failed for you, recently, with the frequency to prove they weren't flukes.

It also changes when practice should stop. The usual rule is "when the problem set is done," which is a statement about the assignment, not about you. A better rule: an error is retired when you can execute that step cold, without prompting, on a problem you haven't seen — and then again a week later, after you've had time to forget. Spacing the recheck is the part everyone skips, and it's the part that tells you whether you learned it or just recently saw it.

What this requires

All of it depends on being able to read the work. Not the answer — the steps, in the order they happened, including the crossed-out branch where the student tried something and abandoned it. That's the trace of reasoning, and it is where every one of these distinctions lives.

Which is the unglamorous reason Lune Synth™ is built around scanning in handwritten work rather than collecting typed answers. Not because paper is virtuous. Because a page of work carries the information needed to tell a slip from a misconception, and a typed final answer throws all of it away. Then missions get generated from what actually broke, and they don't retire until you've shown the fix holds.

Want practice built from your own errors instead of a generic set? Join the beta waitlist on the Lune Synth home page, or reach us at griffin@lunesynth.com.