Chess Homework From Games: Mistakes to Drills

Because the student's own games are the only honest record of what they actually get wrong, and because a child works harder on a position they recognise. Generic puzzle sets train general pattern recognition; homework built from the student's own mistakes trains the specific habit that is currently costing them rating points.

The motivational effect is not a small thing. "Solve these twenty puzzles" is an errand. "This is the position from your game on Saturday, find what you missed" is personal, and students feel the difference immediately. The position carries the memory of the result, and the correction lands with an emotional anchor that a random puzzle never has. Coaches who run review meetings know this from the other direction: the moments students bring up unprompted are always from their own games, never from exercises.

There is also a diagnostic payoff. When homework is generated from games, the homework history becomes a record of the student's recurring weaknesses, and you can watch a mistake category fade over weeks as the drills do their work. This post covers the pipeline from mistake to drill; how to read engine output on student games in the first place is the subject of our pillar on engine analysis for chess coaching, and the 15-minute review method that produces the homework note is covered in its own guide.

Most student mistakes fall into a small taxonomy of recurring types, and each type has a drill format that targets it directly. Matching the format to the mistake is the whole craft of homework design; the wrong pairing, such as assigning tactics puzzles for a planning problem, produces busywork that changes nothing.

The taxonomy earns its keep because classification is fast. Once a coach has named the category during the game review, the drill format follows mechanically, and themed puzzle sets for the tactical categories are free on the Lichess training pages [1]. The judgement call is the dose, which should err small; the table's quantities are a ceiling for a school-age student with a fifteen-minute daily slot, not a floor.

You extract the mistake positions from the student's games and turn them into a private puzzle set, and the entire pipeline runs on free tools. The build takes a coach about ten minutes per student once the routine is familiar, and the result is a set of exercises no app can sell you, because the positions came from this child's play.

1. 1Export the game as PGN from whichever platform the student plays on [4]; if the academy's students are on Lichess, games can be fetched in bulk through the free API [5].
2. 2Run the engine pass with Stockfish and mark the moments where the evaluation swung [3]; these are the same marks the game review already produced, so nothing is analysed twice.
3. 3Create a Lichess study and add one chapter per marked position, set up one move before the mistake [2].
4. 4Hide the moves so the student sees a problem, not an answer: the question is always "find the better move", with the game continuation entered as the line to avoid.
5. 5Add one sentence of instruction per chapter, naming the concept: "White to move. Check every piece nobody is defending."
6. 6Share the study link with the student, and top the set up with themed puzzles from Lichess training when their own games have not yet produced enough material [1].

The study format matters more than it looks. A position the student must solve cold, without seeing what happened in the game, recreates the decision they originally faced, which is exactly the rep you want. Walking through the game continuation first turns the exercise into a memory test, and memory of the specific game is not the skill being trained. Keep each study small, five to eight chapters, so finishing it is achievable inside the week; a forty-chapter monument to the student's errors is demoralising and will not be opened twice. If a game produced more marked positions than the study can hold, keep the spares in a private draft chapter and feed them into later weeks.

Over a term, these studies accumulate into something genuinely valuable: a personal mistake library per student. When the same category resurfaces months later, the old chapters are right there for revision, and a student paging back through positions they now solve instantly is looking at hard evidence of their own progress. That evidence matters for morale in a way no rating graph quite matches, because the student remembers being unable to see these answers.

Three items maximum, fifteen minutes a day, all three tied to the mistake category from the last review. That is the entire format. Fifteen daily minutes beats a Sunday marathon because pattern recognition compounds through frequency of exposure, and because a daily slot is small enough to survive school, exams, and everything else competing for a child's evening.

• Item one: the personal study built from their own positions, five to eight chapters across the week.
• Item two: a themed set from the free training pool matching the same category, around ten to fifteen puzzles [1].
• Item three: one slow online game where the only goal is applying the named habit, with the game submitted for the next review.

Notice that item three feeds the next review, which is what makes this a loop rather than a list. The game played with the scan rule becomes the game that gets pre-scanned next week, and the review either confirms the category is fading or picks the next moment from the same family. Homework, review, and the coaching note form one cycle, and each part exists to feed the next.

Check it in the first two minutes of the next game review, using last week's one-line note as the checklist. The note already says what was assigned; the study shows which chapters were opened, the puzzle themes show a count, and the submitted game shows whether the habit appeared under real conditions. Two minutes is genuinely enough.

The reason to be disciplined about this is blunt: homework that is never checked stops happening, usually within a month. Students read what coaches actually attend to, not what coaches say matters, and a drill set that vanishes into silence tells them the assignment was decorative. The check does not need to be an interrogation; "show me your favourite chapter from the study" opens the same conversation as a checklist and feels like interest rather than audit.

Parents are the quiet ally here. A one-line mention of the week's homework in whatever recap the academy sends home recruits the household to the routine without turning parents into enforcers. The aim is for the daily fifteen minutes to become as unremarkable as instrument practice, and households are good at routines once they know one exists.

Shrink the assignment, tie it even more tightly to their own games, and never punish. A student skipping homework is giving you information about the assignment, not about their character, and the correction is almost always to make the homework smaller and more personal, not to add consequences.

Diagnose before redesigning. The common causes are mundane: the set was too hard, so five minutes in it felt like failure; there was no quiet device time at home on weekdays; or the student never understood what the drill was for. Each has a cheap fix. Drop the difficulty one band. Move the slot to the weekend and halve it. Spend thirty seconds at assignment time connecting the drill to the exact moment in their game it came from, because "this is so Saturday's game does not happen again" is a purpose, and "do these puzzles" is not.

For the persistent non-doer, cut to a single item: one chapter, from their own game, two minutes. Almost every student will do one two-minute exercise about their own position, and a completed tiny assignment rebuilds the habit that a string of ignored big ones destroyed. Punishment, by contrast, attaches bad feelings to training and accelerates the slide; the connection between joyless drilling and students leaving altogether is real enough that we wrote about it separately in our post on why students quit chess academies, and a stalled training loop is also a common backdrop to the rating plateau.