Chess puzzles from your own games, and how they differ from a puzzle set
A puzzle set is a supply of positions somebody else selected. Your own games are a small, slow supply of positions selected by your mistakes. They are different instruments: one has volume and no information about you, the other has information about you and almost no volume. Knowing which is which decides how you use both.
Where a puzzle from your own game comes from
Chess puzzles from your own games are not generated by an idea generator; they are salvage. A game gets an engine pass, a move where the loss was real becomes a counted event, and that event already carries everything a drill needs: the position as it stood, the side that was to move, the move you played and the move the position was asking for. Turning it into a drill is bookkeeping rather than invention.
Two rules run at that point and both are about honesty. The stored position must be one where you were to move, so an opponent's mistake never comes back to you as your drill. And the answer stored is the move the position was asking for, never the move you actually played - with the answer held server-side and stripped out of what your browser receives, so the page you are solving on does not contain the solution.
What this supply can and cannot cover
It covers exactly the mistakes you make, in the positions you reach, against the opponents you face. No puzzle set can offer that. What it cannot do is fill a training session. A week of games produces a handful of usable positions at best, and a bad week produces several from the same recurring situation, which is one lesson repeated rather than several.
It is also biased in a way worth knowing about. Only mistakes that cost something measurable become events, so the quiet strategic drift that lost you a game slowly is under-represented, and the sharp material mistakes are over-represented. That is a fair trade - the measurable half is the half you can grade - but it means your own-game drills are not a full picture of your chess.
A puzzle set is a different instrument
A rated set gives you volume, a difficulty knob and a clean signal about calculation. It also tells you, by existing, that something is there to be found, which is the one condition a real game never gives you. Neither fact is a flaw. It is simply a measurement of a narrower skill than most players read it as.
So the sensible use is unglamorous: the set for repetition and sharpness, your own positions for information. The one thing not to do is treat a rising puzzle rating as evidence that the mistakes in your games are going away. Those are two measurements, and only one of them is taken in the conditions you actually play under.
How the two are mixed
Mikhail serves personalised training from three labelled sources: the rated puzzle bank, generated positions built for a specific skill, and your own games. Own-game positions are capped at one in every ten positions served, and the cap is enforced by the server over a rolling window of your last serves rather than by intention - so nine in ten of your drills come from the bank and the generator, by design.
Early on it is fewer still, and it should be. Before you have reviewed games there is nothing to salvage, and a signed-out visitor gets no own-game drills at all because there is no account for the positions to belong to. The cap is a ceiling on a scarce resource, not a quota that has to be filled.
Doing it yourself, anywhere
The method needs no particular software. After each review, save the two or three positions where you went wrong into a study or a file, with the side to move and nothing else attached - no theme, no title, no note about what kind of mistake it was. A week later, solve them cold. The absence of a label is the whole exercise, and it is the part people helpfully add and thereby remove.
Grade them strictly. A position you looked the answer up on is not solved, and counting it as solved is how a training record turns into a record of your optimism instead of your play. That is the same reason Mikhail raises mastery only on a clean, unassisted solve, and records a revealed answer without ever rewarding it.
Questions
How many puzzles will my own games actually produce?
It depends entirely on how much you play and how much of it you review, so no honest number can be quoted here. The useful way to think about it is as a slow trickle you should not spend all at once, which is exactly why the serving cap exists.
Why not serve me only my own mistakes?
Because there are not enough of them, and because re-meeting the same position often turns into a memory test rather than a solving test. Mikhail serves a position you have not drilled before, and fills the rest of the session from the rated bank matched to your level.
Do these drills show me the mistake I made?
No. You get the position as it stood and nothing else, because being told what kind of mistake is coming removes the part you failed at. The move you played is in your game review if you want to see it afterwards.
What if I disagree with the drill's answer?
Play the position out against the engine from the position you preferred and look at what changes. The stored answer is the move the engine found best at review time, not a claim that your move was unplayable - and a drill you can argue with usefully has already done more than one you clicked through.
Try it on your games
The part of Mikhail this guide describes is Training, which needs a free account — your games and your counts have to live somewhere.