Our method: train what actually improves your chess

ChessPivot is built on a simple conviction: you improve more by working intelligently on your own games than by playing more of them, and more by searching for answers than by reading them. This page explains where every design choice comes from — what research establishes, what great coaches recommend, what we deliberately rejected, and the limits we own.

What research establishes

Structured study beats game volume. The reference study in the field (Charness et al., 2005) measured, across two independent samples, that serious study is the strongest predictor of rating — far ahead of tournament play. In their regression, one (logarithmic) unit of cumulative study is 'worth' about 200 ELO points versus 33 for tournament games. The authors explicitly recommend devoting the majority of chess time to study. That is ChessPivot’s foundation: analyse, understand and re-train your games rather than grinding blitz.

Guided work matters. In Gobet & Campitelli’s study (2007), group practice and having a coach were the significant predictors of national rating. Qualified feedback on your own moves — precisely what ChessPivot’s board-verified coaching does, move by move — is not a luxury: it is one of the measured levers.

Practice is necessary, not sufficient — and nobody should promise you ELO. Macnamara et al.'s meta-analysis (2014) establishes that deliberate practice explains about 26% of performance variance in games — the most favourable domain of all, yet far from everything. Gobet & Campitelli observed an 8-fold spread between players in hours needed to reach master level, and players past 25,000 hours without reaching it. The practical conclusion: an honest tool measures your process and your skills, never a rating promise.

Invested effort is never wasted. The longitudinal study of young Dutch players (de Bruin et al., 2008) shows the benefit of practice is steady and durable — including for those who later quit competition. Every session counts, at your own pace.

Optimal difficulty exists, but it is a heuristic. Wilson et al. (2019) proved mathematically that learning is optimal around 85% accuracy — for learning algorithms on binary classification tasks. Applying it to chess is a reasonable extrapolation, not a law: we use it as a compass (training where you succeed at everything teaches nothing; training where you fail at everything discourages), without making it say more than it does.

The principles that follow

Your games as raw material

Your real weaknesses are detected in your games by engine-verified rules, and your personalised exercises are built from your own positions. Working on a mistake you actually made transfers better than an abstract puzzle — it is also the most unanimous pillar of the coaching profession, from Botvinnik to Dan Heisman: analyse your games, especially your losses. And when we top them up with a bank of verified puzzles, their difficulty is calibrated to your ELO: you train just above your comfort zone, never on positions out of reach or trivial.

Search before you read

On the analysis page, the 'Guess first' mode has you propose your move before revealing the explanation — on the moments where it makes sense (a tactical blunder with a clear-cut answer is guessed freely; a positional error, where several moves are reasonable, is trained as comparative judgement). Active recall anchors better than reading: one of the most robust results in learning psychology, and coaching practice since forever ('find the move before turning the page').

A miss gets replayed, then comes back

When you fail an exercise, the solution is shown and then you REPLAY it yourself on the board before moving on — a missed pattern you replay gets encoded, one you merely watch evaporates. And the failed position returns at the head of your next pass, until a clean solve. Skipping your failures is online puzzling’s number-one anti-pattern.

Spaced repetition for the memorisable — and only that

Your opening lines are reviewed with spaced review (the FSRS algorithm): declarative knowledge, exactly what the technique is for. We do NOT put your tactics on spaced review: a calculation skill grows through variety of positions, not through memorising a set.

Endgames first

Capablanca and Lasker taught endgames before anything else; modern curricula (the Dutch Steps method, Russian programmes) likewise delay opening theory. Our 'Endgames to know' (canonical positions with engine-proven verdicts), pawn endgames and Endgame Rush follow that pedagogical consensus.

Process before outcome

Your plan adjusts to the time you want to spend — not the other way round — and your cycle goals are calibrated to be reachable, weighted first on effort (which you control) and then on results (which you only partly control). That is the direct consequence of the data: practice pays, but nobody can guarantee the slope.

Our exercises, one by one

Every ChessPivot exercise obeys the same reliability rule: a position is only offered if its solution is verified — by the engine, by a strict geometric Mimesis™ detector, or both. No ambiguous labels, no invented solutions.

Mate patterns

Train

Recognising named mates (back rank, ladder, supported queen…) is the most profitable tactical skill below 2000: it finishes won games. Three angles: the ones you delivered, the ones you missed in YOUR games, the ones to know.

Forced mates verified by the engine; patterns classified by a strict motif detector.

Forks

Train

The most frequent tactical motif at every level, broken down by executing piece (knight, queen, pawn…) — because spotting a knight fork and a pawn fork are two different reflexes.

Material gain validated by simulated exchange (SEE); your missed chances re-validated against the engine best move.

Pins, skewers, discovered attacks, batteries

Train

The four long-range-piece tactics, each with its own page, its schemas (by piece, against the king or not) and its three angles (delivered / missed in your games / to know). The battery — doubling rooks, lining up queen and bishop — is the least taught and among the most profitable in attack.

Geometry validated by a conservative detector (no clear target → no label); banks drawn from the Lichess corpus re-filtered by our detectors; your misses re-validated against the engine best move.

Checks, captures, threats (CCT)

Train

Systematically scanning forcing moves is THE anti-blunder discipline taught by every coach: before moving, list checks, captures, threats. The exercise drills that scan into a reflex.

The set of forcing moves is recomputed from the position itself — the answer cannot be wrong.

Loose pieces (LPDO)

Train

'Loose Pieces Drop Off': scanning undefended pieces — yours and your opponent’s — is the base vision that prevents most one-move blunders.

Undefended pieces are recomputed from the position — exact by construction.

Defensive vigilance

Train

Classic training’s blind spot: puzzles always have you attacking. Here you identify the opponent’s threat THEN parry it — on positions where you actually conceded a tactic.

Any defence that holds the evaluation is accepted (engine-graded) — not just “the” single move.

Good move or bad move?

Train

Positional judgement (structures, trades, plans) almost never has a single answer — guessing 'the' move there would be a lottery. The right form: compare the move you played with the best one, with the measured gap. Covers positional errors, pawn structure, castling, development, bad trades…

The gap between the two moves is the stored engine evaluation from YOUR games (≥ 1.5 pawns) — nothing invented.

Endgames to know & pawn endgames

Train

Technical endgame knowledge (opposition, Lucena, Philidor, pawn races, king activation) decides countless half-points. You CONVERT a winning position or HOLD a draw against best play — no multiple choice.

Canonical positions with proven verdicts; in Endgames to know, each move is judged against the endgame tablebases first (exact truth up to 7 pieces), with the engine as fallback.

Stalemates to avoid

Train

Stalemating in a winning position is amateur chess’s most infuriating accident — and it trains very well: convert without falling into the trap, trap revealed afterwards.

Stalemate traps detected by exact rule (the tempting move REALLY stalemates) on verified positions.

Convert under pressure

Train

Reaching a won game and then losing it on the clock is one of the amateur’s most expensive leaks. Here you replay your own winning endings that you lost on time, against a realistic opponent — the Maia human-like engine, calibrated to your level and tempo — until you close them out cleanly.

Winning starting position established by the engine; the Maia opponent replays humanly plausible moves, never an artificial perfect defence.

Remove the defender

Train

Destroying or deflecting the defending piece: the intermediate-level tactical mechanism par excellence — the one that turns 'I see the target' into 'I see how to take it'.

Targets and defenders recomputed from the position; gains re-validated by the engine.

The timed Rushes

Train

Offensive, defensive, endgame: 60 seconds on YOUR motifs. The clock is not there to 'learn fast' — it serves recognition volume and motivation (a record to beat). It is dessert, not dinner: the real work happens unclocked.

Same verified pools as the untimed exercises; the 100-calibrated score is comparable across runs.

The openings trainer

Train

Learn (annotated lines), Practise (replay against the board), Review (spaced review, FSRS algorithm) — plus a 'positions to fix' deck built from YOUR deviations from repertoire and recurring leaks.

Our repertoires were audited with their plans re-proven by the engine; your leaks are measured on your games.

What we don’t offer — and why

A method is also defined by its refusals. Ours are argued:

The '+400 ELO' promise (de la Maza’s method)

'Rapid Chess Improvement' promises spectacular gains through intensive tactic repetition. The result was never replicated, the method is widely criticised (including by coaches who value tactics), and its author stopped playing after his peak. We keep the sound idea (tactics first below 2000) and leave the promise.

Systematic Woodpeckering

Looping the same puzzle set at decreasing intervals runs against what we know about spaced review (increasing intervals), and reports from players who followed it are mixed. Our choice is the reverse: a solved position is never re-shown before the pool is exhausted — only your MISSES come back, first in line, until solved.

Spaced repetition applied to tactics

Spaced repetition excels at the memorisable (opening lines, named patterns); it is the wrong tool for a calculation skill, which grows through variety. So we use it where it works (openings), not as marketing everywhere.

Puzzle-spam as a method

Chaining hundreds of puzzles while skipping your failures produces a rising score and a flat level. With us, failing triggers the solution TO REPLAY, then the position’s return — the opposite of skipping.

Deep opening memorisation at amateur level

Below 2000, almost no game is decided by theory at move 18 — but many are lost on principles (development, castling, the centre). Our repertoires are short, explained and tied to your games; encyclopaedic depth can wait.

Passive video content

Watching is pleasant and ineffective on its own: no study supports passive viewing as a driver of improvement. Every piece of teaching content in ChessPivot leads to an action: a position to play, an exercise, a game to analyse.

In numbers — what you can check

The statements below describe the service as it runs today. Each one can be observed in the app; the figures are updated whenever the method changes.

Engine analysis: Stockfish 18, depth 15 on every move of every imported game, depth 25 on demand; Every game imported from Chess.com, Lichess or a PGN file is analysed automatically, move by move, at depth 15; decisive moves are analysed deeper. On any move, the player can run a depth-25 analysis, with no limit. No cap on the number of analysed games with the Plus plan.
Verified opening lines: Every taught move compared with the engine best move, one-pawn threshold; Every move taught in our opening pages is compared with the engine best move in the same position: beyond a one-pawn gap, the line is repaired or cut before the faulty move — never extended with an invented move. The opponent move shown is the one players actually play at your level, not the ideal theoretical move, and a line stops where theory stops.
Weaknesses detected: 32 detectors, 41 error patterns grouped into 7 areas; Tactics, defence, king safety, opening, strategy, endgame and clock management. Thirty detectors work move by move, two at whole-game level. Each pattern is named, counted across all your games, then ranked by frequency and by engine-measured cost.
Detector verification: 254 positions judged by hand, 21 replayed on every release; The reference corpus was judged by hand: 254 positions reviewed one by one, 139 kept as confirmed flaws, 115 set aside for lack of sufficient certainty or because the game or the category no longer exists — we would rather miss a flaw than invent one. On every release, 21 of these positions are replayed automatically and the release is rejected if any of them changes verdict: zero false positives required.
Tracking over time: Every weakness tracked at 1, 3, 6 and 12 months, with its trend; Number of occurrences, share of games affected and direction (improving, stable, worsening), weakness by weakness.
Exercise bank: Over 110,000 engine-verified exercises; Every exercise has a Stockfish-proven solution and a difficulty rating; the difficulty served adapts to target roughly an 80% success rate per exercise family. Players can vote each exercise "good" or "doubtful", and that signal is reviewed.
Spaced repetition: FSRS algorithm, for openings only; The lines of your repertoire come back on an FSRS schedule; tactics and endgames are not memorised but replayed until solved, as explained in the previous section.
Opponent preparation: Repertoire by colour, exploitable weaknesses, preparation lines you can add to your repertoire; A named opponent’s file (Chess.com or Lichess) uses the same analysis as yours; the suggested lines are added to your personal lines in one click, and a sparring partner imitates their style.
Over-the-board games: Scoresheet photographed, converted to PGN and analysed; Handwriting recognition reads your scoresheet; you correct any doubtful moves before the analysis runs.
Comparison with your rating band: Every weakness compared with players in your 400-point rating band; On your Statistics page, each weakness shows the share of games affected among players in your band (ChessPivot Observatory: 248 players, five bands from 400 to 2399) and says whether you are above, in line with, or below them.
Your data: Games and repertoire exportable as PGN, in one click; From My games and from the opening repertoire, with no request and no delay. Other personal data is available on request (right to portability).
Accessibility: AA contrast checked, keyboard navigation, animations can be switched off; Every control has a visible focus and dialogs close with Escape; changing areas are announced to screen readers (live regions); animations only play when the system allows them; touch targets are at least 44 px. An automated WCAG 2.1 AA audit (axe) runs on the main pages. Known limit: no full screen-reader audit yet.

Our limits, owned

We do not promise ratings: deliberate practice explains only part of the variance between players (≈ 26% in games per the reference meta-analysis), and the slope of progress varies hugely from person to person. What we do guarantee: diagnoses measured on your games, exercises with verified solutions, and a sustainable plan you can actually follow.

Volume remains king: 45 well-spent minutes a week beat zero, but the fast-improving players in the studies study a lot. Our plan minimises friction so the time you HAVE goes to the right place; it will not create time you don’t have.

Finally, overall clock management is trained mainly by playing slower time controls: no exercise replaces the experience of managing your clock in a real game. Its most expensive case — losing a winning position on time — does, however, have a dedicated trainer ('Convert under pressure'), where you replay your own winning endings until you close them out.

A word on your data: ChessPivot only analyses public Chess.com and Lichess games, accessed via your username. No password is ever requested, and no data is ever sold. Evaluations come from Stockfish 18, running in your browser and on our servers, at a depth adapted to your rating.

Sources

The scientific claims on this page link to the studies below. The rest is coaching consensus (Heisman, Aagaard, the Capablanca/Lasker tradition, the Steps method), presented as such.