How the review is calculated

Engine search

For each move, the engine searches its preferred move and the played move from the position before the move. Candidates are compared within a completed search iteration. Important, unstable and tactical positions receive deeper verification, with additional samples throughout the game. Advanced passed pawns can receive up to four times the normal verification budget. Centipawn precision and outcome confidence are assessed separately. Unresolved comparisons always receive verification and are marked provisional if they remain uncertain.

• Quick: 25,000 nodes per first search
• Standard: 100,000 nodes
• Deep: 250,000 nodes

Grades and accuracy

A move grade describes the drop in expected points from the player’s point of view. Best, Excellent and Good preserve most of the position’s value; Inaccuracy, Mistake and Blunder lose progressively more. Ordinary grades use rating-aware expected-points loss; technical centipawn loss is displayed separately without overriding the grade. Special labels such as Brilliant, Great and Miss require extra evidence and are awarded conservatively.

Accuracy is a separate 0–100 score based on each move’s rating-neutral expected-points loss. Move accuracy is 100 × exp(−loss / 7), rounded to one decimal place. Game and phase scores are the ordinary average of their eligible move scores: each decision counts equally, so one error cannot dominate a long game. Smaller errors also receive a meaningful penalty. Book and forced moves are excluded. If any scored decision remains unresolved, the player’s accuracy is Pending and affected phase scores are unavailable; provisional verdicts are counted separately. The score measures decisions, without a bonus for winning or holding an advantage. This is ChessVI’s independent approximation, not a reproduction of private CAPS2.

Exact endgames

Online endgame tablebases are an optional setting, off by default. When enabled, board positions with seven pieces or fewer are sent to Lichess Syzygy for proven win, draw or loss outcomes, including the fifty-move rule. No player names or PGNs are sent. Unavailable or incomplete evidence falls back to local analysis. Move details identify exact evidence; enable the setting before reanalyzing.

Limits and licensing

Analysis method 2.2.0. Older reports retain their scores and offer reanalysis. Accuracy comparisons in Progress use only the current method.

Searches are bounded, so engine versions, speed and depth can change close decisions. These grades are ChessVI’s own method and are not intended to reproduce a different site’s private scoring model. An unseen practice move is checked from the same position before it is scored; unavailable evaluation remains inconclusive.

The bundled Stockfish engine is licensed under GPLv3. Read its license.