Methodology
Chess Owl reports are built from engine analysis and deterministic rules. Every metric below is generated from the published MetricDefinition catalogue and the standard-v2 analysis profile — not from free-form copy.
Analysis profile
- Profile
- standard-v2
- Engine
- Stockfish Stockfish 17
- Nodes per position
- 400,000
- MultiPV (critical)
- 2
- Classification
- wpl-v1
How the score is graded
Score: how to read it
A win is worth 1 point, a draw ½ and a loss 0. Your score is the share of the available points you took.
50% means you took as many points as your opponents. Online games pair you with players of your level, so most results land close to 50%.
Example: 6 wins, 2 draws and 2 losses make 7 points out of 10, a score of 70%.
Grading scale
- A65% or more
- B55 to 65%
- C45 to 55%
- D35 to 45%
- Eunder 35%
Metric definitions
Every rate shows its numerator and denominator. Findings based on fewer games than a metric’s minimum sample are labelled insufficient and never headline a section.
| Metric | Formula | Denominator | Min n | Source | Limitations |
|---|---|---|---|---|---|
| Engine-estimated accuracy Average per-move accuracy derived from expected-score loss against the engine’s best move. | mean over player moves of 103.1668 × e^(−0.04354 × win%-loss) − 3.1669, clamped to 0–100 | Analysed player moves | 30 | Engine | An engine-relative estimate, not a measure of playing strength. Comparable only under the same analysis profile. |
| Average centipawn loss Average evaluation lost per move relative to the engine’s best move, capped at 1000 per move. | sum(min(1000, max(0, eval_best − eval_played))) / player moves | Analysed player moves | 30 | Engine | Inflated in already decided positions; comparable only under the same analysis profile. |
| Versus your baseline Difference between this subset’s score and your overall score, in percentage points (pp): 55% against an overall 50% is +5 pp, 42% is −8 pp. | subset score − overall score (percentage points) | Games in subset | 8 | Statistic | Small subsets produce large swings; check the confidence badge. |
| Leaves book at move Average full-move number of the first move after the last position found in the opening dataset. | mean(ceil(book_departure_ply / 2)) | Games with a reliable book match | 5 | Rule-based | Only games from the standard start position; dataset covers named lines, not full theory. |
| Category score Your score (share of the points taken) in games that reached at least one qualifying position of the category. The “vs baseline” column compares it with your overall score in percentage points (pp): 55% against an overall 50% is +5 pp. | (wins + 0.5 × draws) / games reaching the category | Games with a qualifying position | 8 | Rule-based | Correlation only: reaching a category does not cause the result. |
| Move quality Move classes by expected-score (win %) loss: best ≤ 2, good ≤ 5, inaccuracy ≤ 10, mistake ≤ 20, blunder > 20; missed win when a winning position (≥ 70%) drops to ≤ 50%. | count per class / analysed player moves | Analysed player moves | 30 | Engine | Thresholds are conventions; classification depends on engine depth. |
| Repertoire share Share of analysed games that reached this opening. | games in opening / games in scope | Games in scope | 1 | Statistic | Depends on which games were imported and filtered. |
| Score Share of the available points you took: a win is 1 point, a draw ½, a loss 0. 50% is par. Shown with a letter grade from A (65% or more) to E (under 35%). | (wins + 0.5 × draws) / games | Games in scope | 5 | Statistic | Does not account for opponent strength. |
| Tactic success rate Share of detected tactical opportunities the player converted. | found / encountered | Detected opportunities of this motif | 5 | Rule-based | Only motifs our detectors recognise are counted; engine-preferred moves without a motif are excluded. |
| Win rate Share of games won. | wins / games | Games in scope | 5 | Statistic | Treats draws the same as losses; read together with score. |
Explanations
Short explanations may be written by a language model, but only from facts the analysis has already computed. Each explanation is labelled with its source, and explanations that mention moves or numbers not in the analysis are discarded and replaced by rule-based text.
Chess Owl, the report assistant
Section summaries from Chess Owl are assembled from the findings above and never call a model. Answers to your questions are AI explanations: the model only sees your report’s computed facts, and an answer that mentions a move or number not in those facts is replaced by the closest measured findings. Questions are limited per report and their text is never stored.
What the report cannot tell you
- Your intentions or plans. The engine judges moves, not ideas.
- Time-management findings if your games have no clock data.
- Anything about games outside the selection you paid for.