How ChessTing explains and calculates chess analysis

ChessTing combines public game data, statistics from real games, and browser Stockfish, then turns that evidence into plain-English chess explanations, practical opponent preparation, and targeted training actions. ChessTing is designed to tell a player what happened, why it mattered, and what to do next — not simply show an evaluation number, graph, or engine line.

Plain-English coaching is a core ChessTing feature

ChessTing is intentionally built as an explanation layer on top of chess evidence. Raw engine output is useful to advanced players, but a number such as -2.4does not by itself teach a player what went wrong. ChessTing converts the underlying evidence into readable chess language and connects it to a practical next action.

In Game Review, ChessTing does more than label a move as a mistake or blunder. Where the position supports it, the explanation system can identify and communicate the actual chess mechanism behind the error: for example a missed mate, fork, pin, line tactic, bad capture, loose piece, opening slip, king-safety problem, or endgame mistake. The explanation can connect the move played, the stronger move, the relevant piece or square, the evaluation cost, the tactical or positional reason, and the lesson the player should take from the position.

Review therefore aims to answer questions a chess player actually has after a game:

• What did I do wrong? Identify the important decision rather than only show a score swing.
• Why was it wrong? Explain the tactical or positional mechanism where the evidence supports one.
• What was better? Show the stronger move or continuation and why it solves the problem.
• What should I learn from it? Turn the position into a reusable chess lesson rather than an isolated engine correction.
• What should I practise next? Connect repeated mistakes to retry positions, drills, lessons, or a training prescription.

ChessTing also supports different explanation depths — Beginner, Intermediate, Club, and Pro — so the same underlying chess evidence can be presented with less jargon for a developing player or with more technical detail for a stronger player.

Plain-English opponent scouting

The same philosophy applies to Scout. ChessTing does not stop at opening percentages, win-rate charts, or a list of PGNs. It interprets repeated game patterns and turns them into practical scouting language: how an opponent tends to play, which tendencies have enough evidence to trust, what may be exploitable, how to approach the opening, what to remember before the game, and what kind of positions or habits are worth practising against.

This is what powers features such as the Scouting Report, Victory Plan,10-minute prep plan, pre-game checklist, opening preparation, strengths and weaknesses, style/archetype explanations, composure observations, and opponent-specific practice.

Analysis becomes training, not just text

ChessTing's explanation layer is connected to the rest of the product. A repeated weakness can become a training priority; an important mistake can become a retry position; a tactical pattern can route into a relevant drill; an opening problem can route into study or preparation; and a reviewed game can produce a training prescription. The goal is a loop of evidence → explanation → action → practice rather than a static analysis page.

ChessTing is not an AI chess coach

The phrase plain-English chess coach describes how ChessTing communicates chess analysis. It does not mean ChessTing is powered by ChatGPT, Gemini, Claude, or another large language model.

ChessTing does not currently use an LLM or generative-AI model to write its scouting reports, game-review explanations, Victory Plans, prep advice, or training prescriptions. Those outputs are produced by programmed chess-analysis systems: Stockfish where engine analysis is required, statistical analysis of real games, deterministic chess rules and heuristics, tactical/positional classification logic, evidence thresholds, and programmed mappings from detected chess patterns to human-readable explanations and actions.

In other words: ChessTing explains chess in English, but it is not an AI chatbot.

Quick trust guide

Where analysis runs & what we store

ChessTing has four modes. This matters for privacy and for what you should expect from each feature:

Full detail: Privacy Policy and Cookie Policy.

Where the data comes from

• Chess.com & Lichess public APIs — profiles, ratings, and PGNs for the period you select (default last 3 months).
• Your browser — Stockfish WASM runs locally for game review, move-quality scans, and blunder detection.
• ChessTing analysis logic — deterministic rules, statistical models, evidence gates, tactical/positional classifiers, and explanation mappings turn raw evidence into readable coaching and prep.
• Sample limits — scouting loads up to 600 recent games in the browser to keep the tab responsive. Longer periods may be truncated.

Ratings on the scout dashboard

Listed rating

The number on the player's Chess.com or Lichess profile for their strongest active pool. This is not calculated by ChessTing — we display what the platform reports.

Performance rating

Tournament-style performance over games in the selected period:

points = wins + 0.5 × draws
smoothed_score = (points + 0.5) / (rated_games + 1)
performance = avg_opponent_rating + 400 × log10(smoothed_score / (1 − smoothed_score))

Requires at least 3 rated games with opponent ratings and known results. Draws count as half a point; a half-game prior prevents infinite numbers after a perfect or zero score. Heavily skewed samples are flagged, and a diversified rating is shown only when at least 3 eligible games remain after removing the dominant opponent.

Our best estimate

Uses listed rating when performance lacks three eligible games. Once performance is available, the card explains whether listed and sampled results agree. If one opponent dominates, we lean onlisted and never claim to have a diversified estimate when too few games remain.

Move quality rating (Stockfish)

Not a replacement for your Chess.com rating. It answers: “In this sample, did the moves look cleaner or sloppier than typical for this listed rating?”

1. Replay the last N games (you pick 4, 8, 15, or 30).
2. Evaluate every position before each player move with Stockfish at depth 12.
3. Compute centipawn loss per move from that player's perspective.
4. Skip mate scores or shallow evals that distort averages.
5. Cap each move at 500 cp in the aggregate.
6. Average cp loss across all graded moves, then anchor to listed rating.

expected_cp = benchmark(listed_rating)
adjustment = (expected_cp − avg_cp_loss) × 7.5
move_quality = listed + clamp(adjustment, −280, +350)

Expected centipawn loss by rating

Rough accuracy %

Display-only shortcut: accuracy ≈ max(0, 100 − avg_cp_loss / 6). This is not Chess.com's accuracy formula.

Phase breakdown

Opening = moves 1–12, middlegame = 13–35, endgame = 36+. Same anchor formula applied to each phase's average cp loss.

Game review (paste PGN)

Every position is evaluated once at depth 12. ChessTing then uses the engine result together with the actual board position and its programmed chess-analysis logic to classify the move and, for important moments, explain the reason in plain English rather than stopping at the numeric evaluation.

Precision score

Weighted average by severity (best = 100, blunder = 8, etc.), then averaged per player. Useful for comparing sides in one game — not 1:1 with Chess.com accuracy.

Explanation layer

The classification tells you how costly a move was; the explanation layer attempts to tell youwhy. It can use the played move, best move, affected piece or square, board geometry, engine swing, and recognised chess mechanism to create a readable lesson. Important mistakes can then be retried on the board or routed into a related training prescription.

Blunder gallery & scout engine notes

• Inaccuracy: ≥ 50 cp
• Mistake: ≥ 150 cp
• Blunder: ≥ 300 cp

Opening book (first 8 plies) is skipped for blunder mining to avoid theory noise.

Scout heuristics and plain-English scouting

Most of the scouting dashboard — openings, post-loss patterns, archetypes, style, and Victory Plans — is statistical and deterministic. These features do not depend on Stockfish unless the user runs an explicit engine scan, and they are not generated by an LLM.

ChessTing's job is not merely to display those statistics. The Scout explanation layer converts supported patterns into readable chess conclusions and practical preparation: what the opponent repeatedly does, why a pattern matters, how reliable the signal is, what the player may want to prepare, and which action should come next. Coach phrases are evidence-based hypotheses to verify over the board, not guarantees.

Evidence states

• 0 observations: not enough evidence; no numeric composite is presented.
• 1–4 observations: early read; confidence is capped below 50%.
• 5–11 observations: developing signal; direct evidence remains visible.
• 12+ observations: established overall sample, while each domain still keeps its own denominator.

Scout-score dimensions are pulled toward a neutral 50 until 20 games are available. This prevents two wins, zero timeouts, or no losses from becoming a 90+ strength/composure claim. The raw opening, result, colour, and move counts remain visible.

Domain-specific minimums

• Performance rating: 3 eligible rated games with opponent ratings.
• Accuracy claim: 3 games with reported accuracy.
• Composure score: 3 losses followed by another game.
• Colour edge: at least 4 games as each colour and a 15-point win-rate gap.
• Piece preference: at least 8 games, 120 non-pawn/non-king moves, and an 8-point usage lead.
• Radar: only measured axes are plotted; missing axes are omitted rather than filled with defaults.

ChessTing vs Chess.com

Expect 10–25 percentage-point gaps on accuracy vs Chess.com for the same game — normal, not a bug.

Known limitations

• Browser analysis can freeze on huge accounts — we cap at 600 games per scout.
• Shallow depth misses deep tactics; mates are excluded from move-quality averages.
• Small samples still swing; ChessTing labels and shrinks them instead of hiding the direct observations.
• Playing one opponent repeatedly skews performance, not listed.
• Plain-English heuristic explanations are gated by signal-specific evidence and should be read with their evidence/source games.
• ChessTing does not use an LLM to fill gaps when the evidence is missing; unsupported claims should be omitted or softened instead.

Questions or corrections?

Note the username, period, and which card looked wrong — we tune formulas from real reports. support@chessting.com