What Stockfish depth is enough for opening analysis?

Depth 18-25 is sufficient for almost all opening analysis. Below depth 18, Stockfish still misjudges some quiet positions; above depth 25, diminishing returns kick in and analysis takes much longer without meaningfully changing opening-phase evaluations. For tactical positions (Sicilian Najdorf English Attack, sharp King's Indian lines), aim for depth 22-25 with multi-PV set to 3-5 so you see the top alternatives. For quieter positions (London, QGD, Caro-Kann Classical), depth 18-20 with multi-PV 3 is plenty. Modern cloud engines on Lichess run deeper by default — use their Request computer analysis rather than running Stockfish locally unless you need multi-PV.

How many games should I review per week?

Three to five losses per week, focused on opening-phase deviations, is the sweet spot for most club players. More than that and you lose depth of analysis — you end up skimming rather than extracting fixes. Fewer than three and you miss the pattern repetition needed to solidify fixes. Prioritize games where you lost material or evaluation in moves 1-15, and skip games where the opening was fine but the middlegame collapsed (those belong in a tactical/strategic review, not a deviation review). Spend 30-60 minutes per game following the 6-step process in this guide.

Is centipawn loss the same as accuracy?

No — they're related but measure different things. Centipawn loss (CPL) is the average evaluation drop per move, in units of 0.01 pawns. Accuracy is a derived percentage based on CPL and the position's complexity (Lichess uses a logarithmic formula by Uri Blass). A move in a simple position losing 30 centipawns drops accuracy more than the same loss in a wildly complex position, because the simple move was 'easy' to get right. For opening review, focus on raw CPL per move in the first 15 moves — it's more actionable than a single accuracy number.

How to Find Where Your Opening Preparation Broke Down (Deviation Finder Guide)

One unprepared reply in the opening can cost you the whole game before move 15. The fix is simple: find the first move that left your notes, then build a confident answer. This guide shows how to find where your opening preparation broke down, using databases, engines, and short practice loops so you can tag the exact deviation, measure its cost, and patch your repertoire fast.

You'll replay the game, compare moves to an opening database, run an engine check, study one or two model games, drill the line, and schedule spaced reviews. Expect 30 to 60 focused minutes per game. ChessAtlas automates most of this loop with a built-in deviation finder layered on top of one-click game import from Lichess and Chess.com, so the manual process below collapses into a minute per game once you wire it up.

Prerequisites

Gather these tools before you start:

Chess database access: Lichess Opening Explorer, the Chess.com database, or ChessBase for move frequencies and win percentages at your level.
Chess engine: Stockfish on Lichess or Chess.com, aimed at depth 18 to 25 with multi-PV enabled so you can see 3 to 5 candidate moves at each branch.
Game replay tool: ChessAtlas, Lichess, or Chess.com with PGN import and move-time display.
Note-taking system: A document, notebook, or the comments field on your ChessAtlas chapter where you log deviations, eval changes, and chosen fixes.

ChessAtlas integrates with Lichess and Chess.com for one-click game import, these platforms are complementary tools, not competitors. Use the database and engine from the platform you play on, then store your repertoire in whichever tool best supports spaced-repetition drilling.

You should know your main repertoire well enough to recognize where your prep ended and your improvisation began, that's the whole point of this process.

Step 1: Replay Your Game and Flag the Opening Phase

Import your PGN into Lichess, Chess.com, or ChessAtlas, then open the analysis view with the full move list and board. If you played online, the game is already in your history.

Replay from move 1, focusing on the opening phase, usually moves 1 to 10 in tactical lines and up to 15 to 20 in quieter systems. The phase often ends once the pawn structure stabilizes and clear middlegame plans appear.

Mark the last move you knew cold from your notes. For example: "Prepared through 7...Be7; 8.h3 was new." That is your preparation boundary. Players who burn a lot of time on early moves usually hit these boundaries then and there, so watch your recorded move times.

Write down each move where you hesitated or spent an unusual amount of time. Time spikes on moves 6 to 12 are classic signs you left theory. These positions are your initial deviation suspects.

Step 2: Use Database Tools to Map Main Lines Versus Deviations

Chess opening database analysis showing main line versus deviation frequencies

Open Lichess Opening Explorer or a similar database. Enter your game's moves one by one to see top continuations at each branch, along with play percentages and results at your rating band.

Filter by a rating range near your opponent's level. On Lichess, set the rating filter so statistics reflect likely over-the-board choices at club level, not only grandmaster play.

Check frequency at each move. A practical deviation often shows under 10% frequency in the database or is absent from your notes entirely, a useful rule of thumb for flagging "this is the move that surprised me." For example, you might see "At move 8: 60% c6, 15% b5, 10% Qc7" (illustrative numbers) and realize your opponent's 8...b5 is the rare side branch.

Find the first point your opponent left the main line, then record both choices with numbers. Example: "Deviation at 8: Black played 8...b5 (12%, Black scores 45%) instead of 8...c6 (60%, Black scores 52%)." (illustrative numbers).

Check your opponent's prior games on Lichess, Chess.com, or FIDE databases. If they've used this sideline repeatedly, expect it again. If it's new for them, treat it as a prepared surprise and prepare a clear antidote before the rematch.

Log the deviation move, frequency, win rates, and the main-line alternative. Turning fuzzy impressions into concrete numbers will guide your engine work and practice plan.

Step 3: Run Engine Analysis to Quantify the Breakdown

Load the game into engine analysis. On Lichess, click Request computer analysis; on Chess.com, use Game Review. Use Stockfish at depth 18 to 25, with multi-PV set to 3 to 5 so you see viable alternatives.

Track the evaluation bar and centipawn loss per move. An eval of +0.77 means White is better by about three-quarters of a pawn. Centipawn loss is the gap between the engine's best move and the one you actually played.

Lichess and Chess.com classify moves loosely as inaccuracy, mistake, and blunder based on how much eval they drop. Exact thresholds vary by platform, but as a rough guide: a small drop (under half a pawn) is an inaccuracy, a clear drop (half-pawn to a pawn) is a mistake, and anything above a pawn is a blunder. Note these in moves 1 to 15, where opening errors snowball fastest.

Zoom in on the deviation. Compare evals before and after. For example: "Before 8...b5, eval +0.5. After 8...b5, still +0.5, the deviation was playable. My reply 9.Nd2?! dropped eval to -0.2. That's the real breakdown, not their move 8."

Use a simple checklist around the deviation: list immediate threats, generate 2 to 3 candidate moves prioritizing checks, captures, and threats, then compare your evaluation to the engine's view on material, activity, and pawn structure.

Click the engine's top suggestion and let it run 5 to 10 moves forward. Document the improvement. "At 8...b5, engine recommends 9.c4 (+0.4), hitting d5. I played 9.Nd2?! (+0.1), ceding the initiative."

Finish with clear metrics: exact breakdown move, eval swing, your centipawn loss, and the engine's better line with its evaluation.

Step 4: Investigate the Deviation and Build a Counter

Return to the database at the deviation position and look forward 5 to 10 moves. Click through popular continuations to see practical plans used by strong players in this exact branch.

Extract ideas, not just moves. If Black expands with 8...b5, check whether top games hit back with c4, calmly develop with Nbd2, or counter on the kingside. Note recurring piece placements, pawn breaks, and king-safety choices.

Watch 1 to 2 model games starting from your deviation position. Search for the position in the Lichess masters database or filter to 2400+ games in your database. Track where pieces land by move 12 and which pawn breaks decide the structure.

Update your file with a short counter-repertoire. "Vs 8...b5 in the Caro-Kann: play 9.c4, aiming at d5. Eval +0.3. After 9...Bb7 10.Nc3, prepare d4-d5 or Nf3-e5."

Prepare 2 to 3 concrete responses with one-line reasons. Understanding the plan beats memorizing 20-move trees, because a 20-move tree collapses as soon as the opponent deviates again, and the plan still holds.

Step 5: Practice and Test Your Fixes

Chess player drilling opening repertoire fixes against engine practice games

Set the deviation position on Lichess and choose Play with computer. Play it 2 to 3 times at 10+0 or 5+3, executing your plan under time pressure.

After each game, recheck with the engine. Did you play c4 on time? Did the position's eval match your notes? Log new mistakes and update your file with the corrected move and one sentence on why.

Drill active recall. Use "guess the move" from the key position, or generate puzzles from your games. Active recall, predicting the move before looking, beats passive review every time.

Play targeted online games steering into the same opening to trigger the deviation. Track time use: if the position cost you 8 minutes before, aim to respond in 60 to 120 seconds now, with the prepared move.

Repeat 3 to 5 times for the critical branch. Aim for consistent execution in your test lines. Note comfort level and any lingering confusion directly in your repertoire file.

Step 6: Review and Iterate Your Repertoire

Write a short deviation report. Include move number, the deviating move, the eval change, your reply quality, the engine's fix, and practice results. "Vs opponent123, Caro-Kann. Deviation: 8...b5. My 9.Nd2?! lost about 65cp. Correct: 9.c4 (+0.3). Practiced 4 games, responded correctly in 3/4, average 90 seconds to respond."

Schedule spaced reviews. Reserve one weekly session for sidelines. Tag positions as deviation or needs review in ChessAtlas, then follow the prompts to keep rare lines fresh. See our spaced repetition guide for scheduling details.

Track progress qualitatively across your next 10 games: fewer inaccuracies in moves 1 to 15, lower clock time per opening move, and post-opening evals closer to equal. The improvement compounds when you fix breakdowns systematically rather than memorizing more theory blindly.

Expect surprises to continue, but make them cheaper. A repeatable process, not deeper memorization alone, turns unfamiliar positions into steady, explainable choices.

Common Mistakes to Avoid

Mistake 1: Analyzing too many candidate moves. Looking at 5 to 6 options creates shallow, tangled lines. Limit yourself to 2 to 3 candidates, starting with checks, then captures, then threats.

Mistake 2: Ignoring the opponent's best reply. Skipping zwischenzugs, counter-checks, or key captures leads to optimistic errors. After each candidate move, ask: "What is their best check, best capture, and most dangerous threat?" This catches most tactics.

Mistake 3: Blitzing memorized lines without verifying the position. Move-order shifts and transpositions change evaluations. Use a brief reflect, pause, react check before each move, and calculate to a calm position, not to a "looks fine" stop.

Here is the action plan in one line: identify the first deviation with database stats, quantify the cost with engine checks, build a clear counter-line with one or two model games, drill it, and schedule reviews.

ChessAtlas can streamline this loop: import games, flag the deviation, store your fixes in the repertoire, and get spaced-review reminders. That keeps prepared answers ready when the same sideline appears again.

Find the first off-book move using database frequencies, sub-10% moves are strong deviation candidates.
Measure the damage with engine evaluation; depth 18 to 25 and multi-PV 3 to 5 is enough for opening work.
Study 1 to 2 model games from the deviation, capturing plans, piece placement, and pawn breaks.
Prepare 2 to 3 responses with one-line reasons, then drill under 10+0 or 5+3.
Track accuracy, response time, and evals; review deviations weekly with spaced repetition.

Do a fast run on your latest loss: mark the deviation, run a 10-minute engine check, add one model game, and play two practice runs from the key position today.

Skip the Manual Loop

Every step above becomes one click once ChessAtlas is connected to your accounts. The deviation finder auto-scans imported games, flags the first off-book move, and queues the correct reply for spaced review. No PGN exports, no hand-built spreadsheets.

Try it free →