Choosing an analysis depth
Four settings, one real trade-off, and one place the setting does not apply.
Depth is the only analysis knob in the app, and it is worth understanding because it changes the answers, not just the waiting time. Settings offers four choices:
| Setting | Depth | Use it for |
|---|---|---|
| Fast | 5 | Skimming a long game for the obvious disasters |
| Normal | 10 | The default; sound for most review |
| Deep | 15 | Checking a position you disagree with |
| Very Deep | 20 | Tactical positions where the line matters |
What depth actually means
Depth is how many plies — half-moves — the engine searches ahead before committing to a verdict. Depth 10 is five moves for each side. It is not a time limit: the engine searches until it has finished that depth, so the same setting takes longer in a complicated middlegame than in a sparse endgame.
The search is not exhaustive. Stockfish prunes aggressively, following forcing lines much further than depth 10 suggests while dismissing quiet alternatives early. So "depth 10" is better read as a measure of effort than as a guarantee about what the engine considered.
Why deeper can reverse a verdict
The interesting failure is the horizon effect. If the refutation of a move lies one ply beyond where the search stops, the engine cannot see it, and the move looks fine. Search one ply further and the same move is a blunder. This is not noise — it is the search boundary moving.
The practical consequences:
- Sacrifices are the classic case. A piece given up for an attack that pays off in six moves looks simply losing at depth 5.
- Deeper analysis grades harsher. The same game re-analyzed at depth 20 finds refutations depth 10 missed, so moves that were "good" become inaccuracies. Your play did not get worse.
- Comparing evaluations across depths is meaningless. If you re-run a position at a higher depth, re-run the ones you are comparing it against.
So when a classification surprises you, the right response is to raise the depth on that position before believing it — and equally, before dismissing it.
What it costs
Each additional ply multiplies the work rather than adding to it, so the settings are not evenly spaced in time. Going from Normal to Very Deep is far more than twice the wait, and on a long game the difference is minutes rather than seconds, because every position in the game is analyzed.
The web version runs Stockfish compiled to WebAssembly in a single worker thread, which is slower than a native build with several threads. That is the price of doing the analysis in your browser instead of on a server, and it is why the Android app feels quicker at the same setting.
One thing the setting does not affect: free play. When you are moving pieces on the front page, each position is analyzed at the default depth of 10 regardless of what you have chosen in Settings. The depth choice applies to analyzing an imported game. If you want a position examined more deeply, import it rather than playing it out.
A reasonable habit
Leave it on Normal. Review the game, find the two or three moments that actually decided it, then re-examine those at Deep or Very Deep. Analyzing an entire game at depth 20 spends most of its effort confirming that uncontested moves were fine, and the extra precision lands where you were never going to learn anything.
The exception is a sharp tactical game — a sacrificial attack, a complicated endgame — where depth 10 may be genuinely wrong about the critical line rather than merely imprecise. There, start deep.
Open the analyzer →