Everything published so far
Numbered in the order it was written. The series builds, so the early ones are the place to start.
- What A Chess Engine Actually Does When You Press Analyse 1,604 words A search tree plus an evaluation function is a decision procedure, not an oracle; understanding that Stockfish is ranking variations rather than pronouncing truths is the precondition for every other skill in this site.
- Your First Engine Analysis: A Ten-Minute Workflow For One Game 1,641 words Most improvers either skip analysis or spend an hour scrolling arrows; a fixed ten-minute pass with four defined steps beats both, and this post gives the exact sequence to repeat every week.
- Install Stockfish On Windows, Mac And Linux Without The Terminal Fear 1,714 words The official downloads page is a wall of CPU acronyms that stops people at step one; this decodes which binary to pick for your machine and gets an engine loaded in a GUI in under fifteen minutes.
- Plus 1.20 Explained: Reading The Eval Bar Without Panicking 1,710 words The eval is a pawn-denominated forecast of a perfect game, not a probability that you will win this one; calibrating what each band means for a 1400 changes which positions you bother resigning or fighting.
- Inaccuracy, Mistake, Blunder: What The Labels Actually Measure 1,684 words These are arbitrary centipawn thresholds applied by the website, not chess judgements; knowing the cut-offs reveals why one-move oversights and slow strategic drift get scored identically and why that misleads you.
- Stop Reading The Engine Line First: Guess-Then-Check Analysis 1,767 words Turning the engine on before you commit to your own answer converts active recall into passive reading; writing down your candidate move first is the single habit that makes analysis produce learning.
- Depth Is Not Truth: Why Stockfish Changes Its Mind At Depth 30 1,774 words Reported depth is a heavily pruned average, not a guarantee that every line was searched that far; eval instability across depths is a signal about position sharpness that you should read deliberately.
- Which Engine Should A Club Player Actually Run? 1,411 words Engine choice at 1000-1900 barely matters for finding blunders and matters enormously for explanation quality; the right answer is a two-engine setup, not the strongest single binary.
- Is Playing Bots Cheating Yourself? Sparring Versus Real Games 1,452 words Bots remove the two things that actually make humans improve under pressure — rating risk and opponent psychology — which makes them excellent for drilling and useless as a substitute for rated play.
- Can An LLM Coach Your Chess? Establishing A Baseline 1,746 words Language models cannot calculate but can explain, plan and summarise; separating those two jobs and pairing the model with an engine is the only configuration that survives testing.
- Lichess Analysis Board Versus A Desktop GUI: When The Browser Isn't Enough 1,697 words The browser engine is throttled, single-line by default and forgetful; this identifies the four specific tasks where that genuinely costs you and where installing a desktop GUI is overkill.
- Building A Blunder Log That Actually Changes Your Play 1,644 words Recording positions without recording the reasoning that failed produces a puzzle collection, not a diagnosis; a three-column log capturing the move, the false belief and the missed cue is what makes patterns visible.
- Maia Explained: The Engine Trained To Play Like You 1,872 words Maia optimises for predicting human moves at a target rating rather than for winning, which makes it the only widely available engine whose mistakes are informative about your own.
- Why Stockfish's Best Move Is Often Useless Advice To You 1,497 words The top line frequently depends on a defensive resource fifteen plies deep that no club player would find; the second or third choice is usually the move you should actually study.
- Choosing A Sparring Time Control: Why Rapid Beats Bullet Against Bots 1,597 words Bots don't flag, tilt or premove, so fast time controls against them train nothing but mouse speed; 15+10 sparring is the format where bot practice transfers to human games.
- Hash, Threads And Multi-PV: The Three Settings That Matter 1,783 words Everything else in the options dialog is noise for an improver; this explains what each of the three does to your analysis output and gives concrete values by RAM and core count.
- Multi-PV Is The Improver's Setting: Reading Three Lines Instead Of One 1,613 words Single-line analysis tells you that you were wrong; three lines tell you how wrong and whether your move was a near-miss or a category error, which is the difference between a score and a lesson.
- Tagging Your Losses By Cause, Not By Opening 1,727 words Improvers blame openings because openings are easy to name; a six-category cause taxonomy applied across twenty games almost always reveals the real leak is calculation hygiene or time use.
- Leela Versus Stockfish: Two Different Kinds Of Wrong 1,641 words Policy-guided MCTS and alpha-beta search fail in opposite directions — Leela underrates forced tactics, Stockfish underrates long-term compensation — and knowing which is which tells you when to run both.
- Prompting An LLM For Chess: What Works And What Hallucinates 1,630 words Asking a model for moves produces confident illegal nonsense; asking it to explain an engine line you paste in produces usable coaching, and the prompt structure that enforces this is reproducible.
- The Twenty-Game Audit: Finding Your Pattern, Not Your Worst Move 1,640 words One game is anecdote and the engine will always find something; only aggregating across twenty games separates recurring structural weaknesses from bad-day noise.
- Chess.com Bots Rated: Do The Personalities Play Like Their Numbers? 1,801 words Testing the named bots reveals their ratings are badly non-uniform — strong tactically, artificially weak positionally — so a 1400 bot is not a 1400 opponent and shouldn't be used to gauge progress.
- Cloud Eval And Free Depth: Getting Deeper Analysis Without New Hardware 1,548 words Lichess cloud evals, shared analysis caches and free notebook compute cover most of the depth an improver needs; buying a faster CPU is the last resort, not the first.
- Accuracy Percentage And Game Review Scores: What They Hide 1,766 words Accuracy rewards quiet positions and punishes sharp ones, so a 96% loss and a 78% win can represent identical play quality; treating it as a progress metric actively misleads you.
- NNUE In Plain English: Why Stockfish Grew A Neural Network 1,612 words NNUE replaced hand-written positional rules with a small learned network that runs on CPU, which is why modern Stockfish evaluations of closed and sacrificial positions differ sharply from the versions most club advice was written about.
- Turning Engine Findings Into A Weekly Training Plan 1,848 words Analysis without a conversion step is entertainment; this maps each error category to a specific drill, dose and review date so the week's study is chosen by your games rather than by mood.
- Sparring A Weakness: Building Positions The Bot Must Punish 1,596 words Playing out a position from your own loss against a bot, repeatedly, from the move before the error is far higher yield than another full game — and most GUIs support it in two clicks.
- When The Eval Says Plus Three And You Still Can't Win It 1,420 words The engine's advantage assumes an engine will convert it; the gap between a +3 evaluation and a human-convertible position is a measurable skill, and analysis should target that gap directly.
- Feeding PGNs To An AI Coach: Formats, Context Limits And Accuracy 2,046 words Raw PGN forces the model to track a board it cannot see; supplying FENs at key moments plus engine evals collapses the hallucination rate, and this shows the exact pre-processing that does it.
- Opening Prep With An Engine Without Memorising Twenty Moves 1,655 words Engines are terrible repertoire teachers and excellent repertoire debuggers; use them only at the branch points your own games actually reached, and stop at the move where you can state a plan.