Climbchess
What's confusing, broken, or missing? We read every message.
Weak players burn 5 minutes calculating an irrelevant move. Strong players spend 5 seconds picking the 2-3 moves worth calculating — then go deep on those. This drill trains that skill: look at the position, click 1-3 squares with pieces you'd consider moving, submit. The engine reveals its top-3 moves; you score on whether you spotted the right pieces. 30 seconds per position. Most improving players say this feels harder than puzzles — that's the point.
Bad day? Save your streak with the daily minimum.
20 minutes a day, prescribed by your tier and recent performance. Most progress in chess comes from consistency — daily practice beats sporadic marathon sessions. Streak stays alive as long as you complete one daily session.
You've solved 5,000 puzzles. Your rating barely moved. Here's why.
320 patterns from the world's best chess engine. We measure which ones you can't see, then drill them at exactly your difficulty. Climb faster than random puzzle-rush.
Free forever • No email • No signup for the rating test • Delete your data anytime.
Most trainers conflate four distinct skills into one undifferentiated puzzle stream. Climbchess separates them.
When a coach watches you play, they don't just say "you're a 1500." They say things like "your pattern recognition is solid but your calculation breaks down past three moves," or "you see the right ideas, you just don't prune candidates well — you burn five minutes calculating an irrelevant move." Those are four different skills, and they fail in different ways. A puzzle stream that gives you one rating can't tell them apart.
Climbchess splits training into four skills, each with its own rating, each with its own dedicated mode. Pattern recognition — seeing the right idea on the board. Calculation — reading lines accurately N moves ahead. Memory — holding a sequence in working memory. Move selection — choosing which two or three candidates are even worth calculating before going deep. After about fifty puzzles you can see which one is bottlenecking your play.
The substrate underneath is 4,505 interpretable concepts extracted from Leela Chess Zero via sparse autoencoders. Standard platforms tag puzzles with twenty to sixty manually-curated themes (fork, pin, mate-in-2). Ours come from reading the residual stream of the strongest open-source neural engine and labelling each feature by its top-activating positions. Of the 4,505 concepts, 562 fire on at least one puzzle in the current corpus. When you miss a puzzle, the trainer tells you which concept-rating sits below the puzzle's hardest-firing concept, then queues puzzles tagged with that concept until the rating moves. Full methodology with DOI.
Climbchess scores every pattern against your skill — then drills your weakest at your difficulty.
You've watched the YouTube videos. Bought the Chessable course. Solved 5,000 puzzles. Watched a 90-minute GM analysis on the Caro-Kann that you don't even play. Your rating barely moved. Here's why: none of it targeted the pattern you actually keep missing.
It's never just the subscription fee.
Even paying for all of this, no human coach can measure your 320 neural blind spots.
None of it targets the patterns you're actually weak on.
Climbchess: Free. 15 min/day.
Drills the one pattern that's actually losing you games — at your exact level.
Each mode trains a different skill. The point is that they don't share a rating — so you can see which is actually limiting you.
A 12-rep daily session tailored to your current concept profile. Mixes weak-concept drills with the modes you're underusing.
Adaptive 30-day journey. Three layers: principles, named tactics, SAE patterns. About twenty concepts at your current band, not 4,505.
Open SAE-tagged tactic stream. 9,950 positions, every one decomposed into the concepts that fire on it.
Strips the position to the minimum number of pieces required to read the line, asks you to visualise 2–3 ply, and scales the piece count up as you improve. Calibrating to "minimum required pieces" first lets you train calculation in isolation, without it being bottlenecked by your pattern recognition or short-term visual memory at the same time.
Shows a position and a sequence, hides it, asks you to reconstruct. Your span — longest sequence repeated correctly — is logged separately from your tactical rating.
Pick two or three candidates without calculation; the trainer reveals whether the strongest move was in the chosen set. Distinguishes "spent five minutes on the wrong move" from "spent five seconds pruning, then went deep on the right two."
Bisects your true ELO in roughly ten minutes. No signup.
80 curated endgame studies with concept tags and solving guides. Paste a PGN and the post-mortem extracts the SAE concepts that fired on your mistakes, then feeds them into the curriculum — so the trainer learns what's costing you rating in actual games, not abstract puzzles.
Adaptive 15-puzzle test. 60 seconds. Get a real rating, not a vibe-check. No email needed.
We score every one of 320 patterns with a separate Elo at your tier. The lowest ones are your blind spots — they decide the games you lose.
15 minutes a day. We pick puzzles that hit your weakest pattern — at your difficulty. Watch your per-pattern Elo move week by week.
Peer-reviewed research shows competitive chess players peak around age 43, not 16 (Roring & Charness, 2007, Psychology and Aging).
Documented adult improvers have reached Master starting from scratch in their 30s, 40s, and 50s. The "you're too old" myth is folk wisdom, not data.
We won't promise Grandmaster — the world's top ~2,000 players, rated 2500+. But Expert (2000) and Master (2200) are achievable with the right method. Climbchess is the right method.
Free, no email, runs in your browser. The list, plainly:
We don't know if this works for everyone yet. Things we already know don't work, or work poorly:
No signup. No email. Just a 15-position adaptive test. Then you'll know.
We don't dump 320 concepts on you. You only need ~20 at your current level. Three layers: principles (how to think), named tactics (what every coach teaches), and position patterns (the recurring shapes that decide games at your tier). When you graduate, the curriculum rotates.
Watch a sequence of moves play out. The board resets — you replay the same sequence by dragging. No calculation. Pure visualization. Build the muscle in isolation.
The #1 skill that separates strong players from weak ones isn't calculating deeper — it's picking the right moves to calculate.
Paste your Lichess username. We pull your last 20 games, find each loss, run Stockfish on every move, and identify the critical mistake that lost the game. Each mistake becomes a position in your review queue — drilled like normal training, but with the context of your real game.
Your top missed patterns from real games. Click to generate a 30-puzzle remedial pack targeting that pattern at your optimal difficulty.
Don't memorize moves — internalize the recurring shapes your opening produces. Pick an opening, see the patterns it most commonly creates (extracted from a chess AI's brain), then drill puzzles built on those patterns. That's how GMs actually understand openings.
Build your opening lines, drill them with spaced repetition. Most useful for 1700+ players where opening preparation matters.
Theoretical endgames are taught knowledge, not pattern recognition. Even top engines use tablebases for these. Master ~30 essential positions and you have strong tournament-player endgame technique (roughly the level of a titled "FIDE Master" rated ~2300). Each one is drillable + walks you through the winning sequence.
Most players over-train what they're already good at and ignore their weaknesses. Your rating is bottlenecked by your weakest concept, not your strongest. This panel shows where you've been spending time vs where you actually need it.
A 30-day beginner journey is the right scale to see real movement. Show up, even briefly.
SAE-derived skill ratings for the chess concepts you exercise. Lower scores = where to train next.
Accuracy × Speed × Depth per pattern. Your weakest axis is the one breaking your rating in real games.
Each pattern fits T(N) = T₁ × N−α. We classify whether you're learning normally, plateauing, mastering, or regressing — and recommend what to do.
Auto-advances when you hit 85% accuracy at your current depth's target rating. Steps back if you drop below 65%.
Shows how cluttered the board can be before you need pieces removed to see each pattern. Lower numbers = you spot the pattern even in messy positions (stronger recognition).
Glows the squares the engine's network is "looking at" for this puzzle's concept. Off = clean board for testing yourself.
Right/wrong feedback uses both colour and a ✓/✗ icon — colour is never the only signal. Switches palette to a colour-blind-safe set if you pick a CVD profile.
Uses your browser's built-in voices. Adds an auditory encoding pathway to strengthen pattern memory.
Runs synthetic users at fixed ratings (600–2800) through N training puzzles and verifies per-pattern Elo + speed Elo + tier classification converge correctly. Uses the same code paths as real training. Does NOT touch your real progress — runs against an isolated state copy.
Permanently delete your account and all training data from this device and our servers. This action cannot be undone.
Permanently delete all your training data? This cannot be undone. Your local progress and any cloud-synced state will both be removed.