Make thinking visible
Moves considered, time, hints, the played line and the opponent's best response help reveal where the decision process broke down.
A practical way to learn chess
The CraftChess learning cycle
Correct now. Remember later. Apply it in a new position.
CraftChess is a chess training system built to turn decisions into lasting learning. It combines focused exercises, immediate correction, spaced review, adaptive difficulty and complete games against a level-adjusted opponent. The goal is not to collect easy wins; it is to think more accurately, remember useful patterns and apply them independently.
Why it exists
Training judgment, not answer memorization
A chess exercise is useful only when it changes what the student notices and calculates in a later position. CraftChess therefore records the first decision honestly, teaches the missed idea, and then looks for new evidence over time.
Correct with evidence
Stockfish verifies moves and continuations. Explanations translate those verified facts into practical chess language rather than inventing tactics.
Build independent recall
Immediate correction is followed by delayed review and unfamiliar positions, because seeing an answer once is not the same as being able to find the idea alone.
The complete loop
From one decision to new evidence
Every exercise passes through the same protected sequence. The solution remains hidden while the student is thinking; teaching begins only after the attempt has ended.
- 01
Observe and calculate
The student studies the position, names candidate moves and may request graduated hints.
Legal-move assistance shows possibilities, never the best move.
- 02
Commit to a move
The move or line is played on the board. The original attempt is recorded and never rewritten later.
First-attempt evidence remains separate from every later repeat.
- 03
Verify with the engine
Stockfish compares the decision with the strongest continuation and grades the move's practical cost.
The engine is the chess authority; AI is used only to narrate verified facts.
- 04
Understand the consequence
Replay, explanation and guided board animation show what worked, what the opponent could do and what to check next time.
The explanation is available after the attempt, so it cannot leak the answer.
- 05
Reconstruct and review
Missed positions can return in session reinforcement and later as scheduled reviews.
Immediate reconstruction supports correction; delayed review provides stronger recall evidence.
- 06
Apply on fresh boards
New positions continue to test the same developing skills without depending on memory of one diagram.
The next results update the learner profile and begin the cycle again.
Inside focused practice
What happens during a puzzle session
The main set measures independent work. When it ends, a clearly labelled reinforcement phase may offer exact repeats of missed positions without extending or changing the main progress record.
Main set
A position is selected
Difficulty and subject are chosen from the learner profile, due reviews and a small variety pool.
The student works independently
Candidates, thinking time and requested hints are captured. Hints become progressively more specific but remain optional.
The result is finalized once
The board accepts the complete expected line or a verified equally strong alternative. A deviation ends the attempt and is evaluated safely.
Feedback becomes available
The app can replay the consequence, explain the best move, narrate it and animate a short verified continuation.
Session reinforcement
After the main set, missed positions may reappear under a separate heading. This is a short correction ritual: useful for rebuilding the line while it is fresh, but not treated as proof of long-term retention.
- It is optional and can be ended immediately with “Finish for now.”
- The main progress grid remains frozen; reinforcement adds no extra cells.
- Results do not change Glicko ratings, bands, the coach journal or the original miss.
- A failed reconstruction does not create a recursive chain of new reinforcement items.
- The original scheduled review remains available for a later session.
Four kinds of evidence, kept separate
- 01
First attempt
Records the independent decision, candidates, time, hints, move quality and rating effects.
- 02
Immediate reconstruction
Corrects the exact position at session end; it changes no rating, main progress or original result.
- 03
Scheduled exact review
Tests recall after a delay and updates skill evidence without rewriting the first attempt.
- 04
Fresh position
Asks the learner to calculate again on an unfamiliar board and provides broader evidence of application.
Personalization
How the training adapts without becoming easy
Adaptation changes what the student sees next, not the rules of chess. A lower skill estimate leads to better-matched material, while every move is still judged by the same engine-backed standard.
Evidence collected
- Solved or missed line
- Hints used
- Thinking time
- Skill and puzzle rating
- Move quality and error pattern
CraftChess
Learner profile
Glicko-2 ratings estimate overall and skill-specific performance with uncertainty. Recent evidence matters, but one result never defines the student.
What changes next
- Puzzle difficulty
- Skills receiving attention
- Scheduled reviews
- Calibration for new learners
- CPU strength and pressure style
Most new work
Targeted positions
Fresh exercises emphasize skills whose rating is reliably below the learner's overall profile.
A regular share
Scheduled review
Due positions return after a delay so recall is tested rather than assumed from immediate correction.
A smaller share
Useful variety
Randomized, level-appropriate material prevents the session from becoming too predictable or narrow.
After the attempt
How move quality and feedback are interpreted
For legal deviations, CraftChess compares the played move with Stockfish's best line from the same position. Centipawn ranges help describe impact; mate and decisive transitions take precedence. These labels explain the decision—they are not a judgment of the student.
Expected move
Exercise line completed
The student completed the line selected to teach the exercise's idea.
Equally strong solution
0–30 centipawns
The move differs from the stored line but Stockfish verifies it as practically equal. CraftChess accepts it as fully correct.
Playable move
31–79 centipawns
The move remains reasonable but misses part of the exercise's main opportunity. It receives constructive feedback and does not complete the exercise.
Inaccuracy
80–179 centipawns
The decision gives away a noticeable part of the position without necessarily deciding the game immediately.
Mistake
180–349 centipawns
The decision loses an important advantage, material or tactical opportunity and leaves the position clearly worse.
Critical error
350 or more centipawns, or mate override
The move changes the position decisively, misses or permits forced mate, or loses a major tactical resource. Tactical puzzles naturally produce more of these than quiet positions.
Could not be verified
No reliable engine result
If analysis times out or is incomplete, the attempt is not penalized and does not update the learning rating.
Severity and cause answer different questions
Move quality describes how much the position changed. A diagnosis asks which thinking habit may have produced the decision. CraftChess records a cause only when the board evidence is strong enough.
Vision and safety
Examples include an unprotected piece, a missed recapture, a missed opponent check or stopping the calculation after one move.
Calculation and choice
Examples include missing a candidate, overlooking the opponent's resource, choosing the wrong move order or evaluating the final position too early.
Advanced decisions
Examples include missing prophylaxis or failing to convert an advantage. A severe move without a reliable cause is not forced into a diagnosis.
The hint ladder
- Hints move from orientation toward increasingly concrete guidance.
- The highest rung used is recorded so assisted success is not confused with unaided recall.
- Hints never change chess truth, and no hint or voice reveals the solution before the attempt ends.
Engine-grounded coaching
- Stockfish supplies the move, best reply, verified continuation and tactical facts.
- The language model or trusted compositor explains those facts; it cannot overrule the engine.
- Text, voice and guided replay are post-attempt teaching tools and can be skipped without delaying Next.
Two complementary activities
Focused exercises and complete games
Problems isolate specific decisions; games test whether those decisions survive the complexity, time pressure and uncertainty of a complete game. Their records stay separate so each tells an honest story.
Solve problems
Short, engine-verified positions train candidate generation, calculation, defense, tactics, endgames and other skill groups.
- Adaptive puzzle difficulty and subject
- Candidate capture and graduated hints
- Immediate explanation, replay and reinforcement
- Scheduled review in later sessions
Play against the CPU
A full game uses a Stockfish-based opponent whose target Elo and pressure profile are derived from the learner's established record.
- Blitz, rapid and guided classical pacing
- Adaptive, aggressive, tactical, positional or technical style
- Human-like thinking delays instead of instant replies
- Post-game evaluation chart and teachable moments
CPU games do not directly change puzzle ratings. They provide a separate transfer signal: whether training habits appear during complete play.
For parents and coaches
A record designed for decisions, not surveillance
The coach view organizes evidence so an adult can choose the next teaching focus without reducing the learner to a single score.
Skill profile
Overall and skill-specific ratings include uncertainty, helping distinguish a stable weakness from an area that simply lacks enough observations.
Move quality
Expected, equivalent, playable, inaccurate, mistaken and critical decisions are shown over time with unverified and legacy records kept separate.
Patterns and journal
Repeated, engine-supported diagnoses become hypotheses to observe. They are not permanent labels or claims about personality.
Games and reviews
The coach can examine time controls, recent games, critical moments, hints and the difference between first attempts and later recovery.
Boundaries that protect the learning signal
What CraftChess promises—and what it does not claim
Good educational data depends on preserving context. CraftChess keeps source attempts, later reviews and different activities distinct, and fails safely when an engine result cannot be trusted.
Integrity of the lesson
- The engine is the source of chess truth; generated prose is narration only.
- Answers, candidates and solutions are protected until the attempt is complete.
- A later success adds evidence but never rewrites an earlier miss.
- Immediate reinforcement cannot change ratings, progress or journal evidence.
Honest interpretation
- A familiar-position success demonstrates recall of that position, not automatic transfer.
- One session or one label does not define ability or mastery.
- Confidence, hints and speed are context; confidence never feeds Glicko.
- Only linked guardian accounts can open a learner's coach record.
Common questions
How to read the experience
These short answers clarify the choices students and guardians see most often.
Why did I receive a difficult exercise after making a mistake?
Difficulty responds to a longer evidence history and uncertainty, not to one result. The selector also mixes targeted work, due reviews and variety, so consecutive positions do not form a simple easier-or-harder staircase.
Does using a hint count the same as solving unaided?
No. The line can still be completed, but the highest hint rung is preserved and contributes less evidence than an unaided first attempt or delayed recall.
Why can a legal move still be marked as a missed exercise?
Legal means the rules allow it. The exercise asks for the strongest idea. A playable move may keep the game going while still missing the tactical or strategic opportunity being trained.
Why are many tactical misses classified as critical?
Tactical exercises are usually selected around a decisive resource. Missing it often changes a winning position into equality or a loss. The label describes the position's change, not the student's worth.
Can a student choose the puzzle difficulty?
Not directly. That would weaken the learning measurement. The student chooses the activity and learning aids; the adaptive system chooses a suitable challenge from the current profile.
Does beating the CPU change the puzzle rating?
No. Games and exercises are measured separately. Game results and analysis help coaches observe transfer, while puzzle ratings continue to describe focused exercise performance.
What happens if Stockfish or narration is unavailable?
The app fails safely. An unverifiable deviation is not used for rating, and narration can be retried or skipped. The student must always be able to continue without waiting for a teaching animation or voice.
Choose the next useful activity
Use focused problems when a specific decision skill needs practice. Use a complete game when you want to test pacing, resilience and transfer. The learning record will keep those two forms of evidence separate.
CraftChess · Observe · Calculate · Verify · Correct · Remember · Apply
Back to top