Skip to content
Game Analysis, Engines & DatabasesGuide

60 Comparative Chess Analysis Patterns

Study paired analysis patterns that contrast good and bad human reconstruction, engine use, database research, tablebase verification, annotation and lesson extraction. Train the difference that changes the conclusion rather than memorizing labels.

Train Differences, Not Labels

Comparative patterns are useful because analysis errors often look similar on the surface. Two moves can both be “blunders” while one came from missing a candidate and the other from mis-evaluating an endpoint. Two database statistics can both show 60% while one comes from a robust sample and the other from a handful of games.

For each pair, ask four questions:

  1. What changed?
  2. Why does the analytical conclusion change?
  3. Which evidence is decisive?
  4. What nearby counterexample would reverse the lesson?

Build Cross-Tool Comparisons

The strongest drills combine evidence types. Example: first reconstruct a position without tools, then inspect database practice, then compare engine candidates, then—if covered—probe the tablebase. The training target is not agreement between tools; it is knowing what each layer contributes.

Convert Your Own Games Into Pairs

When you make a recurring error, find or create a contrasting position where the superficial feature is the same but the correct decision differs. A good pair forces you to identify the deciding condition rather than memorize “always do X.”

Pattern Library Categories

Useful families include:

  • consequence vs root cause;
  • human-first vs engine-first workflows;
  • CP/WDL/perspective interpretation;
  • PV and search stability;
  • database sample/filter problems;
  • exact-position vs name-based opening research;
  • tablebase consequence vs human diagnosis;
  • annotation clarity vs variation dumping;
  • game-specific correction vs reusable lesson.

A mature analysis library contains counterexamples as well as success cases.

Reference Framework

1. Same Blunder Label, Tactical Cause

Player never sees back-rank tactic.

Training:

  • recognition/tactical scan.

2. Same Blunder Label, Candidate Cause

Player calculates two moves accurately. Winning third move never considered.

Training:

  • candidate generation.

3. Same Blunder Label, Calculation Cause

Correct candidate found. Opponent zwischenzug missed.

Training:

  • best-defense calculation.

4. Same Blunder Label, Visualization Cause

Captured piece remains in mental board.

Training:

  • board updating.

5. Same Blunder Label, Endpoint Cause

Line calculated correctly. Final ending falsely judged winning.

Training:

  • endpoint evaluation/endgame.

6. Same Blunder Label, Clock Cause

Player knew position required calculation. Played instantly.

Training:

  • critical-moment/time allocation.

7. Human First

Player records candidates before engine.

Engine reveals missing defense.

Lesson: cause remains visible.

8. Engine First

Player sees ?? and best move immediately. Later cannot remember original candidates.

Lesson: learning evidence lost.

9. Engine-First Triage Used Well

10,000 games scanned for eval swings. Selected games then reconstructed manually.

10. +1.00 Without Material Edge

Stockfish favors White though material equal.

Lesson: normalized CP ≠ pawn count.

11. WDL Misread as Human Probability

Engine WDL shows strong self-play win component. Beginner assumes 70% chance personally.

Lesson: model output ≠ human forecast.

12. Perspective Error

GUI shows White perspective. Analyst reads negative as bad for side to move.

Lesson: record perspective.

13. Mate vs CP

Search changes from +8 to mate score.

Lesson: semantic category changed.

14. PV Changes with Depth

Early PV:

  • tactical line A.

Later:

  • quiet defense B.

Lesson: PV is current preferred line.

15. PV Is Not Forced

Opponent has several near-equal replies absent from PV.

16. MultiPV Finds Human Candidate

Three human candidates all appear. One missing engine line is only tiny better.

Lesson: human search was broad enough.

17. MultiPV Finds Missing Candidate

Quiet fourth move dominates.

Lesson: candidate-generation error.

18. Too Many MultiPV Lines

Ten PVs at same time budget. Top-line quality/search confidence degrades.

Lesson: more output costs resources.

19. Depth Worship

Depth rises. Best move/eval still oscillate.

Lesson: depth alone ≠ settled analysis.

20. Stable at Lower Depth

Best move/eval/PV unchanged under more resources.

Lesson: stability is contextual evidence.

21. Engine Disagreement

Two engines disagree. Analyst averages +0.8 and -0.2 to +0.3.

Lesson: investigate; do not average.

22. Tiny Eval Obsession

+0.12 vs +0.18. Same practical/strategic lesson.

Lesson: do not manufacture significance.

23. Engine Optimal vs Human Method

Engine finds obscure only-move sequence. Simpler line keeps decisive advantage.

Lesson: label human-friendly alternative honestly.

24. Database Popularity Trap

Move A played 70%. Move B engine-preferred and theoretically current.

Lesson: frequency ≠ quality.

25. Win-Rate Trap

Rare gambit scores 75% in 12 blitz games.

Lesson: sample/time/player pool matter.

26. Rating Filter

Move performs differently below 1800 and in elite classical games.

Lesson: population matters.

27. Recency Filter

Historical main line dominates all-time frequency. Recent elite practice shifted.

Lesson: frequency and current theory differ.

28. Duplicate Data

Same game imported three times. Move frequency inflated.

29. Player-Name Split

Garry Kasparov, Kasparov, G. counted separately.

Lesson: normalization affects statistics.

30. Engine Dataset vs Game Dataset

One corpus has Stockfish positions. Another has played results.

Lesson: different evidence.

31. Move-Order Search Misses Games

Sequence A not found. Sequence B reaches identical FEN.

Lesson: position search recovers transpositions.

32. Rare ≠ Novel

Move absent from master DB. Found in older broad corpus.

33. Apparent Novelty

No hit in searched databases. Claim remains qualified.

34. Model Game by Fame

World champion game selected. Theme obscured by tactics.

35. Better Model Game

Less famous game shows structure/plan cleanly.

36. Classical Annotation vs Modern Engine

Old note says move forced. Engine finds resource.

Lesson: historical annotation and modern truth can coexist as separate evidence.

37. Similar Structure, Different Opening

Same IQP position from two move orders.

Lesson: thematic research crosses opening names.

38. Tablebase Proves W→D

Played move loses exact win.

39. Tablebase Does Not Diagnose Cause

Player:

  • knew Lucena;
  • rushed and missed move.

Cause:

  • time/technique, not knowledge.

40. Tablebase Knowledge Gap

Player does not know Philidor at all.

Same consequence class, different training.

41. Raw PGN Overwritten

Analyst edits result/moves/comments in only copy.

Lesson: source evidence lost.

42. Raw + Annotated Copy

Original retained. Working copy versioned.

43. ? +2.4

Numeric label gives no transferable reason.

44. Explanatory Annotation

Comment identifies entry square, defender and missed candidate.

Lesson: analysis becomes learning.

45. Variation Dump

Eight PVs pasted. No statement of what matters.

46. Concise Variation

One line shows the tactical refutation. Comment explains mechanism.

47. Missing Provenance

Stockfish says +1.6. No version, FEN, perspective or search context.

48. Reproducible Engine Audit

Exact FEN, engine/version, MultiPV, limit, date and stability recorded.

49. Frozen Tool Fact

Guide says Stockfish supports exactly seven-piece TB forever.

Lesson: date-stamp implementation facts.

50. Good Game-Specific Correction

23...Re8 fails to Re7 penetration.

51. Good Reusable Lesson

Before conceding an open file, identify the opponent's entry square.

52. Repertoire Update

Opening error becomes a versioned FEN-based repertoire branch.

53. Training Task from Candidate Error

Ten positions requiring quiet defensive candidates.

54. Recurring Pattern Across Games

Five games show same missed counter-check.

Lesson: recurrence increases training priority.

55. Good Result, Bad Analysis

Game won. Engine reveals opponent missed forced win. Root cause still analyzed.

56. Bad Result, Good Decision

Sound move loses later. Do not label earlier choice as error by outcome.

57. Three Reusable Lessons—Supported

Game contains:

  • opening gap;
  • visualization error;
  • clock pattern.

58. Three Lessons—Manufactured

Analyst forces three general rules from one accidental tactic.

Lesson: only generalize what evidence supports.

59. Post-Game Tool Use

Engine/database used after game with provenance.

60. Live Assistance Boundary

Same tools during ongoing competitive game.

Lesson: a separate rules/fair-play question.

Key idea

The strongest analysis explains both the chess truth and the human route that failed to reach it.