Use This Page to Separate Nearby Concepts
Many bad analytical claims begin with two useful concepts being treated as synonyms. This reference is designed as a comparison tool: identify the pair, state the difference, then ask which side of the distinction your current evidence actually supports.
Evidence Type vs Interpretation
A database frequency is evidence of historical practice in a selected corpus. An engine score is model/search evidence. A tablebase result is exact within coverage. A human annotation is an explanation or interpretation. None automatically converts into another.
Search Quantity vs Search Confidence
Depth, nodes and time describe aspects of search work. Stability describes how the conclusion behaves as search continues. They are related but not interchangeable. A huge node count does not guarantee that every relevant interpretation has stabilized.
Line vs Proof
A PV shows the current preferred line; a forced line requires that relevant alternatives fail. MultiPV shows several root candidates; it does not prove the omitted candidates are irrelevant unless the search and analytical context justify that conclusion.
Statistics vs Chess Quality
Move frequency, score percentage and average rating describe the selected games. They do not directly measure objective move quality. Conversely, the engine's favorite move may be rare because it is new, difficult, impractical, outside the corpus or only recently appreciated.
Human-First vs Anti-Engine
Human-first analysis preserves learning evidence before external revelation. Anti-engine analysis rejects a powerful verification tool. Those are opposite ideas. The strongest workflow combines independent thinking with rigorous tool audit.
Anti-Dogma Test
When you hear an analysis slogan—“depth 30 is enough,” “the database says this is main line,” “+1 means a pawn,” “the PV is forced”—ask:
- What exactly does the tool output mean?
- What assumptions or configuration produced it?
- What claim is being inferred from it?
- Does that inference require additional evidence?
Reference Framework
Review vs Analysis vs Annotation
Review
- look back at game.
Analysis
- investigate positions, alternatives and causes.
Annotation
- record explanations/evidence durably.
Human-First vs Anti-Engine
Human-first preserves:
- candidates;
- evaluation;
- uncertainty;
- plans.
Engine audit then tests them.
Human-first is not a rejection of engines.
Engine-First Triage vs Engine-First Learning
Triage:
- locate likely critical moments quickly.
Learning:
- still reconstruct human cause.
Error Consequence vs Root Cause
Consequence:
- what changed objectively.
Root cause:
- why the decision process failed.
Blunder can describe consequence. It cannot diagnose cause.
Evaluation Swing vs Cause
A large swing can identify:
- where to investigate.
It cannot tell whether cause was:
- tactic;
- candidate generation;
- calculation;
- visualization;
- evaluation;
- plan;
- knowledge;
- time;
- execution.
Engine vs GUI
Engine:
- searches/evaluates.
GUI:
- hosts, configures and displays.
Engine vs Cloud Service
Cloud service:
- infrastructure/application wrapping one or more analytical engines.
Engine vs Database
Engine:
- computes.
Database:
- stores/retrieves corpus.
Engine vs Tablebase
Engine:
- heuristic search outside exact coverage.
Tablebase:
- exact covered-state information under defined assumptions.
UCI vs Engine
UCI is a communication protocol.
It is not:
- a chess engine;
- a move notation system for humans;
- a database.
CP vs Material
Centipawn-style score:
- normalized engine evaluation scale.
It is not:
- literal count of pawn material.
Stockfish +1.00 vs Human Probability
Current Stockfish normalization:
- tied to engine self-play WDL calibration.
It is not:
this human has a 50% chance to win.
Evaluation Perspective
A sign needs a convention.
Possible displays:
- White perspective;
- side-to-move;
- GUI-defined normalization.
Never publish a bare number without knowing the convention.
Mate vs CP
Mate:
- forced result represented by mate score/search.
CP:
- heuristic numeric evaluation.
Mate is not +infinity pawns.
WDL vs Human Win Rate
Engine WDL:
- model-based output.
Human win rate:
- empirical outcome for a population/context.
Do not conflate.
PV vs Forced Line
PV:
- current preferred continuation.
Forced line:
- chess position leaves no meaningful alternatives / exact forcing sequence.
PV can contain non-forced choices.
PV Stability vs Truth
Stable PV increases confidence for the analytical purpose. It is still version/search-context output unless exact theory/tablebase applies.
MultiPV vs More Truth
MultiPV reveals alternatives. It also consumes search resources.
Use the number of lines required by the question.
Depth vs Confidence
Depth is useful metadata.
Confidence also needs:
- stability;
- time/nodes;
- search setup;
- MultiPV;
- tablebase;
- engine/version.
Seldepth vs Depth
They are distinct search metadata concepts.
Do not compare them as if one were simply better depth.
Nodes vs Strength
More nodes:
- more search work under that engine/configuration.
Cross-engine/version comparisons are not automatically meaningful.
NPS vs Accuracy
High NPS can reflect:
- hardware;
- code;
- settings.
It does not prove stronger analysis.
Tablebase Hits vs Root Exactness
A search can probe tablebases in branches. That does not make the root exact if root is outside coverage.
Search Disagreement vs Average the Scores
When engines disagree:
- inspect settings;
- perspective;
- versions;
- resources;
- stability;
- critical branches.
Do not average two evals mechanically.
Tiny Eval Difference vs Meaningful Difference
A few centipawns can be:
- unstable;
- model-specific;
- irrelevant to the lesson.
Do not invent pedagogical distinctions from noise.
Engine Optimal vs Human Friendly
Engine-optimal:
- maximizes engine objective.
Human-friendly:
- preserves sufficient soundness while producing a reproducible method.
Label the distinction.
Master Database vs Broad Database
Master/reference:
- curated/high-level/theoretical focus.
Broad:
- breadth/practical usage.
Use both for different questions.
Personal/Opponent Database vs Population Evidence
Personal/player sample:
- useful for habits/preparation.
It is heavily selected. Do not generalize it to chess as a whole.
Engine-Evaluation Dataset vs Game Dataset
Engine dataset:
- model outputs.
Game dataset:
- empirical played outcomes.
They answer different questions.
Frequency vs Quality
Frequency:
- what people played.
Quality:
- what analysis supports.
Popular ≠ best.
Win Rate vs Evaluation
Win rate:
- selected-corpus outcomes.
Evaluation:
- analytical assessment.
A move can score well and be objectively inferior.
Sample Size vs Certainty
Small sample:
- wider uncertainty.
Do not trust dramatic percentages from a handful of games.
Recent vs Best
Recent strong games:
- contemporary evidence.
They are not automatic proof that an older move is inferior.
Rare vs Novel
Rare:
- low frequency in corpus.
Apparent novelty:
- absent from current search.
True novelty claim:
- needs wider corpus/date/transposition verification.
Move Order vs Position
Move order:
- path.
Position:
- resulting state.
Transpositions make position search essential.
Model Game vs Famous Game
Model game:
- conceptual clarity.
Famous/high-rated game:
- may or may not teach the theme cleanly.
Tablebase Consequence vs Human Cause
Tablebase can establish:
- WDL changed.
It cannot establish:
- player forgot theory;
- player rushed;
- player miscalculated.
Raw PGN vs Annotated PGN
Raw:
- source evidence.
Annotated:
- derived analysis artifact.
Do not overwrite the raw record.
Annotation vs Variation Dump
Good annotation:
- explains mechanism and lesson.
Variation dump:
- lists engine moves without purpose.
NAG vs Explanation
? can summarize. It cannot explain:
- why;
- what was missed;
- what to train.
Tool Fact vs Evergreen Fact
Version-sensitive:
- database size;
- Stockfish option;
- tablebase coverage;
- platform label threshold.
Evergreen:
- provenance must be recorded;
- frequency is empirical;
- tablebase exactness is coverage-bound.
Post-Game Use vs Live Assistance
this analysis section:
- post-game/study/permitted training.
Live-play rules:
- current competition/platform fair-play boundaries.
The central analytical skill is not knowing which tool to open; it is knowing what kind of evidence that tool can produce.