Game Analysis, Engines & DatabasesGuide

Chess Game Analysis Record: A Reproducible Verification Template

Use a structured analysis record to preserve game identity, human reconstruction, critical moments, database evidence, engine settings, tablebase results, root causes and training outputs. Make analytical claims reproducible instead of tool-dependent.

All Chess Guides

A Reproducible Record Should Be Proportional to the Claim

You do not need a laboratory notebook for every casual blitz game. But when an analysis supports a published guide, repertoire change, historical claim or exact theoretical conclusion, the record should preserve enough context to reproduce the important evidence.

Compact Game Analysis Card

For ordinary improvement work, a compact version is often enough:

Game context

game/source and time control;
result and termination;
relevant clock context.

Human reconstruction

where memory/theory ended;
candidates and expected replies;
believed evaluation and plan;
uncertainty.

Critical moment

move number + FEN;
why the position is critical;
what changed after the decision.

Verification

database evidence if relevant;
engine/version/search context if relevant;
tablebase result if covered.

Diagnosis

consequence;
first broken decision stage;
reusable lesson;
training action.

Evidence Strength Should Match Language Strength

Use strong wording only when the verification is strong. “Forced,” “novelty,” “only move,” “tablebase win,” and precise statistical claims all carry evidence requirements. When the search is incomplete, qualify the language instead of hiding uncertainty.

Reproducibility Is More Than Engine Settings

A reproducible claim may also require the exact root position, evaluation perspective, database filters, corpus date, tablebase assumptions or historical source. Record whatever another analyst would need to understand why you reached the conclusion.

The Record Is a Tool, Not the Goal

Do not let documentation overwhelm training. Use the full template for research-heavy or publishable work and a shorter card for routine games. The important habit is preserving the evidence that explains your decision and the evidence that supports your later correction.

Reference Framework

Analysis Record

Game Analysis

Game Identity

raw PGN path/source:
game ID:
players:
event:
date:
round:
result:
termination:
time control:
clock data source:

Human Reconstruction

theory-memory exit:
surprises:
candidates considered:
expected replies:
believed evaluation:
plans:
uncertainty:
clock spent:
tactical ideas seen/missed:

Critical Moment

move number:
FEN:
side to move:
trigger:
irreversible?;
practical context:

Human Post-Game Analysis

current evaluation:
candidate A:
candidate B:
candidate C:
best defense expected:
endpoint:
explanation:

Database Research

database/source:
snapshot/date:
query position/FEN:
filters:
sample size:
frequency:
W/D/L:
recent strong games:
transpositions:
caveats:

Engine Audit

engine:
version:
GUI/service:
analysis date:
evaluation perspective:
MultiPV:
search limit:
threads/hash if relevant:
NNUE/network if relevant:
Syzygy config:
eval:
WDL:
PVs:
stability note:

Tablebase

covered?;
source:
FEN:
side to move:
WDL:
DTZ/metric:
rule assumptions:
move changed WDL?;

Consequence

negligible / small / clear:
lost advantage?;
W→D?;
D→L?;
W→L?;

Root Cause

Choose specific cause:

tactical oversight;
candidate generation;
calculation;
visualization/state;
endpoint evaluation;
positional evaluation;
strategic plan;
opening knowledge;
endgame knowledge;
time management;
practical risk;
execution/interface;
rule/procedure.

Reusable Output

game-specific correction:
general lesson:
repertoire update:
training action:
model position:
endgame position:
clock/practical lesson:

Raw Record Rule

Never overwrite the only raw game record.

Use:

raw source;
versioned working copy;
export/snapshot when tool formats are mutable.

Clock Evidence

Exact clock data may come from:

authoritative PGN clock comments;
official broadcast/event data;
reliable platform record.

Do not invent exact values from:

memory;
edited video;
delayed commentary.

Engine Reproducibility

When precision matters, record:

exact FEN;
side to move;
engine/version;
search limit;
MultiPV;
relevant options;
tablebase path/config;
perspective;
date.

Do not assume:

depth 30 in one version/hardware equals depth 30 elsewhere.

Search Stability Note

Classify:

stable enough for claim;
still resolving;
engine disagreement;
exact tablebase result.

Explain why.

MultiPV Protocol

Before engine:

write human candidates.

Then MultiPV:

compare same root;
sufficient resources;
number of lines justified.

Ask:

missing human candidate?;
refuted candidate?;
same endpoint?;
tiny/noisy difference?;

Database Research Record

Always store:

database/corpus;
snapshot/date;
position query;
filters;
sample size.

For a claim:

move X scores 58%

also state:

among which games?;
what ratings?;
what period?;
which time controls?;
how many games?;

Novelty Verification

Before calling a move a novelty:

exact-position search;
transposition search;
reference/master corpus;
broader corpus;
date/strong-game search;
opening literature/source where relevant.

If incomplete:

use apparent novelty / not found in searched corpus.

Model Game Record

theme:
structure:
why selected:
reliable score?;
annotation source:
historical context:
engine corrections:
model positions extracted.

Tablebase Record

Use the exact endgame/tablebase semantics described in the endgame guides.

Always record:

FEN;
side to move;
rule assumptions;
WDL;
DTZ/metric if used;
source/date.

Then separately diagnose:

knowledge;
calculation;
time;
technique.

Annotation Standard

Prefer:

23...Re8? abandons control of the e-file and allows White's rook to penetrate on e7. The key error was not the numerical swing but failing to identify the entry square.

Avoid:

23...Re8? +2.73

Variation Standard

Include a variation when it demonstrates:

refutation;
best defense;
alternative plan;
exact transition.

Stop when the lesson is established.

Engine vs Database Conflict

If database popularity favors one move and engine another:

do not force reconciliation;
investigate era, player pool, plans, sample, move order and search.

Historical Analysis

Separate:

what contemporary annotations claimed;
what current engine says;
what later theory established.

Do not invent historical player thought.

Live Assistance Boundary

Do not embed advice that normalizes live engine/database use during play.

Check current event or platform rules separately before using analysis tools during live play.

Final Verification Checklist

[ ] Raw game preserved?
[ ] Human thought captured before engine where feasible?
[ ] Critical moment marked for a reason?
[ ] Consequence separated from cause?
[ ] Engine/GUI/database/tablebase distinguished?
[ ] Eval perspective known?
[ ] WDL described as engine-model output?
[ ] PV not called forced without proof?
[ ] Search stability recorded?
[ ] MultiPV cost acknowledged?
[ ] Database filters/sample recorded?
[ ] Frequency separated from quality?
[ ] Novelty claim verified/qualified?
[ ] Tablebase exactness used only in coverage?
[ ] Annotation explains why?
[ ] Provenance/version retained?
[ ] Analysis ends in reusable output?

Key idea

Reproducibility is the difference between “the computer said this once” and a durable analytical claim.