Tablebases are exact endgame data within their covered material, metric and rule assumptions. They are not just stronger engines: they can provide exhaustive result information that an ordinary centipawn evaluation cannot replace.
Tablebase result, tablebase metric and human winning method are three different layers.
What an Endgame Tablebase Is
An endgame tablebase stores/precomputes exhaustive information for positions within a covered low-material class, allowing exact queries such as win/draw/loss under defined assumptions.
Retrograde / Exhaustive Endgame Information
Tablebases are produced through exhaustive/retrograde-style endgame computation rather than heuristic evaluation alone. Their authority comes from complete analysis of the covered state space and metric.
Exact Result vs Engine Evaluation
An exact tablebase result is not an engine centipawn evaluation. Inside covered conditions it gives discrete theoretical information that a heuristic engine score cannot substitute for.
Current Piece-Coverage Limits Must Be Date-Stamped
Piece-coverage limits are time-sensitive facts. Every publication claim must be date-stamped and tied to the actual tablebase/probing system used.
WDL
WDL classifies the position by win/draw/loss information; Syzygy-style probing can refine this into five values that also encode cursed-win/blessed-loss 50-move context.
DTZ
DTZ measures distance to a zeroing move (capture or pawn move) within Syzygy's rule-aware optimization framework, not distance to mate.
DTZ50 / Fifty-Move Context
DTZ50 / 50-move context adjusts the interpretation of winning/losing lines around the move-count rule. Record the halfmove/rule assumption rather than treating raw DTZ as universal.
DTM
DTM means distance to mate under its metric's optimality assumptions. It answers a different question from DTZ and may recommend different moves.
DTC / Other Historical Metrics
DTC and other historical metrics measure different conversion objectives. They may be useful historically or in other tablebase families, but must never be silently equated with DTZ or DTM.
Do Not Conflate Metrics
Do not conflate WDL, DTZ and DTM: result category, distance to zeroing progress, and distance to mate are different data.
Cursed Win
A cursed win is a theoretical win in the no-move-count sense that becomes a draw under the 50-move-rule context represented by Syzygy WDL/DTZ assumptions.
Blessed Loss
A blessed loss is the mirror category: theoretically losing without the rule constraint, but drawable because the opponent cannot complete the required zeroing progress within the 50-move-rule context.
Syzygy WDL Files
Syzygy WDL files store win/draw/loss-class information (including rule-aware subcategories where applicable). They are distinct from the DTZ files.
Syzygy DTZ Files
Syzygy DTZ files store distance-to-zeroing information used to preserve the tablebase result while making rule-aware progress.
Seven-Piece Syzygy Coverage — Date-Stamped Fact
As of 2026-08-26, standard Syzygy/Stockfish probing supports tablebases up to seven pieces; this is a date-stamped implementation fact, not a permanent architectural assumption.
DTZ-Optimal Can Look Unnatural
DTZ-optimal moves can look unnatural because minimizing/preserving distance to a zeroing move is not the same objective as producing the shortest or clearest human winning plan.
Human-Friendly Winning Method
A human-friendly method should preserve the exact WDL result while emphasizing reproducible concepts: cut off king, build bridge, create shelter, centralize, simplify into known theory.
Record FEN
Record the exact FEN when verifying an endgame. One-square changes can alter WDL, DTZ and the legal move set.
Record Side to Move
Record side to move explicitly. Many theoretical endings reverse result when the move changes.
Record W/D/L
Record the result as W/D/L (and refined WDL category if relevant) from the perspective being used.
Record Rule Assumptions
Record rule assumptions: halfmove clock, 50/75-move context, castling/en-passant validity where a probe/tool supports or excludes them, and whether the metric ignores/uses move-count rules.
Verify Mirrored Position
Mirroring a board can preserve geometry but side to move, pawn direction/file and rule-state differences can still change the result. Re-probe the mirrored position.
Verify Named Position Conditions
Verify the defining conditions of named positions rather than trusting resemblance: Lucena, Philidor, Vancura, wrong-bishop fortress and Troitsky-family setups are condition-sensitive.
Identify Only-Move Defenses
Identify only-move defenses explicitly. A tablebase win/draw can depend on a unique move that human heuristic play would miss.
Verify Fortress
Verify fortress claims with exact probing when the material is covered. If outside coverage, use analysis/theory and label uncertainty honestly.
Engine Is Not Proof Beyond Tablebase
An engine is not proof beyond tablebase coverage. A high depth/large evaluation can support analysis but does not provide exhaustive mathematical certainty in an uncovered position.
Tablebase QA Practice
Tablebase QA practice should compare same WDL/different DTZ, cursed wins, one-square flips, named-position conditions, human-friendly vs DTZ-optimal methods, and positions outside coverage.
What Tablebases Prove—and What They Do Not
Within the exact material and rule model covered by a tablebase, the result information is exhaustive rather than heuristic. That makes tablebases fundamentally different from an engine evaluation. But an exact tablebase result still does not automatically give the clearest human explanation or the easiest practical method.
As of 2026, standard Syzygy tooling supports complete tablebases through seven pieces; active research and public services have begun exposing partial eight-piece coverage. Because coverage and services evolve, any statement about the “largest solved endgame” should be date-stamped and tied to the specific system being used.
WDL and DTZ Answer Different Questions
WDL classifies the result under the tablebase's rule treatment. DTZ measures distance to the next zeroing move—typically a pawn move or capture—under its optimization convention. A DTZ-optimal move may look strange to a human because the metric is not “shortest mate” or “easiest win.”
Do not silently substitute DTM for DTZ. If a source provides mate distance, label it separately.
The Fifty-Move Context Matters
Syzygy-style data can distinguish ordinary wins/losses from cursed wins and blessed losses in relation to the 50-move rule. This means “mathematically winning without a move-count restriction” is not always the same as “claimable win under tournament rules from the current halfmove state.” Record the halfmove clock when it is relevant.
A Reproducible Tablebase Query
For every exact claim, record:
- full FEN, including side to move;
- material count;
- tablebase service or files used;
- date of verification;
- WDL category;
- DTZ where relevant;
- move-count assumptions;
- only moves or critical defensive resources;
- a human-readable method.
Re-probe mirrored or “nearly identical” positions. One square or side-to-move change can flip the result.
<!-- ASSET: chess-g-221-chart-01 -->