Glossary

The terms below recur throughout the rest of the documentation — manual, user guide, FAQ, CLI. This page gathers them in one place rather than redefining them in every section.

Position

A backgammon decision point: board, cube, dice, score and match flags. A position imported twice counts only once — it is its identity, computed automatically, that recognises it, never the file it came from.

Individually imported position

A position entered into the database on its own — written from the board, or read from a position file — rather than brought in by a match. This property is durable: a position imported this way stays so even if a match containing it is imported afterwards.

Flagged position

A position the user flagged as worth studying in the software the match came from — today only eXtreme Gammon records this mark, move by move. A flagged cube decision marks both positions it produces: the double and the take/pass.

Analysis

What an engine (eXtreme Gammon, GNUbg, BGBlitz or the embedded gammonNet evaluator) concluded about a position, stored and read back later. A position carries one analysis per engine; importing an analysis never overwrites another engine’s, nor an analysis already present (see Import: what is written, what never is).

Evaluation

The result of a computation the embedded evaluator performs on the position currently on the board, whether stored or not. An evaluation is valid only for as long as the position does not change; if stored, it becomes an Analysis.

Referential

The scale a figure about a position is expressed in. Money answers “how many points is this worth”, a gammon always being worth two points. Match answers “how much of the match is this worth”, on a normalised scale where a gammon is worth whatever the score makes it — almost everything at 2-away/4-away, nothing to a leader at 1-away. The referential is a property of the position, never a display preference: it is the away score that sets it.

Normalised equity

Match equity, expressed on the 2 × MWC − 1 scale, where ±1 represents the current cube won or lost. It is the scale blunderDB displays and stores at a match score, never the raw win probabilities.

EMG

The acronym under which eXtreme Gammon and GNUbg name the normalised equity above, and under which blunderDB’s statistics count errors: “an error of 0.100 EMG” is an error of 0.100 on this scale. Both terms name the same quantity.

Away score

How many points a player still needs to win the match, from that player’s own side. Two values are sentinels rather than counts: -1 means there is no match — the referential is money — and 0/1 distinguish the game after Crawford (0) from the Crawford game itself (1), for the same distance to victory.

Regime

What kind of answer the panel gives about a race, always stated on screen rather than left for the reader to guess:

• exact — read from a two-sided bearoff database, never estimated;

• evaluated — computed by the embedded evaluator, which plays the trajectory out;

• estimated — a calibrated summary of a trajectory, valid for a win probability, never for a cube verdict.

Bearoff database

A table of race answers, two-sided (the cube verdict) or one-sided (the EPC). blunderDB generates part of it itself on first launch (up to 6 checkers per player); wider domains are generated from the Bearoff tab or the CLI. A generated database is said to be verified when its fingerprint exactly matches the one GNUbg produces.

EPC (Effective Pip Count)

A classic, one-sided estimate of the number of rolls needed to bear off all of a player’s checkers — corrected for wastage against the plain pip count.

Position fact

A quantity that belongs to the board itself, independent of any choice a player might make: pip count, EPC, win/gammon/backgammon chances, cubeless equity. True before the roll.

Pre-roll baseline

The vector of win/gammon/backgammon chances and cubeless equity, shown above the list of candidate moves once the dice are rolled. The gap between this line and a play measures the luck of the roll, never the merit of the move played.

Decision

The answer to the question the board is asking: the ranking of candidate moves when dice are rolled, or the three cube options and their verdict otherwise. A position never asks both questions at once.

Verdict

Which option of a cube decision is the right one: no double, double/take, double/pass or too good (to double). Two special cases are never left blank: no decision (the regime is not entitled to one) and refused (the engine declined the position, typically a score beyond its table’s reach).

Too good

The verdict saying the position is so favourable that doubling would prompt the opponent to pass, whereas playing on without doubling pays more — a cube best kept.

Luck

The dice roll’s contribution to a move’s equity, read from the source file at import (eXtreme Gammon, GNUbg) and never recomputed by blunderDB. Formats that do not carry it (BGF, Jellyfish .mat) never provide it, and the import is not retroactive.

Blunder

A move or cube decision whose error (the equity gap with the best choice) exceeds a threshold; the bl (or blunders) command loads a search’s worst errors directly.

Millipoint (mpt)

A thousandth of an equity point: an equity error of 0.100 is worth 100 mpt. It is the unit a position’s own figures are read in — the blunder threshold, the e> and E> search filters, the error-magnitude histogram, the average luck per roll. PR is not read on this scale: its factor is 500, not 1000, so a PR of 5.0 is worth 10 mpt of average error per decision.

PR (Performance Rating)

A player’s average error cost, in thousandths of an equity point per counted decision, checker play and cube combined. See Annex: Statistics model — XG / gnuBG / blunderDB alignment for the exact counting rules (parity with eXtreme Gammon and GNUbg).

Snowie Error Rate

The same numerator as the PR (the sum of errors), but divided by the number of moves played rather than the number of counted decisions — more stable from one tool to another. See Annex: Statistics model — XG / gnuBG / blunderDB alignment.

MWC cost

A cube error expressed in percentage points of Match Winning Chance, rather than in equity — the natural quantity to compare the same player across different match scores.

Collection

A named, ordered set of positions the user assembles by hand, drag-and-drop included. Collection membership is an explicit gesture, unlike an individually imported position (see Collections Panel).

Anki deck

A set of positions turned into cards reviewed by spaced repetition (the FSRS algorithm). See Anki Panel.

Review card

A position from an Anki deck, presented as a question: the board is the question, the stored analysis is the answer, hidden until the gesture that reveals it.

Comment

Free text attached to a position, whether typed by the user or lifted verbatim from an imported file — the two are indistinguishable. A #keyword inside a comment is a tag, searchable like any other text.

Player

A name exactly as it appears in an imported match. blunderDB does not guess the person behind the name: the same human signing under two spellings is two distinct players, until an explicit merge (“Merge players”).

Watermark

The signed statement, applied by whoever exports a database, of what it is and where it comes from. Unforgeable but not unremovable, it never records anything on the recipient’s side. See Handing out a database: origin and password.

Issuer identity

The key that signs one person’s watermarks — created on its own at the first watermark applied, and movable from one machine to another as a single file.

Protected file (.dbx)

A database exported and encrypted with a password, opened once and then ordinary. It protects the file in transit, not its content: whoever receives the password can open it.

Tenant

In server mode, the owner of a set of positions, matches and collections. Nothing one tenant stores is ever visible to another. On the desktop there is only one, implicit: the person whose file it is. See Headless mode (server).