Reading the analysis

#Equity

Equity is what a position is worth, in points, to the player on roll. An equity of +0.5 means you expect to come out half a point ahead from here, averaged over every way the game could go from this position.

That is what makes moves comparable. The analysis evaluates the position after each legal play, ranks them, and reports every play by what it costs against the best one. The best play costs nothing by definition, so the number beside a move is always a loss, never a gain.

Cube decisions are priced the same way: doubling, taking and passing each have an equity, and the error is the distance from the best of them. Reading one of those panels is The cube.

#The error symbols

A decision is flagged by what it cost:

Checker errors are drawn as a circle and cube errors as a square. One turn can carry both: a clean play made after failing to double is still a mistake, and it must not read as a clean turn.

#Accuracy (DQI)

The Decision Quality Index scores every decision you made and averages them into a single percentage. A perfect decision scores 1.0, and each worse band costs twice as much as the one above it, so a catastrophe scores well below zero and cannot be averaged away by a run of good moves.

#Error rate (ER)

ER is the average equity you lost per decision, in mEMG: thousandths of a point, equivalent to a money game. An ER of 5.0 means you gave away 0.005 points per decision. Lower is better.

An ER is only comparable against another ER measured at the same depth, by the same engine. A deeper pass finds mistakes a shallower one agreed with.

#XR

XR is an error rate on eXtreme Gammon's scale, so it can be compared with a rating from that program. It differs from ER in what it counts: only decisions where you had a meaningful choice, positions with one reasonable move being excluded. That makes it the stricter measure of decision-making, and the more stable one across matches with a lot of forced play.

#Match review

Under the summary, Dice and play splits the result of the match into what the dice did and what the players did. It reads from the player the summary is showing, so switching the seat to the other player reads it from their side. In a match of more than one game, the picker beside the title reads the bars, the key moments and the written review over one game instead of the whole match; the line and the cube record stay the match's.

At match play every number is a chance to win the match: the match starts at the chance the score gave you and ends at 100% or 0%. At money play it is points.

The bars add up, from the start to the result:

• Opening rolls: what the first roll of each game was worth to whoever got it, against an average opening.
• Dice: each player's luck over every other roll. A roll's luck is what the best play of that roll is worth, minus the average of the best plays of all 36 rolls, both judged one ply deep.
• Errors: what each player gave away, every checker play and cube action at what it cost.
• Not accounted for: what is left. Luck is judged one ply deep while errors are judged at the depth of the analysis, and a resignation ends a game at a price no roll measured, so a little is always left. When it is more than a quarter of what the other bars add up to, the card says the split is approximate.

A match analysed before luck was measured, or brought in from eXtreme Gammon or BGBlitz, has no luck to show: its dice rows read "not measured" and the dice sit in the remainder. Re-analyse the match to measure them. A file from GNU Backgammon carries GNU Backgammon's own luck, measured the same way.

#Key moments

The line under the bars is the same chance, move by move, across every game. Its dots are the key moments, the largest single swings of the match, each worth at least 0.02 of a point in money-game terms. Each one is tagged:

• Error: a player's own mistake.
• Luck: a roll.
• Both: a roll that punished the other player's mistake on the move just before it. The row names the size of that mistake, not of the swing.

Click a moment to go to its decision.

#Cube record

Every cube action and every missed double, for both players, with a count of the cube decisions that were close enough to matter. Cube errors are rare and large, so they are listed one by one rather than turned into a rate.

#The written review

Review this match asks for a written account of the same match, from the player the summary is showing: a headline, a line under each key moment it has something to say about, where most of the error cost went, what went well, and one thing to work on. Clicking a key moment shows its position on the board and leaves the review open, so you can keep reading. If the moment is an error, opening the move's details from there finds the Why? explanation already asked for, where that decision has one.

The written review is part of the Standard and Pro plans, and is written on the server, so it needs a connection. A match analysed only on this device can still be reviewed: it is sent along with the request. On the free plan, or offline, the bars, the line and the lists above are all there; only the written part is missing.

#What went well

The review praises two things, and only when they are measured:

• A hard decision found. A decision is hard when the play a quick reading prefers (the engine judging each play without searching ahead) is not the best play and would have cost at least 0.05 of a point in money-game terms at the depth of the analysis. When you played the best move in one, the review says so and links up to two of them, the biggest traps first.
• Clean cube handling, stated as a count: at least three cube decisions in the match, none of them costing as much as an inaccuracy.

Neither is ever said of a forced move, an opening move or a decision where every play is worth the same. The quick reading runs on the server, on the model that analysed the match, so a match analysed on this device, by a model that is no longer offered, by another program, or at the shallowest depth (where the analysis and the quick reading are the same judgement) gets no "hard decision found" line: nothing there means it was not measured. Where it was measured and nothing qualified, the review says that instead.

#Mistakes you have made before

When you review your own match from the side you played, the review also looks through your other analysed matches. If one of your errors here is the same play in the same position as an error in at least one other match of yours, its line says so, with the number of other matches, and links to that mistake on your statistics page. A repeated mistake then comes first as the one thing to work on. Opening moves and replies match at any score.

Only your own matches are read, each once from its deepest analysis by HedgeHog, and the match under review is never counted among the others. It is a count of what happened, not a verdict on your play, and nobody else viewing the match sees it. A match analysed only on this device can be checked against your saved matches too.

#By phase

Once the written review is in, By phase (collapsed under the cube record) shows where the decisions and the error cost fell: the opening, the middle game, a blitz, a priming battle, a holding or back game, a bear-off with contact, the race, the bear-off, and positions where a gammon is at stake. Each row counts the decisions that were close enough to matter, the errors, and what they cost, and the costs add up to the errors in the bars above. The phase is read from the position each decision faced, from the reviewed player's side. It is an account of this match only: no row says you are weak in a phase, because one match holds too few decisions to say it. That verdict needs many matches, and your statistics page is where it lives.