Supremo Player level

What "Supremo" means here

Every member of the club carries a Level of Play, and the level is not a mood or a slider. It is a measured playing strength, expressed in the number the backgammon world uses for exactly that: PR, the Performance Rating. PR is simple to state. Take every decision a player makes in a match, ask the engine how much each one cost compared with the best play, add it up, divide by the number of decisions, multiply by 500. Lower is better; zero is perfect.

A Supremo's play sits in the band PR 2.0 to 2.8 — well under a hundredth of a point per decision. This is above what the best human players sustain over a tournament, and it is the summit of the club: ten members, the fewest of any level, built to be the strongest opponents anywhere that still play like people.

Why a target, and why not every Supremo member plays the same

Each of the ten Supremo members has a target PR of its own inside that band. The target is what that member is tuned to average over many matches — not a promise about the next one. Two things follow.

First, no two Supremo members are the same player. One sits near 2, another near 3; one is a little sharper on the checkers, another a little looser with the cube. Each was given its own fixed sharpness within the level, so that meeting GG Akiyama is not the same as meeting GG Hoffmann.

Second, a single match will swing above and below the target, for a member exactly as it does for you. A match full of easy decisions produces a clean PR; a match full of close ones punishes everybody. The level holds as an average, and nothing is done to force one match into the band — no correction, no "making up for it" after a good game, no error budget being spent. That was tried in an earlier build and taken out, because people do not blunder on a schedule and neither should they.

How far from the top

The club's strongest player, GG Guru, plays the engine's best move every time: PR 0. A Supremo member gives away well under half a hundredth of a point on every decision that Guru does not. Over a match that gap tells — but it does not decide everything, because the dice still exist.

Put in rating terms: a Supremo starts at 2300 — the same rating as Guru. On the club's rating scale that makes the expected result of a Supremo against Guru an even match at any length, 50 % either way, with the small PR gap and the dice deciding the rest. Against everybody else in the club, a Supremo is the favourite, and the longer the match, the heavier a favourite.

How a Supremo member is tuned

The engine underneath every member is the same one: GNU Backgammon, which reads every position at full strength. A Supremo member does not get a weaker engine. What changes is how the member looks at what the engine sees: through a little noise, so that when two plays are close, the wrong one can look right. How much noise, and how often the member is sharp rather than careless, is set for the level and then adjusted per member.

The settings were not guessed. They were fitted to how real people at this strength actually err — around 200,000 decisions by approximately 100 real players, from world class down to PR 15.9 — and then confirmed in engine-graded test matches before the level shipped, played by hand after that, and re-measured every time the cast changes — by the time the doors open, every level will have thousands of graded games behind it. When a Supremo member plays like one, that is because the measurement said so, not because somebody thought it looked about right.

Where they go wrong — and where they don't

A Supremo's remaining imperfection is deliberate: without it, they would be Guru. It is a very small amount of human-shaped noise in positions that are genuinely close, so that a Supremo still, once in a long while, chooses the second-best play — and so that two Supremos are not the same player. Everything clear is clean. There are no gifts, no drift, no lapses of attention.

The cube at Supremo level

Checker play and cube handling are tuned separately, because in people they come apart. A Supremo's cube is, in Jay's playtest word, airtight: doubles exactly on time, takes and passes correct at every score, redoubles that hurt. Their cube decisions are fitted to the top of the human data. If you take a cube from a Supremo and it was wrong, the review will show you precisely how wrong.

Is this your level?

Almost nobody plays at this level, and that is the point of it. Play a Supremo to see what the top looks like from across the table; play the members a level or two below to improve. If your PR is under 3 over a run of matches, welcome — the ten below are waiting.

Your own level is not a label you pick. When you start, the badge beside your name says provisional. As your career goes on it fills in with your average Level of Play, measured the same way the members are: play in the Supremo range and you carry a Supremo badge; play better, and it moves. Try one level below yours for pleasure, and one above it to learn.

Beat them, and they are yours

In Career Mode, beat a member in a tournament or a league and they are unlocked for good — on your account, across every career you ever start. From then on you can sit down with them for a friendly whenever you like, or just say hello. The ten below are the summit.