Commander · Playgroup Experiments

Can one rule change fix turn order?

In 257,261 four-player Commander games on this site the first seat won 29.23% of the time and the last seat 21.5%. This experiment tests one rule against that: First player skips their first draw. Last player draws one extra card.

Playgroup users can opt-in to play with these new rules and collect data for the experiment. The first read is at 3,000 games.

Updated 10 Sep 2026, 06:35 UTC · refreshed hourly
14
of 3,000 games collected
Progress to the first read

0.5% of 3,000 qualifying games

Read point
3,000

games, fixed before collection started. A second read at 8,000 is an option, never a promise.

Calendar stop
31 Dec 2026

Collection stops on this date or at 8,000 games, whichever comes first.

The rule

First player skips their first draw. Last player draws one extra card.

The same rule two-player Magic already uses, applied to the whole table, to find out whether it evens out the seats.

The baseline it is up against is measured on this site: the first seat wins 29.23% of four-player games and the last seat 21.5%, a gap of 7.73 points. Seats two and three are untouched on purpose: they are the control inside every game.

How to join

Opt a four-player table in

  • Playgroup Live: switch on Playgroup Experiments when you create the lobby, or from the table rules while it is waiting. The board will skip player 1's draw and recommend two draws for the fourth player.
  • Web life tracker: switch it on in the setup screen at four players. The first and last seats are asked to acknowledge their draw (or lack thereof) on their first turn.
  • Only for four-player games.
  • The starting player is chosen at random.

Every finished game counts toward the Lab Partner achievement: the Lab Partner card sleeve at 5 games, and a free month of the Premium Supporter Pack at 20.

They are a small thank-you from the two of us for helping answer a question we think the format will benefit from.

Why this shape

Priced in cards, sized to be readable

Somebody wins every game, so the four seats' win rates add up to 100% and no fix can be all bonuses. The first seat is the one furthest from a fair 25%: 4.23 points over, against the last seat's 3.5 under. The only way to bring the first seat down is to take something away from it.

Mulligans give the measuring stick. The last seat wins 21.53% keeping seven, 22.37% after one free mulligan and 19.07% after two, so one card of hand quality is worth about 3 points of win rate, and the 7.73 point gap is about 2.6 cards wide.

The first idea was scry 1 for the third seat and scry 2 for the last. Scry improves a draw rather than adding a card, about a fifth of one, and moves the gap so little that reading it would take about 115,300 enrolled games. A card off the first seat and a card onto the last moves it about 5.54 points and reads in about 3,000.

The third seat gets nothing, and if the central prediction holds, the last seat overtakes it. That is deliberate: one change at a time is the only way to say which change did what, and the third seat's shortfall is about half a card, which is scry-sized. It is the input to the next experiment.

The long version, with the mulligan table and the drift ladder: Playgroup Experiments: testing a fix for going first

Last seat, by mulligans taken

Kept 7

21.53%

One

22.37%

Two

19.07%

The first London mulligan is free. The second costs a card: about 3 points.

Central prediction

Seat Today Predicted
1 29.23% 26.54%
2 25.69% 25.61%
3 23.58% 23.51%
4 21.5% 24.35%
First minus last 7.73 pts 2.19 pts

A prediction written down before the first game, not a result.

How it is judged

One comparison, fixed in advance

The trial is judged on a single number: the first seat's win rate minus the last seat's, over enrolled four-player single-winner games. The pre-registered criterion is met if the 95% confidence interval on that gap excludes the baseline 7.73 points and the gap itself is at most 4.5 points, which is a reduction of at least 40%. Both halves were written down before the first game.

Per-seat rates are shown once 300 games are in, always with their confidence interval; the headline gap from 1,000 games. Neither is a verdict. The verdict is the frozen block that appears at the read point and never moves again.

Win rate by seat

The four seats, with their uncertainty

Per-seat rates appear at 300 games. At that point each seat is still known only to about ±4.9 points, so they are always drawn with their interval. Baseline row until then:

Seat 1
29.23%
Seat 2
25.69%
Seat 3
23.58%
Seat 4
21.5%
The gap

First seat minus last seat

The headline gap appears at 1,000 games, still with its interval and the two reference lines. The baseline gap it is up against is 7.73 points.

Placebo seats
28.57%

Seats two and three combined. Untouched by the rule, so they should stay near their baseline of 49.27%. A drift beyond two points means the measurement itself needs a look before anything else is read.

Tracker compliance
not yet

Web tracker games where both affected seats acknowledged their draw (0 of 0). Playgroup Live games (14 of 14) are enforced by the board.

Same table, many games
not yet

The share of games from the single most active four-player group. Reported so the read can be repeated without any group over 3% of the games; never a cap on anyone's play.

Playgroup Experiments FAQ

What is the rule being tested?
In a four-player Commander pod, the player who goes first does not draw on their first turn, and the player who goes last draws one extra card on their first turn. Seats two and three play exactly as normal. It is the two-player first-draw rule (CR 103.8a) applied to a table, plus a matching card for the last seat.
Why not just read the numbers as they come in?
Because a result read whenever it looks good is not a result. The trial was pre-registered before the first game: the n it is read at, the single comparison it is judged on and the threshold were all fixed in advance. Everyone can watch the numbers move, but the read happens once, at the read point. Below the disclosure thresholds a seat's confidence interval is wider than the whole effect, so a bare number would only mislead.
How do I take part?
On Playgroup Live, turn on Playgroup Experiments when you create a four-player lobby, or from the lobby's table rules while it is waiting. On the web life tracker, switch it on in the setup screen at four players. The starting player is rolled by the server and nobody can pick a seat; Live enforces the draws, the tracker asks the two affected seats to acknowledge them.
What do contributors get?
Every finished qualifying game counts toward the Lab Partner achievement. The Lab Partner card sleeve unlocks at five qualifying games for everyone, and players without a supporter plan get a free Premium month at twenty. Nothing is tied to winning or to which seat you sit in; that would bias the very thing being measured.
Do experiment games count for ratings?
Yes. Games count normally toward ELO and stats and are badged as experiment games. If they counted for nothing, people would stop playing them seriously and the data would be worthless.
What counts as a qualifying game?
Exactly four seats held by four distinct registered accounts, a server-rolled starting player, a game long enough to be a real game, a single winner, and a first-party client. Guest seats, roster players, draws and games where the table was not fully enrolled are not in the dataset.
The question this answers

Does going first matter in Commander?

Read the turn order data

The article that measured the seat advantage across hundreds of thousands of tracked games, and why a one-card tax on the first player was never going to be enough on its own.