How the Tournament Predictor Works

The Tournament Predictor takes an entry list — from a club's registration page, a Google Sheet, or a list of US Chess IDs — looks up every player's current rating, and plays the event out round by round. This guide explains what happens in each game, how rounds are paired, what the 1,200 simulated runs add, and where the model is knowingly simple.

One game

The expected score between two players comes from the standard rating curve. White is treated as 35 rating points stronger, roughly the edge large game databases show for the first move. The curve fixes the expected score, but not how it splits between wins and draws, so the predictor assumes a draw rate for evenly matched players that depends on the rating system, and shrinks it as the gap grows:

These draw rates are assumptions, not measurements, and they never change the expected score — only how it is made up. Two 1600s in a Regular event: White wins 40%, draws 30%, loses 30%. A 1700 with White against a 1500: 66% / 27% / 7%.

Pairing the rounds

A Swiss tournament cannot be predicted with a formula, because each round's pairings depend on the previous round's results. So the predictor pairs one round, decides its games, updates the standings, and pairs the next — never repeating an opponent, balancing colours, and handing a bye to the lowest-placed player without one when the field is odd. Requested byes from the entry list are honoured: those players sit out the rounds they asked for and get the half point.

• Swiss — Everyone plays every round, paired against someone on a similar score and never the same opponent twice.
• Accelerated Swiss — The top half carries a virtual point for the first two rounds, so the strong players meet each other sooner and the field separates faster.
• Round robin — Everyone plays everyone once. The number of rounds is settled by the field size, not chosen.
• Double round robin — Everyone plays everyone twice, once with each colour. Twice the rounds and no colour luck at all.

Events already under way

For an Austin Chess tournament that has started, the predictor reads the rounds already played from the club's published results — every game, colour, bye and forfeit — and checks that each player's points add up to the total the club prints. Those rounds are kept exactly as they happened, a game still being played is predicted like any other, and only the rounds still to come are paired and simulated, from the real standings. Real results are shaded in the cross-table, and every simulated run starts from them, so nobody can finish below the score they already have.

Pairings for rounds the club has not posted yet are the predictor's own, and they are the weakest part of the prediction. Replaying past Austin Chess rounds from the real standings, the predictor matched the tournament director's actual pairings in about half the games: directors weigh colours, rating gaps and other preferences differently from one event to the next, and pairing first by colour helped one event as much as it hurt another. So read the opponents shown for a future round as one possibility. Posted pairings are used as soon as they appear.

The predicted cross-table

The cross-table you see first gives every game its single most likely result. It is the "if everything goes to form" version of the event — and even so, it does not simply return the entry list in rating order:

• In a four-round Swiss with evenly spaced ratings, fields of up to 8 players do come back in rating order. In bigger fields it does not, even with the favourite winning every game: an odd-sized score group has to float someone down to play a weaker opponent, and that player collects an easier point than the people above them who had to play each other.
• Byes do the same thing more bluntly. A free point can lift a player past someone over a hundred points stronger.

1,200 simulated runs

One cross-table is a single story. To see the range, the predictor then replays the whole event 1,200 times, deciding every game at random at the odds above. From those runs it reports, for each player:

• the chance of finishing first, and of finishing in the top three;
• the average finishing place, and the best and worst score reached;
• the best, average and worst rating change, using the official US Chess formula.

The spread is the useful part. A player whose average place is 4th but who finishes first in one run in five is in a very different position from one who is 4th nearly every time.

What whole events do to ratings

Simulating every game at once shows two effects of the US Chess formula that no single result does:

• The bonus adds rating points to a field. In a round robin, the points won and lost cancel exactly before the bonus. But the bonus only ever adds, so a field of identically rated players ends an event with more rating points in total than it started with — the system's deliberate defence against ratings slowly sinking.
• Wide fields lose points; tight fields gain them. Lower ratings move faster, so when the field is spread out, the players losing rating points move further than those gaining them. In a six-player round robin, a field spread over 50 points gains about 54 points in total; one spread over 900 loses about 13.

What it cannot know

• Ratings lag real strength, especially for improving juniors. A player rated 900 who plays like 1200 will beat the prediction, and the prediction cannot tell who that is.
• Entry lists change. Late entries, withdrawals and section changes up to the first round all move the result.
• Tournament directors use tie-breaks, house rules and their own judgement on pairings; the predictor follows standard Swiss practice.
• Entry lists from club sites rarely say how many rounds a section has, so you set the rounds yourself.

Treat a prediction as the likely range, not the result. To try it on a real event, open the Tournament Predictor.