Who actually has your number? Head-to-head records in competitive pinball

There is a player at my local who is rated well below me and has beaten me something like eleven of the last fifteen times we have been drawn together. Ratings say I should be winning that matchup comfortably. The matchup says otherwise, and the matchup has been right for two years.

This happens constantly, and it is the single best argument for keeping head-to-head records rather than just comparing two numbers. Ratings are an average over everybody. A matchup is a specific thing between two specific people, and specific things have texture: machine preferences, nerves, one person who tilts and one who never does, a game that one of you owns and the other has played four times.

So the site keeps all of it.

The tale of the tape

The head-to-head page is the front door. Pick two players, optionally narrow to one venue, and you get the lifetime record: total games shared, wins each way, ties, and then the part that matters, a machine-by-machine breakdown.

Those comparisons have their own permanent addresses, in the form of /head-to-head/name-a-vs-name-b, so you can bookmark a rivalry or drop it in a group chat before league night without making the other person fill in two dropdowns. Which they will not do.

Under the tape is the game history: every single game the two of you have shared, with the date, the tournament, the round, the machine, and who came out ahead. It is the receipts. When somebody tells you they always beat you on a particular game, this is where you find out whether "always" means seven times or twice.

What the win probability is doing

The page also shows an estimated win probability, and it is worth explaining because a naive version of this number would be useless.

If you simply took the historical record, a 3-0 head-to-head would read as 100 percent, which is obviously nonsense. So the estimate starts from a prior based on the two Match Play ratings, treated as a logistic curve where a 400-point gap is a meaningful edge, and then shrinks the actual record toward that prior. With few shared games, the rating prior dominates. As games accumulate, history takes over.

The result is a number that says "roughly this, based on what we know", which is the only honest thing a matchup estimate can say. Three games is not evidence. Thirty is.

Common opponents, the indirect comparison

Two players who have never met still leave a trail. If you have both played the same twenty people, the comparison is available, it is just indirect.

The head-to-head page lists common opponents: everyone you have each faced at least three times, with both records side by side. You beat that player 8-2; they go 3-7 against the same person. That is real information, and it is often more useful than the direct record when the direct record is four games old.

This is also the closest thing to an answer for the perennial question about two players from different venues who never overlap.

Your opponents tab, which is the leaderboard of grudges

Every player profile has an Opponents tab holding the same pairwise records against everyone you have ever shared a game with, sorted so the lopsided ones stand out. Click any row and it opens the full comparison.

The profile also surfaces two panels that people find uncomfortably accurate:

Easiest and hardest opponents. Machine-agnostic pairwise records with a minimum of three shared games, which is where you discover that the person you consider your equal is 14-6 up.

Rivalries, with a nemesis tag. This one is more carefully built than it looks. A rivalry is supposed to mean a matchup that is both frequent and close, so the list prefers opponents you have played a lot with a record near even, over opponents you have played a lot and dominated. Records get grouped into bands by how far they sit from 50/50, and within a band the ones with more games rank higher. A 24-25 regular outranks a 9-9 occasional, because 24-25 across fifty games is a rivalry and 9-9 might be a coin.

The red nemesis tag only appears when the matchup is genuinely real: double digits of shared games and a record close enough to even that neither of you can claim it. If nobody on your list has the tag, you do not have a nemesis. You just have people you have beaten and people who have beaten you.

Machine by machine is where the strategy lives

The overall record is the headline. The per-machine breakdown is the useful part.

You are 9-11 down against someone overall, but 6-1 up on the two electromechanicals in the building and 1-8 down on the modern Sterns. That is not a close matchup, it is two completely different matchups wearing the same name, and in any format where somebody picks the machine, it is worth a great deal to know which is which before the pick comes to you.

That is exactly what the Scout page automates. Give it the venue, yourself and up to three opponents, and it returns machines where you hold the edge and machines where you do not, computed over the lineup that is actually in that building right now. The scouting routine post covers the workflow end to end.

When there are more than two of you

Group formats mean you are rarely in a clean one-on-one. For a group, the compare page puts up to four players side by side: IFPA rank and rating, Match Play rating, win rate, finish quality, tournaments played, wins and best machine, all in one table. Less granular than a head-to-head, faster to read when you have just been drawn into a four-player group and have about ninety seconds.

A caveat about small samples, again

Everything above is subject to the same warning attached to every other stat on the site: a record of 3-1 is a record of four games. Pinball is a high-variance game, groups are drawn semi-randomly, and a bad bounce decides more matches than anybody likes to admit.

The site tries to protect you from over-reading it. Minimum game thresholds on the rivalry and opponent panels, shrinkage on the win probability, game counts sitting next to every record so you can see what the number is built from. Click any game count and the log pops open with the individual games behind it.

But the last line of defence is you, reading the sample size before you make a decision based on the percentage. A 100 percent record is a very exciting-looking way of writing "twice".

The short version

1. Head-to-head for two players, lifetime, by machine, with the full game history.
2. The Opponents tab on a profile for everybody at once.
3. Common opponents when there is no direct record worth reading.
4. Compare when it is a group rather than a pair.
5. Scout when you want it turned into an actual machine pick.

More in the archive on the blog index, including how the machine score is calculated and what a machine page can tell you before you plunge.