Duplicate bridge across ACBL Districts 16 through 23 — every state from Colorado, Wyoming, Montana and New Mexico west, plus all of Texas, Mexico, British Columbia, Alberta, Saskatchewan and the Yukon — club games (open, limited and mentor), plus regional and sectional tournament pair events. Every rating is expressed against an average open field. Includes games through 08/15/26; page rebuilt 09/14/26. Ranked by rating; a rank appears only after a player has 1,000+ recorded deals overall — and, in a district, state, region or single-club view, 200+ of those at that area's own clubs. In the all-regions view it instead takes either enough cross-field play for a district-wide comparison — about one tournament day's worth of boards — or regular play at three or more clubs.
Individual district pages: District 16 · District 17 · District 18 · District 19 · District 20 · District 21 · District 22 · District 23
An experimental skill rating for duplicate bridge across ACBL Districts 16 through 23 — every state from Colorado, Wyoming, Montana and New Mexico west, plus all of Texas, Mexico and Western Canada — built by Doug Couchman from public ACBL results. It estimates each player's matchpoint skill on one common scale (50 = an average player) using a board-by-board statistical model — not a simple average of session scores. The model takes into account your partner's and your opponents' skill levels, so you aren't “penalized” for playing with a weaker partner or against strong opponents. Recent games count more than old ones — a board played ten years ago carries half the weight of one played today — so current form, improvement included, shows up in the rating. Time away never changes a rating, no matter how long — ratings move only when new results come in.
Unofficial. This is a personal project, not affiliated with the ACBL or any club, and not an official rating. It's a periodic snapshot (see the dates at the top), rebuilt from public results now and then — so it won't reflect the very latest games until it's refreshed.
City and masterpoints on a player card. These come from ACBL's own published pages, not from us: a player's city and state appear on club rosters and on tournament masterpoint-award lists, but a running masterpoint total is published only on club game rosters. So a player we have seen only in tournaments, or one whose last club game predates our capture of that field, will show “masterpoints unavailable”. Totals are also a snapshot from the most recent game we have on file, not a live figure — check the ACBL for the current number.
It covers every game these clubs have recorded to ACBL Live for Clubs (club games back to January 2, 2019) plus all regional and sectional tournament pair events back to January 1, 2018 — open, limited (99er through 749er and NLM), and mentor games — so anyone who has played at a covered club or tournament in that time is included; use the search box. Every rating is expressed as the expected score in an average open field. A player's rating in a single district can differ from their rating across the whole region — each is measured against the field that actually plays there, and field strengths differ. If you mostly play limited games, you can expect better scores there than the number shown here — the open-field competition this table assumes is stronger than the competition in limited games. To limit the display to the highest-volume players, check the “regulars only” box. Details on how to read the numbers are at the bottom of the page. Spotted an error in your record? Corrections are welcome.
By the numbers. 58,979 players rated from 167,069,908 board-by-board comparisons and 27,415,039 plays, over 1,768,226 tables, across 234,230 club sessions at 485 clubs plus 24,557 tournament sessions from 2096 tournaments (32 regionals, 2064 sectionals).
The model is tested on 38,368 sessions it has never seen — held out of training entirely, then predicted blind. Every number below comes from that test.
One session at a time, bridge results include a fair amount of luck. The typical gap between a pair’s predicted and actual game is about 5 percentage points — predict a 52% game, and most nights land between 47% and 57%. More than half of all pair-sessions (57%) come within 5 points of prediction; 88% come within 10.
Averages settle slowly, but they settle. The session-to-session swing (about ±6½ points) shrinks as results accumulate: over 10 sessions, the gap between prediction and a pair’s average is typically about ±2 points; over 50 sessions, about ±1. The part that never averages away — how far the model’s opinion of a pair sits from their true level — measures just ±¾ of a point.
Who beats whom. Two pairs in the same game, playing the same deals: rated 3–5 points apart, the higher-rated pair finishes ahead 2 times in 3. Rated 5–10 apart, nearly 4 in 5. More than 10 apart, 9 times in 10. Rated only 1–3 apart it is closer to a coin flip (59%) — small rating differences mean genuinely close matches.
Technical summary: across 458,590 held-out pair-sessions, correlation between predicted and actual matchpoint percentage = +0.57; mean absolute error 5.11pp vs the flat-50% baseline’s 6.23pp (improvement +1.07pp, session-clustered SE 0.007, n = 33,441 sessions); bias +0.23pp; calibration by predicted quintile within +0.1 to +0.3pp in every quintile. The luck/persistent split is measured on the 14,088 pairs with 8+ held-out sessions (median 14). The pooled seven-district model also predicts held-out sessions slightly better than any single-district model (by 0.02 to 0.05 percentage points per pair-session, each comparison 6–11 sigma on identical held-out sessions) — cross-district tournament play knits the scale together. Measured on the evaluation fit; the published model then trains on the full corpus, these sessions included.
▸ Click any player to open their full profile — declarer & defense ratings, every partner, ACBL links, and a shareable link.
Projected matchpoint % for the two playing together, from each player's skill, against the field they would actually face — pick the game above. A regional open pairs is a much tougher field than a club limited game, so the same partnership projects lower there. (Field strength only: we tested whether a pool's internal spread also needs rescaling and it does not — raw ratings already transfer across venues with correct calibration.) Both must be rated (they'll autocomplete); it doesn't model a specific partnership's chemistry. Handy for seeding or stratifying a field.
Players: How did we do? — paste a game's results page, see what every pair was expected to score against that exact field, and how unusual each result was.
Directors: Roster stratification helper — paste a whole game's signup list, get pair strengths and a balanced NS/EW proposal.
Ranked by Overall %. Players showing the same rating share a rank. A rank appears only after a player has 1,000+ recorded deals overall — and, in a district, state, region or single-club view, 200+ of those at that area's own clubs; the all-regions view instead asks for either enough cross-field play to a district-wide comparison — about one tournament day's worth of boards — or regular play at three or more clubs, so the regulars view can skip an occasional number — toggle the box to see who holds it. Click any column heading to re-sort: sorting by # lists ranked players only; sorting by any other column (or searching) includes everyone. The small range under each Overall % is its 95% confidence interval: we can't measure skill exactly from a finite number of boards, so we show the range we're 95% sure the true rating falls in — the more boards, the tighter it gets. One caution about comparing neighbors: each band describes one player's own rating, not the comparison between two players. Differences are measured more precisely than the bands suggest (much of each band's uncertainty is shared and cancels in a comparison), so a modest overlap does not mean the order is in doubt — two players can overlap and still be confidently ranked. Only heavily-overlapping neighbors are a genuine toss-up.
| # | ↓ | Player | Overall % | Declarer %ile | Defense %ile | Sessions | Boards |
|---|---|---|---|---|---|---|---|
| Loading ratings… (a large table — this takes a moment on first open) | |||||||
Reading it. Overall % is the expected session matchpoint % when you play with a partner of similar skill in an average open field — so two equally-rated players compose a pair that scores about this number against open competition, and the average open player's rating is 50. The band is the 95% confidence interval for that player's own rating; differences between players are known more precisely than the bands suggest, so a modest overlap does not put the ranking order in doubt. The # ranks players by Overall %; identical displayed ratings share a rank, and a rank appears only after a player has 1,000+ recorded deals overall — and, in a district, state, region or single-club view, 200+ of those at that area's own clubs; the all-regions view instead asks for either enough cross-field play to a district-wide comparison — about one tournament day's worth of boards — or regular play at three or more clubs. A player without that is still rated, but we don't claim a precise place for them in a district-wide order. Declarer %ile and Defense %ile are percentiles (field standing, 0–100) from double-dummy card-play analysis (declarer is individual; defense is a partnership rating); players with too few card-play boards show “—”. Derived from all club games uploaded to ACBL Live in Districts 16 through 23 since January 2, 2019, plus all regional and sectional tournament pair events played in those districts since January 1, 2018 — open, limited, and mentor. Displayed percentages are calibrated to an average open field (robots excluded from that average); the open competition this benchmark assumes is stronger than in limited games, so limited-game players can expect better scores in the fields they actually play. Method: a joint statistical model — a board-level Bradley-Terry contest model on game results, combined with double-dummy declarer card-play through one latent skill per player (hierarchical shrinkage; every open game at these clubs on one common scale). Results are time-weighted with a ten-year half-life, so recent games count more (a measured setting: prediction is flat from three years to no decay at all, and short half-lives measurably hurt, so we keep the long memory); time away, of any length, never changes a rating by itself — only new results do. The Declarer and Defense percentiles come from the same double-dummy card-play analysis. Robots (sometimes used to fill a half table) are shown with parenthesized figures and excluded from the percentiles. A † marks a player known by us to be deceased; this is not kept up to date.
Doug Couchman is a bridge player and instructor, and a private tutor specializing in graduate admissions exams (LSAT, MCAT, GMAT, and GRE). Click here for more information. Questions or corrections? Get in touch.
© Copyright Doug Couchman 2026.