Opens February 1, 2027
MahjPhD CardLab
The card, measured.
What are your real odds with every hand on the NMJL card? CardLab plays millions of games of American Mah Jongg to find out, hand by hand, every year.
The question
The card tells you what a hand is worth. It doesn’t tell you how often it comes together.
Same value, nearly five times the chance of Mah Jongg. Experienced players build a feel for this over years at the table. CardLab measures it directly, then turns the numbers into plain-language guidance you can use the next time you pick up your tiles.
Real results from the 2026 card.
Progress to date · October 2026
The 2026 card has been measured.
The simulator is built, checked against real play, and has finished its runs on this year’s card. A final quality check caught and fixed a scoring problem, and every game was run again. The April rehearsal is done, and the full 2026 report builds from the results. What’s left is the plain-language notes and a final review.
- Done Taught the computer the card All 55 lines of the 2026 card, written out as 1,077 exact tile combinations and checked line by line.
- Done Built a complete game The deal, the Charleston, calls, joker swaps and Mah Jongg, by NMJL rules, with all 152 tiles.
- Done Checked it before trusting it No rule broken in 20,000 test games; 20 games read move by move against the real card; tiles left in the wall and jokers per player within 25% of real online play.
- Done Played 5.5 million games 100,000 games for every hand on the 2026 card.
- Done Built the report Charts, hand-by-hand tables and a printable PDF, generated straight from the results.
- Done Rehearsed for April A full run, start to finish, in under an hour and a half of computer time, so the 2027 report can arrive about 10 days after the new card.
- In progress Notes for every hand Plain-language notes for all 55 lines are drafted and in review; the hand families are being named.
- Next Final review One last check of everything before the February launch.
A sneak peek at the 2026 card
A few things the numbers already show.
Card-wide findings from this year’s runs. The hand-by-hand details are saved for launch.
About 1 in 7 for the most reliable hand, 1 in 18 for an average hand, and 1 in 100 or worse for the long shots.
Hands dealt two or more jokers win about four times as often: 12.4% on average, against 3.1% with none.
Groups of hands that draw on the same tiles, each with its own Charleston advice on what to pass first.
Neighbors that let you switch course late without starting over.
How far your opening rack starts from a hand matters: the most reliable hand wins about 41% of games from a close start, and 4% from a far one.
Still 5 tiles away after 15 draws? Your chance of winning has dropped to about 1.6%. Time to protect your discards instead.
Every hand throws someone else’s winning tile in about 15–18% of games. Hands differ in what they win, not in what they give away.
The 50- and 75-point hands win so rarely that their value doesn’t make up for it. Several 25- and 30-point hands are the real bargains.
What’s inside
A full report card for every hand.
Each edition covers the current NMJL card, section by section and line by line.
How often it happens
Your win rate when you commit to each hand from the deal.
Joker dependence
Which hands lean hardest on jokers, and the best choices for the jokers you’re dealt.
How long it takes
The typical number of turns each hand needs.
Points vs. difficulty
Bargains, and hands whose points don’t cover the risk.
Backup hands
Which hands share your tiles, so you can pivot.
What to pass
Passing advice for each family of hands.
How the card changed
Returning, familiar and brand-new hands compared with last year.
Ready to teach
Slide-ready charts and an easiest-to-hardest teaching order.
When to commit
How close your opening rack needs to be before a hand is worth chasing.
When to give up
The point where a hand is a long shot and it’s time to play safe.
Where the points come from
What each hand wins, and what you pay when someone else wins first.
Every line, one page
Hand counts and the best partner lines for every line on the card.
The method
Monte Carlo simulation, explained in plain English.
When a question is too complicated to solve with a formula, you can answer it by experiment: play the game over and over, very quickly, and count what happens. That’s Monte Carlo simulation, the same idea used to forecast weather and price insurance.
One player, one hand
A computer player commits to the hand from the deal. The other three play whatever they like.
Same deals for every hand
Every hand faces the identical set of deals, with the committed seat rotating around the table, so no hand gets luckier tiles.
A steady, sensible player
Not a genius, but consistent: it plays every hand the same careful way, so hands can be compared fairly.
Count, then explain
Track which hands get made, how fast and with how many jokers, then turn it into charts and notes for the table.
Read the full method: how it was checked, its limitations, and an FAQ →
Details for Math Fans
A little of the math.
Three ideas behind the numbers, for the curious.
How does the computer know how close a rack is to a hand?
Every line on the card expands into its concrete 14-tile hands (1,077 on the 2026 card). For each one, the distance is the number of tiles still missing. Natural tiles fill singles and pairs first, since jokers can never go there; leftover naturals fill pungs, kongs and larger groups; jokers fill whatever group slots remain. That greedy fill is provably as good as the best possible assignment, which was confirmed against an independent max-flow calculation on 10,000 random racks. A rack is scored against the whole card in about a microsecond.
Why do all the hands get the same deals?
Game i uses the same random seed no matter which hand is being tested. This is called common random numbers. A lucky three-joker deal helps most hands at once, so comparing hands on identical deals cancels much of the luck: Var(A − B) = Var(A) + Var(B) − 2 Cov(A, B), and that covariance is positive.
How precise are the win rates?
With 100,000 games per hand, the typical 95% interval is about ±0.14 percentage points (±0.22 at the widest). Hands closer together than that aren’t ranked apart. A 20,000-game pilot run agreed with the final run almost perfectly (r = 0.999).
How it was checked, its limitations, and the full FAQ are on How CardLab Works →
Launch timeline
The card changes every year. So will the analysis.
The full 2026 edition opens February 1, 2027, and that membership carries straight through to the 2027 card. A free tier will share each year’s top hands.
Every edition is made the same way: see how CardLab works.
- Fall 2026 Simulator built, checked and run 5.5 million games on the 2026 card, checked, fixed and rerun, plus a full April rehearsal.
- Now In progress Notes for every hand and a final review Plain-language notes for all 55 lines, and naming the hand families.
- January 2027 Free preview The 2026 card’s top hands, measured.
- February 1, 2027 MahjPhD CardLab opens The full 2026 edition goes live.
- April 2027 The 2027 card, measured The full report about 10 days after the new card arrives, then a live webinar in spring and a midseason update in June.
Straight talk
Estimates, not promises. These are odds from millions of games; your next game can always surprise you. Measured by commitment: read each number as “how a hand does when you go for it.” You still need your card: CardLab names hands by section and line and never reproduces the card.
Be the first to see the numbers.
Join the list to hear about the free preview and the February launch as soon as they’re ready.
MahjPhD CardLab is an independent project and is not affiliated with or endorsed by the National Mah Jongg League, Inc. Analyses refer to hands by section and line number; the NMJL card itself is not reproduced. Please purchase your card directly from the NMJL.