The euchre-tournament scheduler. It plans your whole night — who partners whom, against whom, at which table, in every game, fairly — then prints everything the night needs: player cards, wall schedule, table cards, the score sheet, and the final tally sheet. The night itself runs on paper. How the math works → · What happens at your headcount →
Fixing a spelling, or swapping in whoever showed up instead — the schedule stays exactly as it is, only the names on the printouts change.
Only for keeping the night on a computer afterwards. The night itself runs on the printed sheets — you never need these.
EuchrePlanner.com · from one table to a full hall
Euchre seats exactly four to a table, so every guest count N splits the same way: ⌊N/4⌋ tables, and the remainder — N mod 4, between 0 and 3 people — sits out that game. EuchrePlanner.com rotates the sit-outs under two guarantees: across the night nobody sits out more than one game more than anyone else, and nobody sits out a second time before every player has sat out once.
The second law of the room: in each game you play, you share a table with exactly 3 other people. So over a night where you play g games, you can meet at most 3g guests — which means everyone can meet everyone only if 3 × (games played) ≥ N − 1. Call the surplus your slack. Measured across live runs at every count, the law is crisp: with 4 or more slots of slack, the scheduler reached full coverage on every single run; at slack 0–1 (19, 21, 22, 24 players) a handful of pairs miss — and where we went looking with exact solvers, that gap turned out to be real, not laziness: at 12 players in 4 games a perfect schedule was proved not to exist, and at 24 in 8 nobody has ever found one. Below zero slack (23, and 25 up) arithmetic itself says no. So the honest ceiling for “everyone plays everyone” in a classic 8-game night is 20 players; past that, the lever is more games, not a better scheduler.
4 — one table, the whole night. Everyone meets everyone in game one; there are
only 3 possible partners, so partnerships must repeat from game 4 on. The tool schedules
it honestly; the printouts still run the night.
5–7 — one table plus rotating sitters; the sit-out rotation is the schedule.
8, 12 — two and three tables, nobody sits, real scheduling begins. 8 is a
solved game: every night length from 3 to 8 games ships precomputed and proved best
possible. At 12, five games is the precomputed schedule that seats every pair; four
games cannot, and that is proved, not a limitation of the tool.
16 — the magic square: four tables, and a mathematically perfect night exists —
every pair meets, no partner repeats, no pair over-meets. EuchrePlanner.com ships it
precomputed (see the math page).
17–24 — sit-outs return (except at 20 and 24). At 20 EuchrePlanner.com ships
precomputed schedules for 6, 7 and 8 games; 7 and 8 are proved best possible, while 6
games cannot seat every pair at all — 18 tablemate seats for 19 other people — so the
6-game design is the best known short night, leaving 16 pairs unmet. At 24, nine
games is precomputed and perfect, while eight games leaves 10 pairs unmet in the best
schedule known. Everywhere else the optimizer chases coverage.
25–30 — past the 8-game frontier: meeting everyone is arithmetically
impossible. Add games, or split into two rooms — at 32, two perfect 16-player halls
beat one crowded one.
Verdicts are empirical, not aspirational — measured across three or more live generations at every count, and at every accented row (8, 12–22 and 24) settled by exact solvers rather than by sampling. The fairness audit on the schedule screen always reports what your actual schedule achieved.
EuchrePlanner.com · the expert-level explainer
A schedule for a euchre night is a sequence of games; in each game the guests split into tables of four, and each table splits into two partnerships. "Fair" decomposes into four measurable promises, in priority order:
1 — everyone shares a table with everyone at least once over the night;
2 — a different partner every game, never repeated;
3 — no pair over-meets — nobody sees the same face game after game;
4 — tables and sit-outs rotate so nobody camps in one seat all night.
These promises compete. The interesting mathematics is exactly where they collide.
With 16 players and tables of 4, each game hands every player exactly 3 tablemates. There are 15 other guests, so meeting all of them needs at least 15/3 = 5 games — and a schedule achieving this bound in exactly 5 games, with every pair meeting exactly once, is a very special object: a resolvable 2-(16,4,1) block design. It exists, it is unique up to renaming the players, and it is the affine plane of order 4, AG(2,4).
The construction: place the 16 players at the points of a 4×4 grid coordinatized by the finite field GF(4). The "lines" of this geometry — sets of 4 points of the form y = mx + c for a slope m, plus the vertical lines — sort into 5 parallel classes (one per slope, including ∞), each class partitioning all 16 points into 4 disjoint lines. Each parallel class is one game's table assignment. Two distinct points determine exactly one line — which is precisely the statement that every pair of players shares a table exactly once across the 5 games.
Games 1–5 of the 16-player schedule shipped in this tool are AG(2,4). If your night ends early after game 5, it was still perfect.
A real night runs 8 games. Partner distinctness is information-theoretically easy (8 partners needed, 15 available) but combinatorially tight, because games 6–8 must reuse pairs that have already met. The shipped schedule was found by exact constraint solving (CP-SAT) subject to: partnerships never repeat (all 8 partners distinct for every player), every pair still meets, and no pair shares a table more than twice all night. For comparison, the incumbent tool's schedule for the same 16-player, 8-game night had 12 pairs who never met and pairs meeting up to 4 times.
What about promise 4 — table rotation? Eight games, four tables: perfection would be every player at every table exactly twice. It cannot be done:
Theorem. No assignment of tables to the five parallel classes of AG(2,4) keeps every player to at most two visits per table. Consequently no 8-game schedule containing the perfect 5-game core can achieve perfect table balance — no matter how games 6–8 are chosen.
Proof sketch. Identify the 4 tables with the group (ℤ₂)² and the 16 players with the vector space V = (ℤ₂)⁴, so each parallel class is the coset partition of a 2-dimensional subspace (its line through the origin; the five such lines form a spread, partitioning the 15 nonzero vectors). Here is the lucky accident that makes the proof go: a 4-element vector space is so small that every permutation of it is affine — AGL(2,2) ≅ S₄, both of order 24. So any way of labeling a class's four blocks with the four tables makes each player's table an affine function of their coordinates, and Fourier analysis applies to all labelings, not just structured ones.
For a player with table-visit counts n(t) over the five games, orthogonality of characters forces the exact identity
Σplayers ( 4·Σt n(t)² − 25 ) = 240,
because the five classes contribute five distinct characters of V (each hyperplane of V contains exactly one spread line — a counting argument — so the cross terms vanish). But if every player visited every table at most twice in five games, each player's term is at most 11, for a total of at most 16 × 11 = 176 < 240. Contradiction. ∎
The same identity is sharp enough to characterize the best possible: capping everyone at three visits per table, exactly four players must carry one triple-visited table, and at optimum they finish 3-2-2-1 while the other twelve finish a perfect 2-2-2-2. The schedule shipped here respects the cap — every player visits every table at least once and at most three times. (The incumbent's sheet, for contrast, left one player without ever visiting table 4.) Closing the last gap to the provably optimal rotation profile is the planned Phase Three upgrade; it changes nothing about promises 1–3, which are already at their mathematical maximum.
Away from the magic numbers, this becomes the social golfer problem — a famously hard combinatorial search problem with no known efficient exact method in general. Perfection is usually arithmetically impossible anyway: with N players and G games, each player gets at most 3G tablemate-slots for N−1 other guests, and divisibility rarely cooperates. When your headcount isn't a multiple of four, N mod 4 players sit out each game, rotated so no one sits out more than one game more than anyone else.
For these cases EuchrePlanner.com runs a randomized greedy construction with local swap-improvement and many restarts, optimizing lexicographically: repeated partnerships first (weighted essentially infinitely), then never-met pairs, then the worst pair's meeting count, then table and sit-out balance. It runs for a fraction of a second in your browser and reports its audit honestly on the schedule screen — the numbers you see are measured from the actual schedule, never assumed.
Sixteen is not the only headcount worth solving once and shipping. The whole club set — 8, 12, 20 and 24 players, at every night length anyone would actually play — was put through the same exact machinery (CP-SAT for bounds and proofs, simulated annealing for construction), and the best schedule found for each is embedded in this page. Twelve of the seventeen are provably the best schedules that exist: all six eight-player lengths, 12 players in 7 and in 8 games, 20 in 7 and in 8, and 24 in 9 and in 10 — each hitting the arithmetic floor on every promise at once.
Eight players is a solved game. From 3 games to 7, every pair meets, no partnership ever repeats, and nobody meets anyone more than three times. Three is not the optimizer settling: the naive floor says two, and CP-SAT proves two unreachable at 3 and at 4 games — a wall, not a shortfall. At 7 games the night is perfect in the strongest sense available: every one of the 28 pairs partners exactly once, the classical eight-player whist tournament. At 8 games the arithmetic runs out — 32 partnerships to hand out and only 28 pairs to hand them to — so four partnerships must repeat, and four is exactly what the design repeats.
The interesting failures are worth stating plainly, because they all look achievable on paper. 12 players in 4 games has exactly one slot of slack (12 tablemate seats for 11 other people), and CP-SAT proved in 37 seconds that no such schedule seats every pair — not this one, not any. Seven unmet pairs is the best known, and a fifth game removes the problem entirely. 24 players in 8 games has the same razor-thin slack; the best schedule anyone here could construct still leaves 10 pairs unmet, and neither a perfect one nor a proof of impossibility turned up. Nine games is the dependable answer for 24 — it seats all 276 pairs at the floor on every metric.
20 players in 6 games is the one design that does not seat everyone, and it cannot: the slack is negative. Six games give each guest 18 tablemate seats for 19 other people, so at least ten pairs must miss no matter what, and the best schedule found leaves 16. It ships because it beats the live optimizer decisively (16 against a typical 21), not because it is complete. A seventh game covers everyone.
The general lesson: the arithmetic floor is a necessary condition, never a sufficient one. It says you have enough seats; it cannot say the seats can be arranged. At two or more slots of slack the floor has been real and reachable everywhere we looked — 24 players in nine games has exactly four, and lands on it on every metric at once. Below zero it is not a floor at all: at 20 players in six games the seats simply do not exist. At exactly one slot — or even zero — the truth turned out to depend on who sits out, which the next section explains.
The counts that are not multiples of four — where one, two or three guests sit out each game — got the same exact-solver treatment, and the results were a surprise. Every headcount from 13 to 23 now ships precomputed designs (46 of them, at every night length from 6 to 10 games, plus a 5-game 13), and the number of games needed for everyone to meet everyone fell at eight of the nine counts: 13 and 14 need only 6 games, 15 and 17 need 7, 18, 19 and 21 need 8, 22 needs 9, and 23 — long measured at twelve — needs 10, proved optimal at the arithmetic floor. At 13 players the answer is exact in the strongest sense: six games suffice and CP-SAT proved five cannot. Every sit-out design also honors the rotation promise above: sit counts within one of even, and nobody sits twice before everyone has sat once.
The surprise is what happened to the razor's edge. On paper, 19 players in 8 games has zero slack for its least-played guests — each must meet a new face in every single seat — and the same arithmetic that is provably fatal at 12/4 said this should be a mirage. It is not: a schedule exists, it is embedded in this page, and it is provably optimal on every metric at once. The freedom to choose who sits out is exactly the freedom the multiples of four lack, and it turns out to be worth more than a slot of slack. The one holdout is 18 players in 7 games: every search ends one pair short — exactly one, every time — and neither a completing schedule nor a proof of impossibility has been found. An eighth game settles it.
| headcount | guarantee |
|---|---|
| 8 players, 3–7 games | precomputed, proved optimal on all three: every pair meets, no repeated partnership, and no pair meets more than three times — three being proved unavoidable. At 7 games every pair partners exactly once |
| 8 players, 8 games | precomputed, proved optimal on all three: 32 partnership slots against 28 pairs, so four repeats are forced — and four is what it repeats |
| 16 players, ≤8 games | precomputed optimal design: games 1–5 perfect (every pair exactly once); 8 distinct partners; every pair meets 1–2 times; all four tables visited |
| 20 players, 7 or 8 games | precomputed, proved optimal on all three: every pair meets, none more than twice, no repeated partnership |
| 24 players, 9 or 10 games | precomputed, proved optimal on all three |
| 12 players, 7 or 8 games | precomputed, proved optimal on all three: every pair meets, none more than three times, no repeated partnership |
| 12 players, 5 or 6 games | precomputed; every pair meets, which is the best possible coverage at that length. Max three meetings is best known, not proved minimal |
| 12 players, 4 games | precomputed best known: 7 pairs never meet, and zero is proved impossible at 4 games |
| 24 players, 8 games | precomputed best known: 10 pairs never meet, no pair over twice; better than the optimizer's typical 17, and still not zero |
| 20 players, 6 games | precomputed best known, and the only design that does not seat every pair: 16 miss, and full coverage at this length is arithmetically impossible (at least 10 must miss). 7 games covers everyone |
| 13–23 players (sit-out counts), 6–10 games | precomputed at every length; at and above each count's coverage minimum (13/14 → 6, 15/17 → 7, 18/19/21 → 8, 22 → 9, 23 → 10) the design is proved optimal: every pair meets, none more than twice, no repeated partnership, sit-outs within one of even — and nobody sits out twice before everyone has sat out once |
| the same counts, below the minimum | precomputed best known, honestly labeled: the audit note says exactly how many pairs must miss and whether that is proved (13/5: zero coverage proved impossible; 18/7 and 22/8: open, best known 1 and 6 unmet) |
| any of the above with a couples policy | optimizer, not the design — a precomputed schedule knows nothing about who is married to whom. Ignoring couples is recommended: it is the only setting that gets the verified schedule, and a couples rule usually costs repeated partners or pairs that never meet |
| everything else | optimizer + honest audit; repeated partners avoided whenever arithmetically possible; couples policy enforced as a hard constraint |