scaffoldrummy.com | Est. 2004
A mathematically superior ruleset for two-player Rummikub. Twenty years of playtesting. Seven amendments. One connected structure.
"Two hours of Rummikub math at midnight. Ryan has no idea what's coming for him."
— The AI 2:21 AM May 26, 2026Page One
You know the game. You've been setting it up on that table for twenty years — the worn board, the classic tiles, that riser that's probably older than some of our cars. We play it slow. We play it serious. We play it like two men who have absolutely nothing to prove and also everything to prove.
At some point — neither of us can remember exactly when — we stopped playing the game they put in the box. We started playing ours. A tile couldn't just sit there unconnected. A wildcard had to be earnable, releasable. A demolition had consequences. These weren't rules we discussed. They were rules we arrived at, the way you arrive at a handshake you've used so long you've forgotten inventing it.
I started wondering, recently, whether we'd done it right. Whether our gut had tracked the math, or whether we'd been playing a beautifully flawed game all this time and calling it wisdom. So I ran the numbers.
Ryan. We did it right.
Not perfectly — there are two places where we left a door open that we shouldn't have, and I'll show you exactly where. But the bones of what we built? The scaffold? That's real. The math agrees with us. Every amendment we made moves the game in the same direction: less luck, more skill, more tension, more of the thing that makes you put the tiles down at 1 AM and say "one more."
This is a record of what we built, why it works, and what we should fix. It's also a challenge: read this, play me, and tell me I'm wrong.
— David
When Ryan finds this page, he gets edit access. This document becomes a conversation.
Page Two
Each amendment is scored on three axes against standard Rummikub rules.
L = Luck Weight (lower is better) • E = Exploit Surface (lower is better) • S = Skill Ceiling (higher is better)
No tile on the table belongs to anyone. Any player may dismantle any structure at any time, provided the board is fully valid at the end of their turn.
† Open Board introduced an exploit (strategic demolition harvest) that we under-penalized. See Amendment III. The skill ceiling gain is real; the exploit door should be closed.
The first meld placed on the table is the anchor. All subsequent plays must connect to the existing structure. The board is one thing, not many.
The most consequential amendment. Standard 2-player Rummikub degenerates into parallel solitaire — both players optimizing private groups with minimal interaction. The Scaffold Rule forces shared dependency. Your move changes what I can do. This is better game design, and the math confirms it.
If a player demolishes the board and cannot reassemble all tiles into valid sets, they must pick up every orphaned tile plus draw one tile from the pile.
The exploit: a player already on the board can engineer a demolition, cherry-pick useful tiles from the wreckage, accept the single-draw penalty as the cost of business, and come out ahead. We never ran this play — too honest — but the door is open. Proposed fix: orphans + two draws. The risk/reward math closes at two.
There is no turn timer. A player may take as long as needed to plan and execute a turn, including full board demolition and reconstruction.
Standard tournament Rummikub allows 60 seconds per turn. With Open Board and Scaffold Rule, valid turns require genuine board-state computation. No Clock is not laziness — it's matching the time allowance to the actual decision complexity.
A wildcard may be retrieved from the board only by playing the exact tile it represents. No substitutions. No partial matches.
Without this rule, a wildcard placed on the board exits the playable pool permanently. With 4 wildcards in a 108-tile deck (3.7%), each buried wildcard represents a permanent reduction in the game's highest-utility tiles. Release keeps them circulating. This is the amendment we're most proud of.
Drawing from the pile is your full play for that turn. The drawn tile goes into your rack. It may not be played until your next turn.
In standard Rummikub, 23.1% of draws are immediately playable — the deck rescues you from a stuck turn. This amendment zeros that out. Being stuck costs a full turn, every time. It makes board-reading more valuable and luck-recovery impossible. The Monte Carlo study quantifies the skill-transfer precisely.
Both players know each other's tile count at all times. No hidden rack state. Endgame is visible to both players simultaneously.
We backed into this one through hardware — our old board shows the rack partially. But the effect is real: you can't be ambushed. You see the endgame coming and play defense accordingly. The surprise-win is a luck event. Removing it moves one more outcome from chance to skill.
Page Three
10,000 simulated games under standard rules vs. Offutt-Hensley rules. The simulation tracks opening hand probability, draw-when-stuck playability, game length distribution, and luck-attributable win rates.
Full simulation results including first-mover advantage, comeback frequency, and demolition EV publishing soon. The simulation code is open source.