Plain-English field guide
A field guide to the simulator’s 17 named strategies.
The simulator's names are shorthand for recognizable habits: unload the most tiles, reduce penalty points, save a joker, protect a lead, or avoid a complicated rebuild. This page translates all 17 into decisions a person can notice and practice.
How the simulator works
The engine finds legal moves; the strategy chooses among them.
The mechanical search finds legal ways to use your rack and rebuild the table. A strategy then chooses among those candidates according to a consistent tendency. Every serious strategy takes an immediate win when the engine finds one; the differences below matter when several non-winning choices remain.
For human play, try one tendency at a time. For example, shed points when two moves are otherwise close. There is no need to memorize the simulator’s weights.
Why the report tests more than these 17 names: it also switches five tendencies on and off in 32 controlled combinations. Those combinations test ingredients; the named strategies remain the ones a person can choose and recognize.
Family 01
Strong general play
Simple priorities that are useful on nearly every turn.
Balanced
All-purpose baseline
Tries to play a useful number of tiles, shed meaningful points, preserve a flexible rack, and avoid making the table needlessly easy for the next player.
- Try this at the table
- Use this as your default mental checklist: Can I go out? What can I play? Which choice lowers my risk without wrecking my next turn?
- Watch out for
- A compromise can lose to a more decisive priority when the rack or score clearly calls for one.
Greedy Tiles
Rack emptier
Chooses the legal move that gets the most physical tiles off the rack; points break a tie.
- Try this at the table
- Count tiles before admiring patterns. A five-tile move usually deserves attention before a prettier two-tile move.
- Watch out for
- It can spend useful connectors or leave high-value tiles behind.
Greedy Points
Penalty reducer
Chooses the legal move that removes the most rack points; tile count breaks a tie.
- Try this at the table
- When two moves are both sound, unload the 10s through 13s and any joker that could become a painful penalty.
- Watch out for
- Pure point shedding can break a more flexible low-number rack.
Family 02
Risk and timing
Different answers to when you should commit tiles or keep waiting.
Conservative
Downside controller
Emphasizes high-point tiles and favors less complicated table work, aiming to reduce the damage if another player ends the game.
- Try this at the table
- If an opponent is nearly out, stop saving an elegant rack and reduce the score you could be caught holding.
- Watch out for
- Playing too safely can pass up the fastest route to your own win.
Strategic Drawer
Selective waiter
Requires a stronger opening and may draw instead of making a marginal play, especially when the rack still has several ways to improve.
- Try this at the table
- Occasionally keep a genuinely versatile rack intact when the available play is small and opponents are not threatening.
- Watch out for
- Waiting feels strategic long after it has become expensive; this style performed poorly in the pilot.
Late Opening Hoarder
Cautionary extreme
Waits for a 45-point opening and is unusually willing to keep drawing rather than enter the table.
- Try this at the table
- Use this strategy to test the common temptation to wait for a spectacular opening. The pilot results do not support copying it in human play.
- Watch out for
- It can finish with a large, expensive rack; this was one of the pilot's clearest losing tendencies.
Endgame Liquidator
Emergency unloader
Aggressively sheds points and high tiles, and becomes comfortable spending jokers rather than holding them for a perfect moment.
- Try this at the table
- Switch into this mindset when a player has only a few tiles or when protecting a match lead matters more than building the ideal rack.
- Watch out for
- Used too early, it can spend flexible pieces before they have done their best work.
Match Aware
Scoreboard adapter
Plays normally in a single game, protects a large late-match lead by shedding points, and accepts more variance when far behind.
- Try this at the table
- With one game left, use the match score to decide whether to protect a lead or accept more risk in pursuit of a comeback.
- Watch out for
- The pilot did not establish reliable score-gap switching rules; its current thresholds are hypotheses to test.
Family 03
Tile-shape specialists
Strategies that value particular resources or the shape left behind.
Joker Holder
Joker saver
Strongly prefers to keep a joker when another good legal play exists, preserving its flexibility for later.
- Try this at the table
- Do not spend a joker merely because you can; ask what concrete future problem it is solving.
- Watch out for
- A held joker is worth 30 penalty points, and the game can end before its flexibility pays off.
Joker Farmer
Joker retriever
Values keeping jokers and places an especially high value on legally replacing and recovering one from the table.
- Try this at the table
- Scan every exposed joker for a replacement from your rack before evaluating ordinary plays.
- Watch out for
- Retrieving a joker is not automatically progress; the resulting move still needs to improve the whole rack.
Pivot Copy Keeper
Flexibility preserver
Values tiles that connect to several possible runs or groups, while discounting duplicate copies that add little new potential.
- Try this at the table
- When choosing what to play, keep the tile that participates in several plausible future melds and release the more isolated copy.
- Watch out for
- Future flexibility is only valuable if you survive long enough to use it.
Table Denial
Opponent friction
Prefers final table layouts with fewer obvious extension points for the next players.
- Try this at the table
- If two moves unload similar racks, prefer the table that exposes fewer easy ends and fewer reusable pivots.
- Watch out for
- Do not sacrifice substantial immediate progress for a small and uncertain inconvenience to opponents.
Family 04
Human-style constraints
Useful models of how real players simplify a difficult table.
Eager Beginner
Play-now simplifier
Opens at the legal 30-point threshold and mainly tries to place tiles, without valuing sophisticated rack shape.
- Try this at the table
- Its good habit is decisiveness: take clear legal progress instead of waiting for a masterpiece.
- Watch out for
- It can use flexible tiles carelessly because it does not think much about what remains.
Manipulation Averse
Minimal rearranger
Places a very large penalty on moving or rebuilding melds already on the table.
- Try this at the table
- Use it to model your current comfort level, then practice finding one more table-rearrangement option than feels natural.
- Watch out for
- This restriction left a great deal of playable value unused and produced the pilot's largest score penalty.
Family 05
Adaptive and search
Policies that use public context or a small amount of look-ahead.
Opponent Modeler
Public-pressure reader
Uses visible rack counts and recent plays to become more reluctant to leave an easy table when an opponent appears close.
- Try this at the table
- Notice who is down to a few tiles and tighten up choices that would hand that player an obvious extension.
- Watch out for
- Public clues are incomplete; imagined opponent knowledge can overwhelm better immediate play.
Shallow Planner
One-draw look-ahead
Evaluates the exact mix of unseen tile types one draw ahead, rewarding racks with likely completions and penalizing tables that likely help opponents.
- Try this at the table
- After a close move, count how many unseen numbers can complete what remains rather than saying only that the rack 'looks flexible.'
- Watch out for
- One draw is still a short horizon and cannot forecast the full multiplayer game.
Random Legal
Research control
Selects randomly among legal candidates instead of expressing a coherent preference.
- Try this at the table
- Use it as a baseline opponent that shows whether a proposed strategy adds value beyond legality alone.
- Watch out for
- It is not intended as advice for human play.
Glossary
Six terms used in the report
- Strategy / policy
- A repeatable preference for choosing among legal moves. It is not a claim to have solved the whole game.
- Mechanical tier
- How thoroughly the engine searches for legal table arrangements. This is separate from which legal move a strategy prefers.
- Pilot
- An exploratory study used to find promising patterns and problems before a larger confirmation run.
- Independent deal
- A genuinely new shuffled game. Rotating seats on the same deal improves fairness, but does not create new independent evidence.
- Table composition
- The particular mix of strategies seated together, such as Balanced against three different specialists.
- Percentage point
- A direct difference between win rates. A move from 25% to 28% is three percentage points.