Read this first: these are even-match rules
Every rule below assumes you and your opponent throw at roughly the same level. That caveat is not a footnote — it changes the answer. Take one fixed board — you two numbers up but down 51–0, your throw. Against a weaker opponent you win it about 96% of the time; against a stronger one, under 1%. Same darts in your hand, opposite call — the skill gap swamps the tactics. So when that gap is wide at least three of these rules invert: the farm cap collapses, and you deny and stall instead of racing to close. If you are clearly the weaker thrower, read Playing Outgunned first — it is a different game. These numbers are from the Teacher, which simulates the position at whatever skill levels you enter; the rules below are measured between evenly matched players unless one says otherwise.
The card, in play order
If you read nothing else: the rules by where you are in the game. Each is costed by an exact endgame solve or a measured tournament — the reasoning is below.
- Faucet
- A closed number that only one player has — that player scores points on every hit while the other still can’t answer. It pays out like a faucet, for whoever owns it.
- Open a faucet
- Close a both-open number so it becomes your scoring lane — income only you can collect. (The offensive move.)
- Deny a faucet
- Put your marks on a number the opponent has already closed — the lane they’ve been scoring on. Once you close it too, the door shuts and their income stops. (The defensive move — the +9.4pp mechanic.)
- Farming
- Piling up points by hitting your open faucets over and over instead of closing out.
- Camping
- Sitting on your own last number trying to close it, when you should be scoring on an opponent’s lane first.
- Closing race
- Who will close all seven numbers first — a race of marks, not points. Count marks-to-done on both sides to know if you’re ahead in it.
Start — opening the board
- Open in value order: 20 first, then down; bull last.
- Three marks to close a number? Aim the triple — it closes in one dart.
Middle — scoring and covering
- Count marks to close on both sides (not points). Losing that race? Stop farming — if they close first, your points are decoration.
- Ahead? Deny their faucets before you open your own. Worth +9.4pp.
- Even skill, big lead? Stop farming past ~80 points ahead and go close.
End — closing it out
- Score level and you can close your last number? Close it — ties go to the closer, and the close wins now.
- One mark from closing — including the bull? Take the easiest single, not the double: one mark is all you need.
- Behind or level with one lane of your own left? Don’t camp it — score on an opponent’s lane first.
Outgunned? Three of these invert — see Playing Outgunned. Or put any board to the solve on your phone: ask The Teacher for perfect play →
The law that outranks all others: ties go to the closer
Start with the one rule that is not advice but arithmetic. In standard cricket, you win by closing all seven numbers with a score at least equal to your opponent's — at least equal. A tie in points is a win for whoever closes out first. So when the score is level and you can close your last number, close it. Do not throw for points, do not throw at a double to set up a bigger finish. The single wins the game on the spot.
The exact endgame solve (Ground Truth) put a precise cost on getting this wrong. The largest single-decision blunder in the entire audit catalog is exactly this position: score level, one dart in hand, one number left to close, opponent a mark from closing out. The safe single is worth 0.966 win probability; the double the strategy actually threw is worth 0.403 — 0.563 of a game forfeited in one throw. The Shape Reader (X188), the old site's champion, has the identical blind spot: at a tied score with 15 two-thirds closed, it plays the double bull and forfeits 0.542 .
Both discovery lineages had a chance to find this rule. The rerun found the full mechanic — close the last number, take the minimum-miss single, treat a tie as a winning position — at iteration 3 of 114 . The original lineage never found it in 115 entries. If you take one sentence from this site to league night, take this one.
The closer's law
Score level and your last number closeable: close it. A tie goes to the closer, so the single wins outright — worth 0.966 win probability against 0.403 for the double, in the exact solve's worst catalogued blunder .
The pocket rule card
The rest of the card comes from two sources, blended. The Textbook (DIST) is a neural policy written down as rules a human can read: the interpretable distillation of the PPO run that seeded the RL arc (PPO_run2, Elo 1205.8) . It is useful precisely because it is legible — not because it is the strongest. On the unified ladder the Textbook sits mid-pack at Elo 1085 (rank 12 of 27, 24,000 games) : it beats all seven classic-era bots it played (55.7–70.9%) but loses about 62% of games to the best hand-coded champion, The Shutout (X181) , and it sits below The Closer as well. The Closer (it3) is the actual project champion — the neural policy that, as of July 2, 2026, tops the unified ladder at Elo 1222 (24,000 rated games) and matches the exact solve on 74.0% of 71,961 audited endgame decisions . So where the Textbook's doctrine is silent or beaten, the card borrows from champion doctrine and from the solve itself — and every rule cites the number that justifies it.
Why bother with rules at all? Because the measured price of ordinary decisions is large. Splicing a perfect endgame onto the Shape Reader — changing nothing else — beat the original 54.40% of the time (+4.4pp, 50,000 games) , a bigger edge than most championship margins on the ladder. The decisions on this card are where that edge lives.
| # | When | Do this | The number |
|---|---|---|---|
| 1 | Score level, last number closeable | Close it — ties go to the closer | EV 0.966 vs 0.403 for the double |
| 2 | Choosing what to open | Close in value order: 20 down, bull last | Re-derived independently by both discovery lineages |
| 3 | Tempted to keep scoring | Count marks (not points) to close on both sides; if they'd close first, stop farming and disrupt | This race-gated farming carried The Shutout (X181) to Elo 1176 |
| 4 | Big lead, even skill | Stop farming past roughly an 80-point lead; go close | Champion doctrine's cap; shrinks to 10 when outgunned |
| 5 | One mark from closing — including the bull | Aim the minimum-miss single, never the double | +0.27pp with no losing matchup; the double-at-bull habit forfeits up to 0.542 in one throw |
| 6 | Ahead and covering | Shut their income lanes before claiming empty ones | +9.4pp in one step — the largest single mechanic found |
| 7 | Behind or level, one lane left of your own | Don’t camp your last number; score on an opponent lane first | The dominant blunder motif: 188 of 260 catalogued exemplars |
Three of these deserve a word of expansion. Value order (rule 2): both the original discovery loop and its context-isolated rerun converged on the same lane-selection law — take numbers in descending value — without sharing a line of code or a journal entry . Convergence from independent searches is weaker evidence than a solved number, and we present it as such; but every champion in the project walks the board top-down, and no kept experiment ever beat that ordering.
The race (rule 3): before you farm another 60 points on your open 20, count the marks each side still needs to close out. If your opponent would finish the closing race first, your points are decoration — closing wins games, points only break the tie you just learned goes to the closer anyway. The Shutout (X181), the best hand-coded strategy in the project at Elo 1176 (94,000 rated games) , gates every scoring decision behind exactly this race check .
The single, even at the bull (rule 5): when one mark closes a number, the play is the aim-point you are least likely to miss — and at the bull that means the single bull, not the double. The original champion inherited a double-at-the-bull habit at its third-ever experiment and carried it, unexamined, through all 115 entries of its lineage . The rerun measured the correction at +0.27pp with no losing matchup , and the solve showed the worst case of the old habit is rule 1's forfeited game.
A lead makes everyone sloppy — plan for it
The endgame audit's most humbling finding is not about any one strategy. When the exact solve scored every audited artifact's endgame decisions — champions and baselines alike, 52,000–75,000 decisions each — agreement with perfect play collapsed whenever the player was ahead: 52–62% agreement ahead, versus up to 90.2% when behind . Behind, your instincts are trustworthy: chase, close, deny — the desperate moves coincide with the optimal ones. Ahead, instinct says race to close out, and instinct is often wrong: perfect play keeps scoring and keeps denying on the opponent's open lanes, holding the door shut instead of sprinting for it.
The blunder catalog gives the impatience a shape. Of the 260 worst catalogued decisions across all thirteen audited strategies, 188 are "closed when they should have scored" — typically throwing at one's own last open number (usually the bull) while level or behind, instead of scoring first on a lane the opponent cannot answer. Only 55 run the other way, and 17 chose the wrong lane to close. Two different game states, one error: drifting toward your own last number before the position has earned it. Ahead, the fix is rules 3 and 6 — keep scoring and keep their lanes shut. Level or behind with one number left, the fix is rule 7 — build points they must answer before you camp the finish.
| Blunder motif | Count (of 260) |
|---|---|
| Closed when they should have scored | 188 |
| Scored when they should have closed | 55 |
| Closed the wrong lane | 17 |
When to deny: shut the faucet before you mop the floor
Rule 6 is worth its own section, because it is the single most valuable mechanic either discovery loop ever found, and it is playable by a human without any arithmetic. Think of every number your opponent has closed and you haven't as an open faucet: every turn it stays open, they can pour points through it. When you are comfortably ahead and covering, the instinct is to claim fresh numbers — empty real estate no one is using. The measured play is the opposite: close their numbers first. An empty number threatens nothing this turn; an open faucet costs you every turn it survives.
When the rerun's loop added exactly this rule and nothing else, its champion's mean win rate jumped from 55.0% to 64.4% — +9.4pp in a single step , which the loop's own journal called the largest single mechanic found in the entire project. The gain was biggest against exactly the opponents you'd expect: pure scorers and point-hoarders, where the faucet matters most (+17.9pp against the points-first baseline PS, +15.7pp against S6, +13.9pp against E1) .
And when should you score instead of deny? The audit's answer from the other direction (rule 7): while the game is level or you are behind, points on a lane the opponent has not closed are pressure they must answer. Denial is a winner's tool — it converts a lead into a locked door. Scoring is how you build the lead worth locking. The failure mode to avoid is the one in the blunder catalog: neither scoring nor denying, just camping your own last number while level, hoping to hit the finish before they do.
The scope note again, because it matters most here: every number on this card is from simulated players with calibrated accuracy profiles, at matched skill unless stated. Denial is a lead-holding tool, and how big a lead is worth holding depends on who you're throwing against. When you are the clearly weaker thrower the caps and thresholds all move — that is the next chapter, where The Grinder (X214_UD) cuts the 80-point farm cap to 10 and denies at any deficit, not just when ahead .
The exact solve behind these rules also powers The Teacher — a position-by-position tool: set up any board on your phone and it returns The Closer’s instant call, then the exact tablebase answer, the win-rate odds for every throw, and the named laws in force. Try it: ask perfection from your phone →