The Game
Cricket is one of the most popular darts games played in bars, leagues, and tournaments worldwide. Two players take turns throwing three darts per turn at a board with seven targets: the numbers 15, 16, 17, 18, 19, 20, and the Bullseye. The objective is to "close" all seven targets before your opponent while accumulating at least as many points.
Each target requires three marks to close. A single hit scores one mark, a double scores two, and a triple scores three — so a single triple can close a number in one dart. Once you close a target that your opponent has not yet closed, every additional hit on that number scores points equal to its face value (15 points for 15s, 20 points for 20s, 25 for Bullseye). Your opponent can stop the bleeding by closing that number themselves.
To win, a player must close all seven targets and have a score equal to or greater than their opponent's. This dual requirement creates a deep strategic tension: do you focus on closing numbers quickly, or do you try to build a scoring lead first? How you balance offense and defense — and when you switch between them — is the central question of cricket strategy.
Frongello's Framework
In 2018, Andrew Frongello published a systematic analysis of darts cricket strategy that defined 17 distinct strategies based on three parameters:
- Lead threshold — How far ahead in points you need to be before switching from scoring mode (building a lead) to covering mode (closing numbers).
- Extra darts — When you cannot close a number with the remaining darts in your turn, redirect those extra darts toward scoring on an already-closed number instead of wasting them.
- Chase — Prioritize closing numbers your opponent has already closed, aiming to neutralize their scoring opportunities.
By simulating 20,000 games per matchup between every pair of strategies in a round-robin tournament — at both equal skill and with a skill advantage — Frongello arrived at four key findings:
- Score first, then cover — Build a point lead before shifting focus to closing numbers. S2 (“take any lead, then cover”) is optimal for equally skilled opponents. Purely defensive strategies consistently lose.
- Use extra darts when more skilled — If you are the stronger player, S6 (S2 + extra darts) outperforms S2 by extending the game and leveraging your accuracy advantage. Against an equal opponent, S2 edges out S6.
- Never chase — Do not waste darts trying to close numbers your opponent has already closed. Every flavor of chasing is dominated by S2 in every scenario.
- Weaker players want short games — When outmatched, play S2 and hope your opponent does the same. Building large point cushions (S3–S5) extends the game and favors the stronger player.
What We Did
Frongello's original analysis tested strategies at uniform accuracy (equal skill) and with one player at 95% relative skill. Real darts players miss in complex, target-dependent ways: triples downgrade to doubles or singles, doubles to singles, and some throws miss entirely. We extended his framework with realistic probabilistic skill profiles to test whether his conclusions hold across a full range of ability levels.
- Implemented all 17 Frongello strategies in Python using a single parameterized class, ensuring exact fidelity to the original definitions.
- Added realistic skill profiles that model how darts actually land: miss rates, downgrades (aiming for a triple but hitting a single), and face-value probability distributions calibrated against published research.
- Tested at 11 skill levels from beginner (MPR ~0.8) to pro (MPR ~5.3), spanning the full range of real-world darts ability.
- Ran 20,000 games per matchup across a full 30-strategy round-robin tournament at each skill level — over 89 million simulated games in total.
- Designed and tested 12 experimental strategies beyond Frongello's framework: Early Bull (E1), Honeypot (E2), Greedy Close-and-Score (E3), Adaptive Threshold (E4), Smart Aim (E5), Always Single (E6), Double or Nothing (E7), Score-Triple Cover-Single (E8), Phase Shift (E9), Score Surge (E10), Kitchen Sink (E11), and Finish Opponent-Closed (E12) — each exploring a different hypothesis about where the original parameter space might leave value on the table. E6 and E7 serve as baselines confirming that hit-type selection matters; the rest probe phase timing, aim adaptation, and hybrid approaches.
- Discovered and optimized a new strategy we call Phase Switch, which emerged from grid-searching over a two-phase scoring-to-covering transition.
- Introduced a bull difficulty multiplier (0.75×) to model the geometric reality that the bullseye is a much smaller target than numbered segments. This single change dramatically reshapes the strategy landscape — the previously dominant Early Bull (E1) strategy collapses, revealing a healthier competitive field.
Key Findings
1. Frongello's Core Findings Hold Under Realistic Conditions
All four of Frongello's central conclusions survive testing across 11 skill levels with probabilistic miss rates: score first then cover (S2 remains optimal among the original 17), never chase (the gap widens at higher skill), extra darts don't help at equal skill (they disrupt closing tempo), and weaker players want short games. His framework is sound — but it doesn't tell the whole story.
2. A New Strategy Dethrones the 8-Year Champion
Phase Switch — a novel two-phase approach that scores aggressively (13× threshold) then makes a single, irreversible transition to pure covering — beats Frongello's optimal S2 by +5.2 percentage points at pro level. That gap is comparable to the gap between S2 and the worst Frongello strategy. The key innovation isn't the threshold; it's the commitment. By never oscillating back to scoring, Phase Switch avoids the tempo-destroying mode-switching that dooms high-threshold strategies like S5. The result is remarkably robust: any threshold from 12× to 20× wins 55–56% against S2. How it was discovered →
3. A Hidden Assumption Was Masking the True Competitive Landscape
Frongello's model — and every prior simulation — treated the bullseye as equally easy to hit as numbered segments. When we introduced a bull difficulty multiplier (0.75×) to reflect the bull's smaller geometric target, the entire strategy landscape shifted. The Early Bull strategy (E1), previously #1 at 9 of 11 skill levels, collapsed to dead last at high skill — a drop of 8.4 percentage points at pro level. No other strategy moved more than ±1.3pp. E1's dominance was an artifact of an unrealistic assumption, and removing it revealed a far more interesting competitive field. Full analysis →
4. Three Distinct Strategic Regimes Emerge
Under realistic bull difficulty, the optimal strategy depends on your skill level:
- Low skill (MPR 0.8–1.5): Phase Switch dominates decisively, with E10 (Score Surge) and E9 (Phase Shift) rounding out the top tier. PS’s committed two-phase approach capitalizes on long, low-accuracy games.
- Mid skill (MPR 2.0–4.0): Phase Switch still leads, but E10, E3 (Greedy Close-and-Score), E2 (Honeypot), and S2 compete closely behind it. Strategy matters, but several approaches perform near-optimally.
- Pro skill (MPR 4.9–5.6): E3 and S2 overtake Phase Switch. At the highest skill level, E3 ranks #1 and S2 #2 — the tight closing tempo at pro accuracy rewards aggressive close-and-score tactics over Phase Switch’s heavy early scoring.
Five strategies — PS, E3, E2, S2, and E10 — form a remarkably stable top tier across all skill levels, though their internal ordering shifts with accuracy.
5. Bull Difficulty Equalizes Unequal Matchups
In every unequal-skill matchup tested, the weaker player gains win rate when bull is harder, while the stronger player loses it. The effect grows with skill: negligible at low MPR, but worth +2.15pp for the underdog at the highest skill gaps. Harder bull also changes which strategy is optimal in 83% of unequal matchup slots — real players facing a skill gap may need different strategy advice than Frongello's framework suggests.
6. At Low Skill, Strategy Barely Matters
When most darts miss their target, luck dominates. The gap between the best and worst strategies narrows dramatically at low MPR, and Phase Switch's advantage over S2 is effectively zero below MPR 3.4. For casual players, any reasonable strategy works. For competitive players, the right strategy is worth 5+ percentage points — the difference between winning and losing a match.
7. A Reinforcement Learning Agent Discovered Something No Human Designed
We trained a neural network to play darts cricket from scratch — no strategy rules, just wins and losses. After 19 versions, 12 training bugs, and millions of games against the hardest hand-crafted bots, it discovered a turn-order-dependent opening: always open 20 when going first, always open 18 when going second. None of the 30 hand-crafted strategies adapt to turn order. The agent also independently confirmed Frongello’s principles (score before cover, never chase) through pure trial and error — and revealed that opponent curriculum, not architecture, is the dominant variable in training game-playing agents. The full story →
Bull Sensitivity Sweep Complete
Full tournaments at all four bull multiplier levels (1.0×, 0.75×, 0.5×, 0.25×) are now complete. E1’s collapse is gradual (rank #2 at 1.0× to #16 at 0.25×, −5.9pp). Chase strategies (S10–S17) gain up to +5.7pp at extreme bull difficulty. PS remains #1 at every bull level, losing only 0.7pp. See the full analysis.
Explore the Results
30 Strategies
17 Frongello strategies plus 12 experimental approaches and the Phase Switch discovery. Each one defined by its parameters and tactical philosophy.
Tournament Results
Full 30x30 win-rate matrices at 11 skill levels. Interactive tables with rankings and head-to-head comparisons.
The Discovery
How Phase Switch was found through grid search optimization and why a two-phase approach outperforms single-threshold strategies.
Bull Analysis
How bull difficulty reshapes the strategy landscape. E1's collapse, ranking stability, game length impact, and the equalizing effect on unequal matchups.
The AI Agent
A reinforcement learning agent trained from scratch discovers a turn-order-dependent strategy no human designed. 19 versions, 12 bugs, 3 breakthroughs.
Strategy Advisor
Enter a game state and see what all 30 strategies recommend for your next throw. Interactive scoreboard with real-time consensus analysis.