The Shape Reader

Ranks first at every one of eleven skill levels we tested. At matched pro skill it averages 62.8% across the field, seven points above the next-best strategy.

Everything up to this point has been a list of strategies. Seventeen from Frongello's original paper, twelve of our own, plus the two-phase commit called Phase Switch. Each one is a short recipe: check a priority, pick a dart. Thirty rows in a table.

This page is about a different kind of object. One of the strategies we built later beats every strategy we benchmarked against, at every skill level we tested. We called it the Shape Reader. In the full round-robin against the thirty originals, it ranks first at all eleven skill levels, from novice (MPR 0.8) through elite pro (MPR 5.6). At matched pro skill (MPR 4.9) it averages 62.8% across the field and never loses a head-to-head — its worst matchup is S13 at 51.1%. The margin over second place widens with skill: about 1 point at novice, more than 7 at pro, almost 10 at elite. It reads as a stack of four checks rather than another row, each one inherited from something earlier in the research, each one with a narrow job. Only the combination wins.

The rest of this page explains what the stack is, in the order the strategy runs it.

What it does, every dart

As long as your opponent isn't racing through closes without scoring, the strategy runs a short priority list on every dart.

  1. Score on a number you've closed that they haven't.
  2. If there isn't one, finish any of their closed numbers you're one mark away from, with a single.
  3. If there isn't one of those either, and only if they've closed more numbers than you have, chase their highest.
  4. Otherwise, open a new one.

The list runs its full length in the opening turns, when lanes don't exist yet and both players are still claiming numbers. From the third turn on, once somebody has closed a number solo, the first clause takes over. It fires on more than 99% of darts for the rest of the game. Clauses two through four are almost entirely an opening-game story.

That list keeps running as long as your points lead stays within a wide envelope: 13 times the highest number your opponent hasn't yet closed. With 20 still open on their side that's a 260-point envelope, which basically means always, since nobody has a 260-point lead in the opening third of a game. As numbers close, the envelope shrinks with the board. When the game enters its closing stretch, meaning three or fewer numbers left to close and only a handful of marks between you and a full board, it collapses to zero. From that moment on, any lead at all flips the strategy into cover-and-finish mode.

Walk a scene

Imagine the board part-way through. You've closed 20. They've also closed 20, so it's dead for both of you. Neither side scores on it any more. They've closed 18 too. You have two marks on 18 yourself. One more and it's closed for you. Nothing else on either side. Scores are tied at thirty.

Walk the list. Do you have a lane — a number you've closed that they haven't? No. 20 is dead. So clause one doesn't apply. Are you one mark from closing one of their closed numbers? Yes: 18. So clause two applies, and the answer is a single on 18, because here you want a finish and the triple is the throw for points or mass progress.

Clause three doesn't get consulted. Neither does clause four. The moment clause two fires, everything below it is skipped.

Why this order

That single-on-18 move is doing something specific. A one-mark finish on a number your opponent has already closed is the cheapest durable move on the board. One reliable single, and that lane of theirs is permanently dead. They never score on it again. You've already paid for two of the three marks; you're buying the last one.

Opening a new number is a three-mark investment from scratch. Done with triples, it's one good turn; done with singles, it's three good turns. Either way, the opponent's existing lanes keep bleeding points while you spend those darts.

The list prefers the finish because the finish is already half-paid-for work. Opening a new number is starting from zero. Any time both are on offer, the half-finished thing is the better buy.

The chase that Frongello warned against

You may have noticed the priority list does something the strategies page told you not to. Clause three is a chase. Every page of the Frongello research agreed: chasing is generally suboptimal. So why is it here?

The difference is what "chasing" means. The Frongello chase strategies made chasing their top priority on every dart. Any opp-closed number you hadn't finished, go close it now, before anything else.

This one is third on the list. It fires only when you have no lane, no cheap one-mark finish, and they've closed more numbers than you have. It also needs your points lead to be inside the wide scoring envelope, which, in the opening game, it almost always is. Put those conditions together and the chase lives in a narrow slice of the game: the opening turns, when neither side has settled on their own numbers yet. Past the third turn, once lanes exist, it barely fires at all.

When covering actually fires

One question you might be carrying: if the priority list is all about scoring and finishing and chasing, when does the strategy actually cover? Every clause on the list sits in the score path.

The honest answer is almost never in the first half of the game. The scoring envelope is so wide — 13 times the highest number your opponent hasn't closed, which basically means 260 points while 20 is still open — that nobody builds that kind of lead during the scoring phase. Cover fires on roughly one in twenty darts in the early and middle phases.

Then the closing stretch hits: three or fewer numbers left to close, nine or fewer marks between you and a full board. The envelope collapses to zero. We call this the phase gate — the on/off switch between scoring mode and covering mode. From that moment on, any lead at all flips the strategy to cover, and cover takes over, running on seven of every ten darts for the rest of the game.

Reading the opponent

Every section so far has begun with the same qualifier: as long as your opponent isn't racing through closes without scoring. Time to explain what that condition is.

It's a cheap pattern match on the opponent's board. They've closed at least four numbers, and their score is unusually thin for how many they've closed: fewer than about twelve points per close. In plain terms: lots of closed numbers, not many points. We call this the pattern detector.

When the detector trips, the strategy makes one decision: are you ahead or behind on points? If you're behind or tied, nothing changes. Keep scoring on any lane you have, or open one. Points are what you're short on, so catch up.

If you're ahead, the priority flips to denial. First, finish any of their closed numbers you're one mark away from. Next, chase their highest closed one. Last resort, cover a shared-unclosed number. The scoring logic goes away, and the aim shifts from running up the score to shutting them out.

The four layers

Pull the camera back and the whole strategy is four layers stacked on the same board state. First, the pattern detector asks whether the opponent is racing through closes without scoring. Second, the phase gate checks whether the game is in its opening stretch or its closing one. Third, a lead check looks at whether your points lead is inside the envelope or past it. Fourth, the priority list picks what you actually aim at: score on a lane, finish a cheap one, chase their highest, or open a new number.

Every branch, every firing rate, every trace signature sits on the decision tree page. Click a node to see its condition, its visibility, and how often each clause fires against a realistic opponent pool.

What this is and isn't

One last thing. The champion label belongs to a specific fight: the full thirty-strategy round-robin, at matched skill for both sides, across eleven skill levels with 20,000 games per matchup. That is where the Shape Reader ranks first everywhere.

Matched skill is the important qualifier. In asymmetric matchups where the Shape Reader is the weaker player, skill dominates strategy and it loses — the same way any strategy in this field loses to a much stronger opponent. That is not a claim about the strategy, it is a claim about dart accuracy.

The architecture is also conditional on the opposition. Against an opponent whose board never trips the pattern detector, the whole first layer sits inert. The strategy is a priority list with a wider envelope, nothing more. The win rate holds up in the tested field because the field is diverse enough that the detector fires against a meaningful slice.

We didn't solve cricket. We found a way to think about it in layers that turn out to compose well.