bocch Strategy Guide

First, try the mechanics by hand

The distance a stone can move is “the number of stones in the 8 surrounding squares (either color).” Placing a stone also changes the numbers of the stones next to it. This is the single most important idea in the game.

Strong shapes

In AI-vs-AI games, the share of games won by “the side that had more of that shape than the opponent.” The center line is 50%; the further right the bar, the better the shape.

Black's 1st move (openings)

Evaluations from the improved “Strongest” (v1.7), which read each of the 14 possible first moves for 15 seconds (15–18 moves deep); mirror-image moves are averaged together. Only the two edge-stone moves, Snuggle and Corner Guard, are even; every move that pushes an inner stone two squares loses a little. For White's 2nd move, “Shoulder to Shoulder” (sliding an edge stone diagonally inward) is best no matter what Black plays. These readings are built into “Strongest” as its opening book.

Openings up to move 3

For all 196 combinations of the first two moves, the latest “Strongest” read Black's 3rd move 15–18 moves deep. On top of that, it read every possible 3rd move (3,006 in total) for 4 seconds each (11–14 moves deep) and sorted them into strong and weak. Scores are from Black's point of view: 0 is even, + favors Black (100 points ≈ a 78% win rate).

• Black's 3rd move: “bring in the other edge too” is the rule: in 159 of the 196 cases (81%), the best move slid an edge stone one row inward. If your 1st move brought in one edge, bring in the other edge on move 3 so both edge stones get to work.
• White loses big by playing anything but Shoulder to Shoulder on move 2: against Black's Snuggle, any other White move gives Black an advantage of +77 to +148 points. Shoulder to Shoulder is White's only good 2nd move.
• Move 3: moving a home stone just one row is safe: across all 3,006 moves, moving a home-row stone a single row averaged 43 points behind the best move, with almost no mistakes. Pushing a home stone two rows at once costs 77 points on average, three rows 90 points, and the mistake rate jumps to 8% and 18%. Long jumps in the opening tend to just hand your opponent footholds and time.
• Top-tier moves are rare: of the 3,006 moves, only 16% were best or good (within 20 points). Nearly half were dubious (60–120 points behind). Differences already appear by move 3.
• The main line is even: Black Snuggle → White Shoulder to Shoulder (on the edge opposite Black's) → Black brings in the other edge too, scores 0 points. The real fight starts from here.

Explore on the board (with variations)

Tap a candidate move to advance the board to it. The number to the right of each move is Black's estimated win rate after that move (latest AI evaluation). The mark on the left shows the gap from the best move: best, good, normal, dubious, mistake. For move 3, every legal move is listed.

The top row is White's home (Black's goal) and the bottom row is Black's home. Hover or focus a candidate to see its move on the board. From move 4 on there is no book; “Strongest” reads and plays on the spot.

Strategies (opening plans)

We had the latest “Strongest” play itself 26,000 times and grouped both sides' positions at move 8 by similarity: they fell into 5 types. Each diagram shows the typical shape of that type, drawn with 6 stones from the most common arrangement of your own stones at move 8 (opponent stones omitted; your home is at the bottom and the attacking side is on the left).

Matchups

The share of games the row strategy won against the column strategy.

How the stronger AI plays

Every time we made the AI stronger, the fashion in how it plays changed. These are the changes in the latest 26,000 games.

• All-Out disappeared and Two-Tier emerged: “All-Out,” which pushes nearly all home stones out and spreads them sideways, is now almost never played by the strong AI. Instead, “Two-Tier,” which keeps 2–3 stones at home and lines up the rest vertically in the middle rows, now makes up about 20% of games.
• The gap between Center Tower and Edge Attack widened: Center Tower beats Edge Attack 74% of the time (previously 67%). The stronger the AI, the better it draws out the power of a central cluster and punishes the weakness of the edge files.
• Half Tower became mainstream: the strong AI builds its tower in the center while keeping 2–3 stones at home. Full Tower, which uses almost every home stone, declined, and Half Tower now wins 4 points more often.
• Edge Attack and Fortress declined: in the learned values, stones left on the edges of your home row are a big minus, while stones in the middle of your home row are a plus. The strong AI slides edge stones inward early and keeps central stones at home.
• Stones left at home become defense: they block the opponent's landing squares (the “Home wall” tactic) and also give you room to build your tower one square away from the opponent.

About 6,000–12,000 players per type (both sides of each game are counted). The shape at move 8 also depends on how the opponent plays, so this does not mean “pick this type and you'll always get this win rate.”

Games as seen in 26,000 games between the latest AIs

Results counted from 26,000 games between copies of the improved latest “Strongest” (v1.7); the first 2–6 moves were random to vary the positions. The latest AI beats the AI from before the upgrade about 60% of the time.

• Stronger play means longer games: both sides block reaches early, so stones don't arrive easily. Half of all games end between moves 31 and 48.
• The first stone is huge: the side that gets a stone across first wins about two-thirds of the time (66.5%). Taking one from a solid defense is worth that much.
• Draws happen too, 6.5%: with deep reading, both sides avoid risky moves and positions tend to repeat.

Game length

Number of games by total moves of both sides (60 or more are grouped).

Win rate of the side that arrives first (by when it arrived)

Arriving first wins over 60% whenever it happens, but it's worth more the later it is: the side that arrives first at move 24 or later wins nearly 80% (77.6%). In the endgame, with fewer defenders left, that one stone is almost decisive.

Key squares

Win rate of the side that had its own stone on that square at move 12 (your home at the bottom; left and right averaged). Greener squares win more, redder squares lose more.

How to stop the opponent's reach

Black's stone (number 2) reaches the goal two rows up. It reaches in three directions — up, up-right and up-left — so there are up to 3 landing squares.

Advanced: the system of tactics

What strong players watch is not the numbers on their own stones but the “gap” of both sides' stones. Every tactic is built from this idea.

Diagrams show the whole board (top row is Black's goal, bottom row is White's goal). Stone numbers, reaches and reachable squares are all computed automatically by the rules.

Moves you must not play

In 26,000 games between the latest AIs (about 890,000 moves), we counted how often the side that played a given move went on to win. The overall average is about 50%. Even for the strong AI, these moves clearly drop the win rate.

• Creating the opponent's finishing move yourself (0% win rate): when the opponent already has one stone across, a move that changes the number of an opponent stone so it can now reach for the second (a backfire). It happened 2,047 times in 26,000 games, and the side that played it never won once. Once the opponent has one stone across, always check before moving: “Is there an opponent stone next to my landing square or next to my starting square?”
• Playing elsewhere while the opponent's finishing move remains (0% win rate): the opponent has one stone across and a stone that can reach for the second next move, and you don't block it. This includes positions with no way to block, but before it gets there, nipping the opponent's second reach in the bud comes first.
• Moves that leave the moved stone a loner (37% win rate): a “0” stone with no neighbors can't move. The side that played a move leaving its moved stone a loner won 13 points below average. Moves that leave a stone behind as a loner score 41%. Creating two or more loners at once drops it to 27%.
• Stopping one row short of the goal (46% win rate): moves onto row b (one row short of your goal) win a bit below average. A row-b stone can only arrive when its number is exactly 1, and if neighbors pile up it jumps past. Unless you can surely arrive next move, it's better to wait on row c than to enter row b.
• Jumping home stones far in the opening: in the analysis of all 3rd-move candidates, pushing a home stone 2–3 rows at once averaged 34–47 points worse than a one-row move and tended to be a mistake (details in “Shapes & openings,” openings up to move 3).

Mindset: once you're only defending, you're starting to lose

• The side that plays a blocking move wins 34%: blocking the opponent's reach is necessary, but if you're being forced to block, the initiative is theirs. The side that blocked the finishing move after the opponent's first arrival won 27%.
• That's why you “block and build”: a pure blocking move gives the opponent time to plan their next attack. When choosing a block, first look for a landing square that also creates a reach of your own (the “Attack and defense in one” tactic).
• If you have a reach too, you can sometimes ignore the opponent's: sides that left the opponent's reach alone while holding a reach of their own won 60%. Giving up one stone isn't an instant loss. Compare “the opponent gets one across” with “I get one across” and pick whichever is better.

The opponent's reach: block it or let it go?

When the opponent can arrive next move, we counted in 26,000 games how the Strongest AI decides.

• Always block the opponent's second stone: when the opponent already had one across, a move that could block but let it go never happened even once. The second stone is an immediate loss.
• The opponent's first stone can sometimes be let through: sides that could block but deliberately let it go won 69.8%. In particular, 95.9% when they already had one across themselves, and 72.3% when they had a reach of their own. If you'll get two across first, spend the blocking move on attack instead.
• When neither applies, “block and build”: with no stones across and no reach of your own, sides that blocked while creating a reach won 51.7%. Just blocking: 39.6%; making a reach without blocking: 46.9%; doing neither: 34.1%.

There's an order to blocking, too

• 1. Push-up (49.7% win rate): place your stone next to the opponent's reach stone to raise its number. The row it reaches changes, so it jumps past the goal.
• 2. Pull-away (43.7%): remove a stone next to the opponent's reach stone to lower its number.
• 3. Blocking the landing square (33.8%): it looks the most natural, but it's the worst way to block. The opponent's number doesn't change, so they just look for another landing square.

Moving an arrived stone back out

• Moving your arrived stone back out of the opponent's home is a last-ditch defense: the side that did it won 4.6% (7,969 times). 62% of those gave up their first stone to block the opponent's finishing move. If the AI plays this, it's already in trouble.
• Moving sideways within the goal row is a normal move: without losing an arrived stone, it changes neighboring numbers to block the opponent or open the way for your next stone (50.4% win rate).

Check these in order, every move

1. Does the opponent have a gap-0 stone (a reach)?If so, block it. Use a push-up or pull-away to move its gap off 0.
2. Will this move backfire?Check that there's no opponent gap +1 stone next to your landing square, and no opponent gap −1 stone next to your starting square.
3. Can you block and create your own reach at the same time?Look for your own gap +1 stone next to the landing square (attack and defense in one).
4. Can you make reaches in two separate places?A double reach clustered in one spot can be broken with a single move. The real thing is two separate places.
5. Will any of your stones become a loner?Sides with many stones that have only 0–1 neighbors drop to around a 45% win rate.