Mancala Strategy: Extra Turns, Captures and the Endgame Sweep (2026)

1. The board this build actually uses

Stone Pits is Kalah, the best-known member of the mancala family. You and the computer each own six pits and one store. Every pit starts with four stones, so there are 48 stones in play, and the stores start empty. You are the bottom row; the computer is the top row. The bottom row reads left to right as pits 1 to 6, with your store on the right. The computer's row mirrors it.

Internally the engine numbers the bottom pits 0-5, your store 6, the computer's pits 7-12 and its store 13. This guide uses the on-screen numbers 1-6 for your pits and adds the engine index only when a counting example needs it.

A move is simple. Pick a non-empty pit in your bottom row, lift every stone from it, and drop one stone into each following pit, counter-clockwise. You go right along your row, into your store, then left along the computer's row, and back to your first pit. The computer's store is skipped: a stone never lands there. The move ends when your hand is empty.

Two events decide the game:

• Extra turn. If the last stone lands in your own store, you move again.
• Capture. If the last stone lands in an empty pit of yours, and the pit directly across the board is not empty, you take your landing stone plus every stone in that facing pit. All of them go to your store.

When either side has no stones left in its pits, the game ends immediately. The remaining stones on both sides go to their owners' stores, and the fuller store wins.

There is no hidden information and no dice. The board is fully visible, the stone count is exact, and the only randomness in the build is the small noise added to the Easy and Medium computer's move scores. That means a loss is usually a decision, not bad luck. It also means counting is a real skill here: the pit numbers, not the dots, are the truth. The dots cap at 12, so a pit holding 14 stones still shows 12 dots.

2. The extra turn: the third pit is the opening tempo move

On the opening board every pit has four stones. Only one of your six pits reaches your store: pit 3, the third pit from your left (engine index 2). Its four stones land in pits 4, 5 and 6, then in your store. You move again.

The other five opening moves do not reach the store:

So the opening has a clear tempo move. The shipped medium search takes it: on a fresh board its best move is the pit that lands in the store. That is not a rule, just the search agreeing that a free stone and another move are worth more than the alternatives.

The opening is only the first example. The general rule is to count the stones in the pit you are considering and count the landings forward, remembering that the computer's store is skipped. If the last landing is your store, you keep the turn. A pit needs exactly enough stones to reach the store: 1 stone in pit 6, 2 in pit 5, 3 in pit 4, 4 in pit 3, 5 in pit 2, or 6 in pit 1. If a pit has already gone around the board, add 13 for each full circuit — a lap has 13 landings because the computer's store is skipped, not 14.

Extra turns are not a rare bonus. In the 120-game medium sample, 24.1% of all moves were extra turns. Roughly one move in four keeps the same player on the board. That changes how you read a position: the turn does not automatically alternate, and a player who can chain store landings can pull ahead in tempo before the opponent gets another choice.

The value of an extra turn is not just the stone it banks. It is the second decision. Use it to set up a capture, to empty a pit safely, or to take a better move than the one you would have had. The mistake is chasing the extra turn when the sowing feeds a capture or leaves a loaded pit facing an empty computer pit. Compare the free move with the board it creates, not with the stone count alone.

3. Captures: one stone becomes a whole pit

A capture is the only way to take stones out of the computer's pits before the final sweep. The condition has two halves and both must be true:

1. Your last stone lands in an empty pit on your side. After the drop that pit holds exactly one stone.
2. The pit directly across the board is not empty.

Then you take your landing stone plus every stone in the facing pit, and all of it goes to your store. The landing pit is emptied again as part of the capture, so the same pit can become a capture cup more than once in a game.

A concrete shape: suppose your pit 3 is empty and the computer's pit directly across from it holds 5 stones. A 2-stone pile in your pit 1 sows one stone into pit 2 and the last stone into pit 3. That is a capture of 6 stones: your landing stone plus the 5 across the board. The stones go straight to your store and cannot be captured back.

If the facing pit is empty, there is no capture. You simply leave one stone in your pit. Check both halves before you count the gain — the empty-pit half alone is not enough.

Captures are common, but large captures are not. In the same 120 medium games there were 631 captures, and they broke down like this:

The average capture was 3.54 stones, and 14.1% of all moves ended in a capture. The shape of that table matters more than the average. Almost six in ten captures are the minimum two-stone grab: one stone across the board plus your landing stone. The big 8-12 stone haul is real but rare. Do not spend the whole game building a ten-stone trap that never arrives; the constant threat of a two- or three-stone capture is what actually shapes the board.

Each captured stone is a two-point swing: you gain one in your store and the computer loses one from its pit. A three-stone capture is therefore a six-point swing in the final score. That is why captures decide close games even when the stone count looks level.

4. The shape of a safe board

The capture rule is asymmetric, and that asymmetry is the whole defensive game. You capture when your empty pit faces a loaded computer pit. The computer captures when its empty pit faces a loaded pit of yours. So a big pile on your side is dangerous only when the pit directly across from it is empty on the computer's side.

Before every move, scan the facing pairs. If one of your loaded pits is opposite an empty computer pit, adding stones to that pile is giving the computer a bigger target. If the computer can land in its empty pit on the next turn, it may take the whole pile. The safer move is somewhere else, even if it looks less active.

The same rule makes your own empty pit an asset when the computer's facing pit is loaded. That is the capture shape from the previous section: your empty cup, their full pit, and a source with the right count. The window can close. If the computer sows into your empty pit before you use it, the cup is no longer empty and the capture is gone. This is why captures often happen immediately after a pit empties, before the opponent can repair the shape.

Two practical habits follow:

• Count the source, not the target. You cannot choose where the last stone lands directly; you choose the source pit. Count its stones forward, skip the computer's store, and see whether the last landing is your empty cup. If it is, the capture is available now.
• Do not confuse a loan with a payment. Seeds you sow into the computer's row are not points for the computer yet. They sit in its pits, where you may capture them later. They become computer points only if the game ends with them still there. That is why a quiet move into the computer's side is not automatically bad. It is a loan of custody, not a gift.

The endgame makes that loan call due. Until the sweep, a stone on the computer's side is still contestable. After the sweep, it is gone. The next section is about not letting that happen on the wrong turn.

A single stone in pit 6 deserves a special mention. From pit 6, one stone lands directly in your store, so it is a guaranteed extra turn. It is a tempo button, not a scoring move: you bank one stone and keep the turn. Use it when the extra decision is worth more than the stone, especially near the end when one more move can set up a capture or avoid emptying your side at the wrong moment.

5. The endgame sweep: count before you empty

The game ends the moment either side has no stones left in its pits. The build then sweeps both sides: every remaining pit stone goes to its owner's store. The stores are added up, and the higher store wins.

The asymmetry from captures returns here, but bigger:

• If you empty your side, your final score is your store, and the computer adds every stone left in its pits. You get nothing more from the board.
• If the computer empties its side, your remaining pit stones go to your store. The computer gets nothing more from the board.

So emptying your own side is a bet: your store must already be bigger than the computer's store plus every stone in its pits. Do that addition before the move, not after. If the sum says you are behind, keep a stone on your side and look for a capture instead.

Emptying the computer's side is the opposite kind of move. Your remaining pits become points, so it is usually the trigger you want — but only if the resulting total is actually ahead. The sweep does not care who triggered it; it only adds the stones to their owners.

A capture is a two-point swing per stone, so the last few captures can change the endgame arithmetic faster than a quiet move. For example, suppose your store has 24, the computer's has 20, you have 3 stones left in your pits and the computer has 6. If your next move empties your side without adding to your store, the sweep makes it 24-26 and you lose. If instead you capture 3 stones first, the stores swing to 27-17; even after the computer's remaining 6 are swept, the final is 27-23 and you win. The numbers are an illustration, but the arithmetic is the real decision: store plus remaining pits, compared before the trigger.

The 120-game medium sample averaged 37.6 moves, with a maximum of 64. Long games are normal. There is no clock, no sudden death and no bonus for finishing early. Do not force the sweep because the board looks messy; force it only when the count says it helps you.

The computer's own evaluation uses the same idea. Its evalFor scores a stone in the store at 1.0 and a stone still in a pit at 0.7, because pit stones can still be captured. That is the sweep risk in one line: the build's search already prefers the bank. If you are choosing between banking a stone and leaving it in a pit, assume the computer values the bank more, and check whether your move changes that.

6. Difficulty, hint, and how to practice

The three difficulty chips change the search, not the rules:

Noise is added to the move score. Easy can miss the best pit, but it never plays an illegal move or a move outside the rules. Hard has no noise, so for a given seed it is deterministic.

The Hint button always runs the hard search, even if you are playing Easy. It highlights the pit and gives a reason: an extra turn, a capture, or the strongest quiet move. It does not move for you. That makes it a useful check on your own plan: decide first, then press Hint and compare the reason, not just the pit.

Two URL parameters help practice. ?level=easy, ?level=medium or ?level=hard sets the starting difficulty, and ?seed= reproduces the computer's tie-breaks. Changing the difficulty starts a new game. The record stored in localStorage under pn.stone-pits.v1 tracks wins, losses, draws, streak and best margin; it does not feed the AI.

The 120-game medium self-play is also the honest baseline for the first-player question. In a mirror match between the same policy, the south seat (the first player) won 51, lost 64 and drew 5. That is not a promise about your win rate against the computer or a claim that moving first is bad in general. It is a warning that the opening extra turn is not a free win in this build: the second player gets information and position in return. Treat the opening as a tempo move, not as a solved advantage.

A practice routine that uses the build:

1. Play Medium. Before every Hint, say your move and your reason out loud.
2. Press Hint and compare the reason. If it says "extra turn" and you missed it, recount the source pit. If it says "captures", check the facing pit before you count the gain.
3. When your reason matches the hint's reason for ten moves in a row, move to Hard.
4. Use ?seed= to replay the same opening with a different plan. The computer's tie-breaks repeat, so the comparison is fair.

FAQ

How many stones are in a Stone Pits game?

48. Six pits per side times four stones per pit is 48, and the two stores start empty. The build conserves all 48 stones until the final sweep.

Which pit gives the extra turn at the start?

Pit 3, the third pit from your left (engine index 2). Its four stones land in pits 4, 5 and 6 and then in your store, so you move again. No other opening pit reaches the store: the other five land in pits 5, 6, or the computer's row.

Why did the computer move twice in a row?

It landed its last stone in its own store. That is the extra-turn rule, and it applies to both sides. The status line will say "Extra turn" when it happens.

Does landing in any empty pit capture?

No. The pit must be yours, and the pit directly across the board must contain at least one stone. If the facing pit is empty, you just leave your one stone there. Check both halves before counting a capture.

Can I capture the computer's store?

No. Stores are never sown into and never captured. A stone that reaches a store is final; only the stones still in pits can be captured or swept.

Is there a time limit?

No. The clock counts up and nothing forces a move. The game ends only when one side's pits are empty.

What does the Hint button actually do?

It runs the hard search — 12 plies, 160 ms budget, no noise — and highlights the best pit with a reason. It is the same policy the Hard computer uses, not a scripted answer. It never plays the move for you.

What happens when one side runs out of stones?

The other side's remaining pit stones go to their owner's store, both stores are compared, and the higher store wins. If you empty your own side, you get no more stones from the board; if the computer empties its side, your remaining pits are banked for you.