To beat a checkers computer, check its complete reply before committing your move. Look for compulsory captures, make exchanges that improve the final position and convert an advantage with a concrete plan. Trying to surprise the bot with a move that merely looks unusual is less useful than proving it has no satisfactory response.
On Pure Checkers, the three difficulty levels share one movement engine and evaluation. They differ in how far they are allowed to search. That makes them useful practice opponents: you can learn what a short search misses, then repeat the same kind of calculation against a deeper one. It does not give you a move that wins every game from the starting position.
What the bot actually looks at
The search begins with the legal moves in the current position. For each candidate, it explores replies and further continuations. It treats its own choices as attempts to improve its score and your choices as attempts to reduce it. This is minimax search; alpha-beta pruning skips branches that cannot improve a decision already established in that part of the tree.
The bot searches progressively deeper. It keeps results only when an entire pass over its candidate moves is complete. That avoids comparing one move searched deeply with another whose analysis was cut off halfway through. A real time limit still means the final completed depth can be below the difficulty level’s ceiling.
Depth counts completed turns. If a piece must jump again, the search stays with that piece and the same player’s turn until the chain finishes. It also examines choices at a branching capture. Counting each hop as a new turn would misread a basic combination, so “Easy depth two” does not mean that the third hop of your triple jump is invisible to it.
Easy, Medium and Hard
These are the current registry settings for the bots you can choose from the lobby. Their names and starting ratings identify the app’s opponents; the depth limits describe their search. The ratings have not been independently calibrated as equivalent human tournament strengths.
| Level | Opponent | Maximum completed-turn depth | Starting app rating |
|---|---|---|---|
| Easy | Pip | 2 | 600 |
| Medium | Marge | 4 | 1000 |
| Hard | The Colonel | 6 | 1400 |
All three obey compulsory captures, backward captures by men and flying-king movement. A lower level does not receive an easier ruleset. If an unexpected backward jump costs you a piece, changing opponents will not remove that rule; use the movement guide to check the position instead.
Live search has a 350-millisecond thinking budget, with additional limits around background work. The computer is not guaranteed to spend that whole interval or reach its maximum depth. A forced move can be selected immediately. If background search fails, the server can use a legal fallback so an unlimited-time game is not left waiting forever.
What its score means
At a nonterminal search boundary, the evaluation starts with material: 1 for a man and 5 for a king. An ordinary man gains 0.1 for each row advanced from its own back edge. A piece in one of the four middle columns gains 0.05. The bot adds its own pieces’ values and subtracts the opponent’s.
The calculation uses exact integer units internally at 0.05 precision, then returns the familiar score scale. This keeps equivalent positions tied instead of letting tiny floating-point differences decide between them. When multiple best replies have the same score, the bot can select among them randomly.
A score of 1 is not a predicted one-piece win, and it is not a probability. A king’s value of 5 is a heuristic chosen for this game, not a universal exchange rate. The search also recognises actual terminal wins, losses and draws. Those results matter more than an attractive material estimate at a position that has not finished.
See the search horizon in an opening exchange
After red plays 11-15, black can offer 24-19. Red must capture 15x24, and black then recaptures 28x19. At a depth of two turns from black’s decision, the search reaches the position after red’s capture, before black’s recapture. At greater depths it can see beyond that temporary material loss.
The following recorded experiment uses the current engine, a 150,000-node limit and no wall-clock cutoff, so its completed depths can be compared without depending on machine speed. The scores are from black’s perspective. They describe this experiment, rather than promising that every live game will return the same reply.
| Requested and completed depth | Score for 24-19 | Highest-scoring replies |
|---|---|---|
| 2 | -1.45 | 24-20, 23-19, 23-18 or 21-17 |
| 4 | 0.15 | 24-19 |
| 6 | 0 | 24-19 |
The complete experiment record includes all scores, budgets, node counts and source hashes. The first-move guide has a board that steps through this exchange. Notice that the depth-four and depth-six scores differ too: looking farther can change the evaluation after the obvious recapture has already been found.
This is a useful example of the horizon problem. A search evaluates the position where it stops, even when an important reply comes just beyond that boundary. It does not follow that every delayed sacrifice fools Easy. Another legal response may avoid your idea, or a forced chain may bring the decisive continuation inside the search’s current turn.
Practise a shot with a forced reply
In this constructed position, try 15-18 against Easy. Black’s only capture is 22x15. Red then has 10x19x26, taking two men after offering one. The point is to calculate the complete forced sequence; it works because of the board’s geometry, not because Easy agrees to make an illegal or arbitrary mistake.
- Square 6: red man
- Square 10: red man
- Square 14: red man
- Square 15: red man
- Square 22: black man
- Square 23: black man
- Square 32: black man
You play red; the easy bot (searches up to depth 2 with a time limit) answers for black. Click a piece, then a highlighted square.
The supporting men are part of the combination. Red on 14 prevents another black landing, and red on 6 stops black from continuing over the intended attacker on 10. Remove either safeguard in your mental calculation and you must recheck the entire idea. A sacrifice copied without its supporting pieces is a different position.
After the double jump, stop and count what remains. You have gained one man through the exchange, but black still has a piece and the game continues. The double-jump guide explains each hop and the guard on 6. Use it if the bot’s forced reply is easy to see but your own continuation is not.
Turn a material lead into a useful position
Trading pieces when ahead can reduce the number of threats you need to manage. It can also release an opposing man toward promotion or leave your own pieces blocked. Before exchanging, count the final material, identify the next player to move and check whether either side gains a king or a forced capture.
When behind, seek concrete chances instead of refusing every trade. A trade that forces the opponent into a capturable position can help you, while preserving all your pieces in a blocked formation may not. The practical test is the position after the full sequence, not a slogan about whether the leading side should always exchange.
With flying kings, a material lead often needs coordinated control of diagonals. Chasing the opponent with one king while the other does nothing can repeat the same position until a draw. The two-kings lesson gives a verified winning plan in which each king has a separate job and every defensive reply is covered.
A practice routine that produces useful feedback
Choose one familiar first move and play several games at the same level. Before each decision, name your captures, the opponent’s likely captures after your move and the final landing of any chain. If you cannot finish that calculation, prefer a position whose replies you can explain over a speculative offer that depends on the bot cooperating.
After a loss, identify the first move where your expected reply differed from the actual one. Was a backward capture missing from your calculation? Did you count a multi-jump as several turns? Did you stop at a temporary material gain before the recapture? Those questions point to a specific exercise you can repeat.
Use My games in the recent-games list, then choose Replay beside a saved result. Step to the position before the exchange you misunderstood. Predict the next capture and its final landing before advancing the replay, then compare your prediction with the recorded move. Stepping backward lets you check which supporting piece or empty square you overlooked without starting another game.
Keep the comparison small enough to explain. Instead of changing your first move, difficulty and overall plan together, repeat one opening at the next level and watch the first exchange. If the reply changes, reconstruct the board after both possible continuations. You will learn more from understanding that one difference than from collecting a list of bot moves without their positions.
When a result surprises you, separate the legal-move question from the quality question. First confirm that the move was allowed under this site’s rules. Then ask what the bot saw after the exchange and what your own calculation missed. A legal move can still be weak, while an unfamiliar backward capture can be entirely correct.
Move up a level when you can consistently follow the exchanges in your current games, rather than because one lucky trap worked. The site FAQ covers the available playing options. A win against a practice bot is a result from that game; a sound calculation is something you can reuse against the next opponent.