Testing locally

The macpoker CLI plays full games on your machine using the same engine the tournament runs, in the same duplicate format: one game per seat over the same decks, every bot shifted one seat each game, and a fresh instance of every bot for each game.

Play a game

macpoker play main.py house:call house:random --deals 50

Every argument after play is one seat at the table. Pass 2 to 9 seats in any mix of:

• a path to one of your bot files: main.py, v2.py, ../old/bot.py
• a built-in house bot, written with the house: prefix

House bot	Style
house:call	Calls everything, never raises. The validation opponent.
house:checkfold	Checks when free, folds to any bet.
house:allin	Shoves every chance it gets.
house:random	Mixes calls, folds and odd raises.

Options

Flag	Default	Meaning
--deals N	50	Hands per game, each from a fresh deck.
--games N	one per seat	Games in the set. Fewer than one per seat leaves the cards unbalanced.
--seed S	local	Deck seed. Same seed, same cards.
--stack N	200	Starting stack each hand (blinds are 1 and 2).
--time-ms N	30000	Starting time bank per bot.
--increment-ms N	100	Clock increment per hand.
--subprocess	off	Run bots as separate processes over the real wire protocol.
--history FILE	off	Write the full hand history as JSON.

Comparing your own bots

Any seat can be one of your files, and you can pass several at once:

macpoker play v1.py v2.py aggressive.py house:call --deals 100 --seed fight

Because every bot plays every deck from every seat over the set, all of your variants play the same cards from the same positions. The final chip totals are a fair head-to-head comparison, not deck luck. Keep old versions around (v1.py, v2.py, …) and A/B them: pin --seed while you iterate so runs repeat exactly, then change the seed to confirm an improvement holds up.

A bot can also play its own copy:

macpoker play main.py main.py --deals 200

Each seat gets an independent instance, and self-play should net close to zero. A large swing means your bot behaves inconsistently between positions.

Reading the result

Output

100 hands (2 games x 50 hands, same decks, seats shifted each game)
bot                                chips    mbb/hand  verdict
main.py                             +140       700.0  OK
house:call                          -140      -700.0  OK

Chips is the score. mbb/hand is milli-big-blinds per hand, a rate that is comparable across game lengths. Chips and mbb/hand are totals over the whole set. A verdict other than OK names the game it happened in and means your bot stopped playing for the rest of that game: see clocks and verdicts.

Debugging

• print() from your bot goes to stderr and is shown after the game if it had problems.
• --subprocess catches an entire class of bugs (protocol misuse, crashes on startup, slow imports) exactly as the tournament would.
• --seed makes runs reproducible while you iterate.