Each bot at the table has a configurable strategy. Click the ⚙ next to any bot
to open its settings. Pick a preset, then fine-tune the knobs if you want.
Presets
Expert Strongest
Pure trained neural network. Beats every other preset and most humans. Use this when you want a real challenge.
Balanced
Pure heuristic bot. Uses rules of thumb (lead bowers, save trump, follow suit) with no card-counting memory. A solid baseline opponent.
Aggressive
Pure ML with extra bid noise. Will overbid and underbid. High variance, fun to play against.
Loose
Mostly rule-based, forgets which trump and aces have been played, makes some mistakes. Easier opponent.
Beginner
Pure rules, no memory, lots of random mistakes. Great for learning Pepper for the first time.
Knobs
Bid (R → M)
Bidding strategy. R (0) means pure rule-based bidding — conservative, predictable.
M (1) means pure ML model — uses the trained network's bid recommendation. Values
in between are a probabilistic mix.
Play (R → M)
Card play strategy. Same scale as Bid: R = rules, M = ML. The ML model
plays significantly stronger, especially in mid-trick decisions.
Slop (0 → R)
Mistake rate. 0 = no mistakes. Higher values add randomness to bids and plays.
Use this to make a bot easier without changing its strategy.
Trump memory
How much trump tracking the rules-based fallback uses. Full remembers every
trump played. Bowers only tracks just jacks. None ignores it.
Only matters when Bid/Play are below 1.
Aces memory
Whether the rules-based fallback tracks which off-suit aces have dropped.
Remember = will lead kings as new tops. Forget = doesn't track.
Table Controls
⚙ gear — opens a bot's settings panel. The × closes it.
⏸ Pause — stops bots from auto-playing. ▶ Step advances one action at a time.
↻ New Bot — replaces the player with a fresh bot.
👤 Open Seat — removes the bot so a human can take the seat.
👴 Ask Grandpa — get ML advice on what to do (twice per game).
🤖 Autoplay — let the bot take over your seat for a hand.