Open source · In development

The cribbage engine
that shows its work

Muggins is an open-source cribbage AI with provably optimal discards, a trained neural pegging model, and transparent RNG. No ads. No rigged deals. Just strong, honest cribbage.

69.5%
Win rate vs Myrmidon
+1.21
Points/hand advantage
95.1%
Neural model accuracy
10,000
Games benchmarked

Tested against the best
open-source cribbage AI

10,000 games against Myrmidon, the strongest agent in the Moulton cribbage RL framework. Alternating seats, seeded RNG, Wilson confidence intervals. Reproducible and open.

Matchup Win Rate 95% CI Per-Hand Differential
Muggins vs Myrmidon 69.5% 68.6% – 70.4% +1.21 pts/hand
Muggins vs Random 98.8% 98.2% – 99.2% —
Myrmidon vs Random 96.7% 95.8% – 97.4% —

Two systems, one engine

Provably optimal discards and learned pegging. The exact EV calculator evaluates all 15 possible keeps across all 46 starters. The neural model handles the rest.

Σ

Exact EV Calculator

Every possible keep scored against every possible starter. Position-aware endgame strategy. Not sampling, not heuristics. Exact combinatorics.

ψ

Neural Pegging

116-dimensional feature vector encoding threats, opponent inference, and board position. Trained on millions of games. 95.1% prediction accuracy.

#

Transparent Dealing

Cryptographic RNG with verifiable deal hashes. Every deal can be independently verified. The cards aren't rigged — and we can prove it.

Δ

Position-Aware Strategy

Endgame go-out calculations, offensive and defensive modes based on board position. Adapts strategy to the score, not just the cards.

≡

Coaching Engine

Every discard and pegging decision analysed. Named mistake categories, EV explanations, and practical guidance. Learn from every hand you play.

⊥

Open Source

Engine and trained weights published. Benchmark methodology documented. Build on it, challenge it, improve it. Think your engine is better? Bring it on.

Discard Analyser

Select 6 cards. Muggins evaluates all 15 possible keeps and shows you the optimal discard with full EV breakdown.

Select your hand (6 cards)
0 / 6 selected
Position
All 15 keeps ranked by expected value

What's coming

Done

Core engine & benchmarks

EV calculator, neural pegging model, position-aware discards, Moulton benchmark.

Active

Web discard analyser

Full-featured discard analysis tool. Use it during your real-world games.

Planned

Play Muggins online

A web opponent at play.muggins.app. Test your skills against the engine.

Planned

iOS & Android apps

On-device AI. Premium, one-time purchase. No ads, no subscription.

Planned

Open-source engine release

Python engine, trained weights, benchmark suite. Challenge accepted.

Think your engine is stronger?

We've published our numbers. We've shown our methodology. Now we're inviting every cribbage engine — commercial, open-source, or private — to do the same.
✓ Your format, your rules. We'll adapt to any reasonable benchmark protocol.
✓ 10,000 games minimum, alternating seats, seeded for reproducibility.
✓ Full results published — win or lose. No cherry-picking.
✓ DLL, API, stdin/stdout, manual relay — whatever interface your engine supports.
Cribbage Pro, HALSCRIB, or anyone else — the table is open. Reach out at [email protected]

Get notified at launch

No spam. One email when the app ships and one when the web opponent goes live.