Inside the Deck Lab
Can a computer discover
a deck worth playing?
Give it a card pool, let it build decks, and see what survives repeated games. Then ask whether the survivors are actually getting stronger.
The Lab is the part of MtgBattles where the computer does the deckbuilding too. You can watch and inspect the results without an account.
Build. Play. Change. Check.
Start without established deck lists
Blind self-play begins with a fixed legal card pool and printed card information, such as mana costs and rules text. Human card rankings, tournament results, and familiar deck recipes are not its starting instructions.
Let Forge play both sides
Game results help estimate which decks are promising. The search keeps survivors and proposes changed lists. A parent is an existing deck; its child is a proposed version. A mutation might replace cards, change copy counts, adjust mana, or restart with a fresh deck.
Test improvement against a fixed yardstick
A deck can rise in a changing population without getting better against outside opponents. Separate validation games face fixed reference decks. In the evolution league, these checks measure progress without selecting the survivors.
The deck lists evolve. Forge’s computer player stays the pilot. Lab scores describe a particular experiment and do not change the public Arena or Daily Draft ratings.
Two related questions
From evolving decks to testing the changes
Does the population improve?
The Unlimited League V2 archive records a search through cards from Unlimited Edition under the experiment’s legality rules. Inspect deck families, past generations, and independent validation to see how it developed.
Explore the league archive →Which changes help?
The balanced mutation study gives each kind of change equal opportunities. It compares children with their parents against fixed opponents. Discovery games find promising changes; separate confirmation games ask whether that apparent gain repeats.
See the current study and its status →Read the result with its unfinished work.
- What does a gain mean?
- A gain of +5 percentage points means the changed deck’s measured score was five points above its parent’s score against the study’s opponents. It is not five extra rating points or a universal bonus for that kind of edit.
- What has the experiment established?
- Read observed gains alongside missing-outcome bounds and the independent confirmation column. A promising discovery result that disappears in confirmation is evidence of uncertainty. A range spanning both loss and gain does not establish a winner.
- Does Stopped mean complete?
- Stopped describes execution. The completed/planned game count tells you whether all games finished. A stopped study can still have incomplete, provisional results. Updates delayed means the dashboard has not received a recent update; it is not a new result.
- Can I enter my deck?
- The Lab’s controlled populations are configured by the site. Use the Arena to enter and tune your own constructed deck, or Daily Draft to compare your drafting decisions.
Forge’s playing limitations apply throughout. The charts can reveal useful patterns in these simulations, but a narrow estimate does not account for a systematic AI or rules-engine mistake. Each experiment keeps its own card pool, rules, and comparison field.