Your first draft
You build the deck.
The computer plays the games.
Start with the same packs as everyone else, make your own picks, and see what your deck can do across repeated Magic games.
Everything here is free. Draft and build in your browser; make an account with a recovery email when you want to run your deck. No cards to buy or software to install.
From the first pick to the last game
Make something worth testing.
Choose an environment
Current, Classic, Cube, and Crazy rotate daily. Pack It Up is an extra chaos challenge every day. Each format tells you its pack count, table size, and pick rules. Familiarity with Magic helps; the site handles passing packs and playing the games.
Draft with the bots
Take a card, pass the pack, and build a pool. Most tables have you and seven bots; special formats can change the seats or cards per pick. You set the pace. Each confirmed pick is saved and final.
Build your forty
Choose at least forty cards from your pool, including lands. You can add unlimited Plains, Islands, Swamps, Mountains, and Forests. The builder’s Add lands button helps fill out the deck. Review it before submitting: your submitted list is locked.
Run your deck
Create a free account or sign in, then select Run my deck. Forge’s AI pilots both sides on the server. You can close the page and return through My Drafts while the queued games are evaluated.
Follow the evidence
Watch the benchmark results arrive, inspect matchups, replay your picks, and compare builds. Ratings can keep moving as more games finish. The archive preserves the final ranking after the daily window closes.
Draft at your own pace, but leave time before the ranking deadline. Simulation wait time depends on the queue and how long the games take; a game-count target is not a promise of an instant result.
Same starting point. Your choices change the table.
Within one challenge, the opening packs and bot roster are fixed, and each human takes the same seat in their own draft. You are drafting asynchronously with bots, so there is no need to wait for other people.
One shared opening packThe same cards are available at the start.
You take the creatureThe removal spell travels to a bot.
Someone else takes the removalThe creature travels to their bot.
Those decisions can change the bots’ colors and later picks. Two players may see different cards farther into the draft. Pick comparisons only match decisions where the offered cards were identical. In Pack It Up, choosing the order of your three assigned packs adds another decision before drafting starts.
Three groups of decks, three different roles.
The bots you drafted with
Your picks helped shape these decks. Explore their pools and picks to understand how the draft unfolded. They are not the shared rating benchmark.
The same test for everyone
A separate all-bot draft creates fixed reference builds, also called standard candles. Every entrant faces these exact lists. The Mirror Drafter is the reference bot that took your seat and opening packs.
Other answers to the challenge
Submitted human decks join the same rating field as the references. Additional games help compare the field. Player placement compares people; the full ranking also includes reference decks.
A normal eight-seat draft schedules twenty games against each of eight references: 160 benchmark games. Other table sizes adjust that allocation. The results page shows the target for your format. Extra ranking games add evidence to the same rating.
An actual archived draft · September 9, 2026
What can you learn from one deck?
In Shedletsky’s MtgForge Competency Cube, shedletsky first-picked Topiary Stomper and built a black-green ramp deck. Here is the frozen result, with the different measurements kept separate.
The benchmark record shows how the deck did in its fixed test. The rating uses the broader game evidence to estimate strength against this particular field. The likely range shows uncertainty in that estimate.
Finishing first among one human tells you very little about other people’s drafting. The reference decks still provide useful opposition, and the overlapping ranges mean the exact order is uncertain. A successful deck also does not prove that every individual pick was best.
The daily clock and your saved draft
New challenges arrive at 00:00 UTC. The next scheduled reset is . Each challenge’s countdown is the best place to check its remaining time.
- When do entries and rankings close?
- New ranked entries close at the daily reset. Existing entrants have one further hour to submit and qualify. A ranked finish requires submission and at least 3 completed benchmark matchups with successful games before ranking closes. Starting late may leave too little simulation time.
- Can I pause, restart, or change my deck?
- You can resume saved picks. A challenge keeps one run per player; opening it again resumes that run rather than dealing new packs. Picks cannot be undone, and a submitted deck cannot be replaced. You can revise your build before submission.
- What if I miss the deadline?
- You can normally finish an existing run after ranking closes, but it cannot change the final standings. Completed challenges also accept unranked replay entries for players without a run there. A challenge retired by the site stops accepting further draft work; its saved history remains.
- Where is my guest draft?
- Your picks are stored on the server, with guest access tied to a cookie in this browser. Keep using the same browser until signing up or signing in attaches the draft to your account. Clearing cookies or switching devices first can lose access. If the account already has a run in that challenge, its existing run takes precedence.
What the computer is doing
MtgBattles’ drafting bots choose cards using their drafting preferences. Forge is the separate community-built rules engine and AI that plays the submitted decks. Battle Rating combines those game results; the Deck Lab experiments with changing deck lists. The Lab does not train Forge to make better in-game decisions.
These are played-out games, but Forge can make poor choices or encounter a rules bug. A result measures how a deck performs with this computer pilot in this field. More games reduce sampling uncertainty; they cannot remove a systematic mistake. If a play looks wrong, send a bug report with the deck, game link, and what you expected.
Queued means waiting for simulation capacity; playing means a game is underway. Failed or timed-out attempts do not automatically become losses. Results remain provisional while evidence is incomplete. Check the progress and matchup counts, and report a stalled result with its link rather than submitting again.
How close are the cards and packs to the original format?
The format’s Rules label describes its recorded draft procedure and fidelity. “Exact” does not mean an unfiltered physical booster: cards this Forge version cannot play are excluded. Cube lists are frozen snapshots; Crazy recipes deliberately change draft rules. Open a format’s history, source links, and cube snapshot to see what it uses.