Questions and answers

Commander decks with a fixed budget play against each other here – not at the kitchen table, but in the Forge simulator. Each price class has the 64 most popular commanders on EDHREC plus four official precons. Every game has four players, and each deck plays 10 games per tournament. So far there have been 147 tournaments.

The question behind it is simple: which commander gets the most out of little money? Every deck comes with its full list and a Cardmarket shopping list. It's a private hobby project, with no ads.

Prices and EDHREC data are refreshed every day, combos from Commander Spellbook once a week and card data from Scryfall every 30 days. The site rebuilds after every tournament and once a day with fresh prices.

Yes, join the server here. We talk about results and deckbuilding, share budget tips, look for players and collect bugs and ideas. New results and changes are announced there first. Deck requests and your own tests still go through the forms on the “Get involved” page.

For the programming, yes: Claude by Anthropic helped me. I decide what gets built, and I check and test everything myself.

The content isn't made by AI. Decks, rankings, Elo, win rates, strengths and weaknesses come from real games in the simulator, evaluated by fixed rules. We don't generate images; the card art comes from Scryfall. When this site talks about the “Forge AI”, it only means the computer opponent that plays the cards in Forge.

They're sorted by Elo, like in chess. Every commander starts at 1500. After each game the winner gains points and the others lose some. How many depends on the opponents: beating strong decks is worth a lot, beating weak ones little. Placement counts too – going out second costs less than going out fourth.

Example: a 1700 deck beats decks at 1600, 1550 and 1500 and gains about 6 points. If the 1500 deck wins instead, it gains about 13 and the favourite loses points.

The tier comes from the Elo: S from 1700, A from 1600, B from 1500 (the average), C from 1400, D below that. Every deck page shows it as a verdict at the top, from “Top tier” to “Weak”. The table also shows how the Elo changed since the last tournament and the form of the last 10 games (gold, silver, bronze, red for places 1 to 4).

The win rate is left out of the table on purpose, because on its own it misleads. Every deck plays opponents near it in the rankings (about ±12 places), so #1 never meets #64. 49% against the top is worth more than 55% against the middle. At a four-player table the average is 25%. You'll find the win rate on each deck page.

Keep in mind what the rankings measure: how well a deck does in Forge under equal conditions. Politics, bluffing and your own playgroup aren't part of it.

Games that hit the time limit count as neither a win nor a loss and are replayed up to three times. Until 2 Nov 2026 the Elo was calculated from all games at once; since then it goes game by game.

As a fixed benchmark. Four official precons play in every tournament and every price class, exactly as sold in stores. The built decks keep changing, the precons never do. That shows how far ahead a built deck is of one you buy ready-made, and what a bigger budget gets you. That precons do worse in the expensive classes is intended – that gap is exactly what the extra money is worth.

They aren't in the rankings themselves but below them under “Benchmark”, with the place they would have. Their games count normally, but beating a precon earns only a little Elo.

We test the 64 most played commanders on EDHREC, so every tournament has enough well-known, strong opponents. When the EDHREC list changes, a new commander moves up. Commanders that drop out are listed below the rankings under “No longer in the field”, and their deck pages stay available.

If yours is missing, you can have your own deck tested or have one built for you. You'll find both under “Get involved”.

From Forge's game logs. We read every game: which card dealt how much damage, who knocked out whom, what got blocked, lands, mulligans and every spell cast. Then we compare each deck with all others in the same price class.

We also look at what a deck struggles against, such as lots of flyers, counterspells, board wipes or graveyard hate. Only very clear differences are listed as weaknesses (99% certainty). If nothing is that clear, we show a tendency at most and say that it isn't certain yet.

Logged means it's in the game log as such. Indication means something that keeps showing up but isn't proof. A card that mostly comes into play in games already being won looks good even if it adds little.

The “Games” tab of a deck page shows damage per game, damage when used, top finishers, best interaction, the biggest obstacles (opposing cards that block most often), rarely cast cards and wins by turn.

They're based on the decklists of thousands of players on EDHREC. Our deckbuilder turns them into the strongest deck that fits the budget, with enough ramp, card draw, removal, board wipes and protection. 100 cards, colour identity and the official ban list are always checked.

After that the program learns from every game. In each tournament every deck tries out one card. If it wins more often with it, the card stays; otherwise it goes. The same swap is tried again after three weeks at the earliest. A change only makes it into the shopping list once it has proven itself.

All 100 cards including the commander, at the Cardmarket average price, updated daily. That's why all prices are in euros – on TCGplayer in the US some cards can be noticeably more or less expensive. The limit is strict: a deck “up to €100” costs at most €100. If prices rise later, we rebuild the shopping list automatically.

Up to €100: at most 3 Game Changers (Bracket 3). Game Changers are especially strong cards from the official Wizards of the Coast list. Combos are allowed and the official ban list applies. Other bracket rules, such as no early two-card combos, aren't checked by the deckbuilder.

Every deck page has the Cardmarket shopping list in the “Buy” tab. Copy it, create a new list under “Wants” on Cardmarket and paste it in – done. The “1 Card name” format also imports into Moxfield or Archidekt.

The list is exactly the version that did best in the tournament. Cards the Forge AI can't play were replaced for the games. The originals and a list to copy are in the “Decklist” tab under “Adjusted for the Forge AI”.

All games are played by Forge's computer opponent. Out of the box it plays decently, but not like a human: it rarely attacked the player with the most life and ignored win-the-game cards like Felidar Sovereign. So at the end of October 2026 we made it play more like a person, tested in hundreds of comparison games:

• Whoever is ahead gets attacked – especially decks that win through their life total.
• Threats are taken seriously. Win-the-game cards and infinite combos are removed or countered as soon as they land. If someone is about to win, everyone attacks them but keeps a blocker back.
• Mulligans and board wipes like at a real table: hands with no or too many lands go back, board wipes come when they hit the whole table.

It still doesn't play quite like a person. It can't play 2,526 cards at all; they're replaced by similar ones in the games but stay on the shopping list. It plays another 2,100 only to a limited degree. Combo and engine decks like Kinnan or Kenrith, politics and bluffing aren't its strength, so such decks tend to do worse than at a real table.

So the rankings honestly show how strong a deck is in Forge. A good guide, but not the final word on your table. See the list of affected cards

Yes. Because of the many requests, currently for supporters on Ko-fi only – otherwise we can't keep up with testing. Once things calm down, it will be free again. You tell us the commander and budget, for example Kaalia up to €250. Then this happens:

1. Build: just like the tournament decks, from the EDHREC lists, with Cardmarket prices and never over your budget.
2. Test: the deck plays 120 games against decks of every tier in the same budget, with the same AI as in the tournaments. If a swap could improve it, it plays the same tables again with the swap. We only mention it if the deck does noticeably better.
3. Delivery: a short email with the key points and a link to the deck page with the list, shopping list and all the analysis. The deck only appears publicly if you agree.

Requested decks aren't in the rankings, but their page shows roughly where they would land.

Yes. Because of the many requests this is also for supporters only right now, up to 3 decks a week. Anyone who submitted before the change will of course still be tested for free. Here's how it works:

1. Submit: commander, decklist and email in the form – as a link to a public Moxfield or Archidekt deck or as text, one card per line.
2. Check: we check 100 cards, colour identity and the ban list and get prices from Cardmarket. If something's wrong, you get an email listing the problems.
3. Test: your deck plays 120 games against 20 commanders from the rankings, four from each tier, all in the same budget as yours. That's as many games as in about 12 tournaments, so luck doesn't decide. A test usually takes three to four hours; testing runs around the clock in order of arrival.
4. Re-test swaps: candidates are up to two cards that rarely came into play in your test – each only if a better-fitting card with the same role fits the budget. Your deck plays the same 120 tables again with these swaps. We only list what makes your deck at least 0.3 places better on average; we never fill up the list. We never swap combo pieces or cards many players of this commander choose on purpose.
5. Result: a short email with the verdict, Elo, placements and the key tips. Everything else is on your private page: win rate against each tier, which cards dealt damage or ended games, who knocked you out and what rarely came into play.

Our emails sometimes end up in the spam folder. If nothing arrives, please check there.

Your deck doesn't appear in the rankings. The private page isn't linked anywhere and is blocked for search engines. Cards the AI can't play are replaced by similar ones in the test, and we say so.