How to playtest a Commander deck online

What goldfishing is, and what it tells you

Goldfishing is the most basic way to test a Magic deck. A 2010 feature on Wizards of the Coast’s site calls it testing “against a defenseless opponent”, and the name says the rest: like a goldfish, it is an opponent who does nothing, so “all you have to do is deal 20 points of damage as quickly as possible.” Commander raises the target — the playtester seats every player at 40 life — but the idea is unchanged: take the opponents’ decisions out of the picture and watch your own deck work.

What it tells you is whether the deck works at all: how fast it gets going, whether the curve lets you use your mana every turn, whether the lands arrive in the right number and the right colours. What it does not tell you is how the deck fares in a real game, where three opponents remove your threats, counter your spells and race you.

Deck sites have their own tools. Moxfield’s playtest mode deals opening hands and draws, to catch mana and curve problems before a real game. Archidekt’s playtester has an optional interaction simulator, in which a hypothetical opponent may attack, remove or counter something at the end of your turns — not meant, Archidekt says, to replicate a real game. Lotus Vault’s solo playtester deals the whole deck onto a full Commander table.

The solo playtester, step by step

Everything happens on the Playtest page, and a game is dealt in a few seconds.

1. Import. Open the import window and give it the URL of a public Moxfield deck, or paste the list as text, one card per line: “1 Sol Ring”, “1x Arcane Signet”, or an export line with its set and collector number, such as “1 Command Tower (BLC) 130”. Archidekt links are not accepted here; export the list as text and paste it instead.
2. Pick the commander. Add *CMDR* at the end of its line, or let the playtester work it out. When the list does not say and several cards could lead the deck, it shows you the candidates with its best guess already selected, and you confirm.
3. Deal. Your commander goes to the command zone, your library is shuffled and you draw seven. You start at 40 life and take the first turn, with no draw on it.
4. Mulligan. Before your first turn, take a mulligan as often as you need. Each one shuffles your hand back into the library and deals you a fresh seven.
5. Play. Draw, play lands and spells, tap and untap, put counters on permanents, create tokens, roll a die or flip a coin, shuffle your library, and move any card between library, hand, battlefield, graveyard, exile and the command zone. Undo takes back up to five steps.
6. End your turn. Bot Alpha, Bot Beta and Bot Gamma take their turns automatically, then it is your turn again and you draw.
7. Restart. Restart deals the same list again, freshly shuffled, so you can compare one opening with the next.

Nothing you import is saved: the list lives only in the game itself, which is deleted after seven days without any action. A pasted list can run to 40,000 characters and 500 lines. Members can also play a deck saved in their account, pop their hand and library out into separate windows, and stream the table to OBS in one click.

The three bots: a routine with nothing to play

Bot Alpha, Bot Beta and Bot Gamma fill the three other seats, each at 40 life, but they are dealt no cards: the solo playtester deals your list and no other, so the bots have no library and no hand, and only your side of the table is on screen. When you end your turn they take theirs automatically, one after the other. Each bot’s turn is scripted to:

• Play a land from its hand, a non-basic one first, and tap its lands for mana.
• Cast at most two spells its lands can pay for, by priority: ramp first while the board is quiet, card draw, removal that climbs as the average salt score of the battlefield rises, board wipes once that score is high, then creatures by power.
• Attack with every untapped creature that is not summoning sick, all at the opponent it rates most threatening, mostly by the salt on their battlefield.
• End its turn.

With no cards, only the last step ever happens: a bot’s turn passes without a land, a spell or an attack. The solo table is a goldfish in the strict sense: four seats, and no one playing back. For opponents that really play, see the Forge playtester below.

What to watch in the first turns

A goldfish game’s value lies in what you write down. Play four or five turns of a hand, restart, and track the same things each game:

• Mana by turn three. Is a third land, or a mana rock, on the battlefield on your third turn? A deck that keeps stalling on two is short of lands or of cheap ramp.
• The commander on curve. Can you cast your commander on the turn its mana value allows, with the right colours? For a four-drop, look at your fourth turn.
• Keepable hands. Of the sevens you are dealt, how many would you keep, and how many mulligans does it take to find one?
• Colours. Do your lands produce the colours your hand asks for, or do spells wait for a missing one?
• Dead cards. Which cards sit in hand for turns, waiting for a board, a combo piece or a colour?

Then set your notes against the numbers. Analyse computes, for the same list, the exact odds of a keepable opening seven (two to five lands), of one land or fewer, of six or more, and of making your land drops on turns 2, 3 and 4 — a hypergeometric calculation, not a simulation. Those odds count lands only, assume no mulligan and no extra draw, and, like the playtester, have you play first with no draw on turn 1. The table shows what they leave out: whether a keepable hand actually does something, and what your mana rocks and ramp spells add. A dozen hands is a small sample — trust the odds for how often, and the table for how it feels. Analyse reads a public Moxfield or Archidekt link.

The limits of a manual table

The solo playtester runs on a small engine of its own and you move the cards yourself, the way you would at a kitchen table: it does not enforce every rule. That keeps it quick, but some things are left to you:

• There is no stack. A card you play goes where you put it, and nothing waits for a response. A counterspell, a trigger or an instant-speed trick is resolved by hand, in the order you decide.
• Nothing is checked. The playtester does not check that you can pay for a spell, that you have made a single land drop this turn, or what your commander tax is. The count is yours.
• Combat is arithmetic. Tap your attackers and count the damage yourself. In a goldfish game the question is how fast the deck could win, not what an opponent would block.
• There is no scry button. Scry by hand. On a computer, choose View Top X… in your library’s menu and send each card you do not want to the bottom with its B button; Put on Top closes the window and leaves the others in their order, and a card’s right-click Move to Top of Library changes it. On a phone, tap your library to see it top card first, then double-tap a card to move it.
• Mulligans are free. Every mulligan deals a fresh seven. The tabletop rule, the London mulligan, then has you put cards on the bottom of your library; the playtester does not, so a hand kept after several mulligans is larger than a real table would allow.

None of this gets in the way of what the tool is for: dealing hands and playing out turns. For games where the rules are applied for you, there is a second playtester.

The Forge playtester: rules enforced, members only

Lotus Vault’s second playtester is built on Forge, an independent open-source rules engine for Magic, developed by its own community and unaffiliated with Lotus Vault. Forge knows the cards and enforces almost the whole rulebook — the stack, triggers, replacement effects, state-based actions, combat — and seats you at a four-player Commander table against three bots. Those are Forge’s own bots, a different system from Bot Alpha, Beta and Gamma, and they play preconstructed decks picked at random: they block, they respond, and they can win if you let them.

It is not a public tool yet. Access is granted to member accounts one by one while the integration is a work in progress, and a game needs a deck saved in your account. The engine runs on Lotus Vault’s own server, with a limited number of tables; when all are in use, your game waits in a queue. And Forge, however close it comes to the real game, is not perfect: some interactions are missing or poorly implemented. The details are in How Lotus Vault works.

Goldfish, solo table, Forge or virtual tabletop?

Each way of testing answers a different question; the right one depends on what you want to learn.

A sensible order: goldfish until the deck opens the way you want, check the odds in Analyse, then — if you have access — take it to Forge, or to a table of real players, to see what it does when someone plays back.

Paste a Moxfield link or a decklist and deal your first hand: no account, nothing saved.

Related guides

• Commander brackets explained (1–5)
• How the on-device MTG card scanner works

Sources

• Playing Your Pet: Rough-Testing a Magic Deck — Wizards of the Coast (2010) (opens in a new tab)
• What is Goldfishing in Magic: The Gathering? — TCGplayer (opens in a new tab)
• Development Update – Playtester interaction simulator — Archidekt (opens in a new tab)
• Forge, an open-source rules engine for Magic — Card-Forge on GitHub (opens in a new tab)

Last updated: September 26, 2026