A keep is a claim about the opening turns.

Name the opening job first

Before evaluating a single card, complete a sentence: in the first three turns, this hand needs to ____. The answer might be establish initiative, prevent an early snowball, assemble enough structure to reach a stronger phase, or preserve a specific form of interaction. The job must describe the position you want, not simply the cards you hope to draw.

Use the information you actually have. Your own list gives you redundancy, curve, and dependencies. Going first or second changes time and resources. The opposing class or revealed format context may suggest a range of plans, but it rarely proves one exact deck. Treat that range as weighted uncertainty, not permission to imagine the most convenient opponent.

Once the job is named, each keep must earn its place. Ask whether it performs the job alone, enables another keep, protects against a common failure, or merely feels too powerful to return. A strong late card can be a weak opening keep if it narrows every redraw and leaves the early job undone.

Evaluate the hand as a system

Cards change value beside one another. Two pieces may form a stable sequence, or they may compete for the same turn. A conditional card can be keepable when its activator is already present and poor when it asks future draws to rescue it. Several individually acceptable cards can create a hand with no flexibility, no response, or no way to spend resources efficiently.

Describe the hand along four axes: curve, role fit, dependency, and redraw cost. Curve asks whether the opening can use its resources. Role fit asks whether those plays advance the job. Dependency asks how many things must go right before a keep works. Redraw cost asks what you give up by refusing another chance to find a higher-priority piece.

Do not demand a perfect curve from every opening. A flexible answer or a powerful setup may justify an uneven sequence. The point is to know why. If the hand needs two specific draws by a specific turn, name that fragility. If it has several acceptable branches, name that resilience.

• Curve: can the hand act on the turns that matter?
• Role fit: do those actions perform the opening job?
• Dependency: how much unseen help must arrive?
• Redundancy: how many future draws can cover the same need?
• Failure mode: what common opponent start makes this hand collapse?

Compare keep sets, not yes-or-no cards

Build two or three candidate keep sets. The first is the safest coherent opening. The second may accept more risk for a stronger payoff. A third may target a credible opposing plan. For each set, say what redraws remain useful and what opening sequence becomes possible. This keeps the comparison grounded in the whole hand.

Then test a common-success case and a common-failure case. In success, do not assume the perfect draw; use any one of several reasonable draws. In failure, choose a plausible opposing start rather than an extreme. A keep set is robust when it performs acceptably across the weighted range, not when it creates the best screenshot under ideal conditions.

Commit with a short reason before confirming the mulligan. For example: I keep the flexible early play and return the conditional payoff because my job is to contest the first two turns, and several redraws can supply a later threat. This reason can be reviewed. I always keep this card cannot.

Review the keep before the draw sequence

After the match, reconstruct the mulligan with later cards hidden. Restate the opponent range you reasonably assigned, the opening job, and the keep sets you considered. Ask whether you ignored a common failure, overvalued a dependency, or underestimated how many redraws could perform the same job.

Only then reveal what happened. A returned card appearing immediately is not proof that returning it was wrong. A risky keep working is not proof that the risk was justified. The result helps you discover missing branches, but the keep should be judged against the information and range available at the opening.

Track reasons rather than raw keep outcomes. After five similar openings, look for repeated assumptions. Perhaps you repeatedly keep power without a bridge to the relevant turn. Perhaps you overreact to one opposing class and give up your own plan. Convert the pattern into one practice behavior, such as naming the opening job before looking at individual card quality.