Limited study

How to read consensus without treating it as truth

Use the shape of model support to distinguish strong signals from legitimately close Limited decisions.

The gap matters as much as the winner.

A leaderboard-style list of cards can tempt you to focus only on which card is first. Pack One is more informative when you look at how support is distributed. If the top two cards are nearly tied, the model is describing a cohort with meaningful disagreement. If the first card is far ahead, the cohort is much more concentrated.

Three kinds of disagreement

Candidate-set disagreement happens when your preferred card is outside the cluster the model considers strongest. Ordering disagreement happens when you and the model identify the same strong cards but rank them differently. Confidence disagreement happens when you think a pick is obvious but the model is split, or vice versa.

Those three cases deserve different review. Candidate-set misses often point to valuation or contextual blind spots. Ordering differences are more likely to be subtle preferences. Confidence differences are invitations to inspect why one side believes the pack is more decisive.

Do not convert support into win rate.

Normalized support only compares the cards in front of you under this model. It does not say that selecting a 55% card produces a 55% match win rate, or that a 20% card is wrong 80% of the time. The numbers are a language for relative preference.

Best use: treat consensus as a second opinion from a reproducible model of strong-player behavior. A second opinion is valuable precisely because you can disagree with it.

Look for patterns across sessions.

One disagreement can be noise. Ten similar disagreements can be a tendency. If your misses repeatedly involve multicolor cards, expensive interaction, narrow synergy pieces, or flexible lands, that pattern is more useful than any single grade.