Balance Report

The gap between greedy and naive is the measurable value of the effect text. If it is small, the cards are decorative — players could ignore the rules text and do nearly as well.

Pick rate — how often a card is played when it is in hand. A card seen constantly but rarely picked is a dead draw.
Lift — average final score minus base value. What the effects and Shadow Cards actually add to the printed number. Only sees a card's effect on its own score, so denial cards read near zero.
Rel lift — improvement in the player's margin over the rest of the board. This is the one that sees denial: dragging an opponent down never moves your own score, but it does move your margin.
Win lift — how much more (or less) often winners played this card than the average player, corrected for winners surviving longer and playing more cards overall. 1.00 is neutral.
Greedy win lift — the same figure measured only among greedy seats. Prefer this one. Because greedy wins most games, plain win lift rewards any card greedy happens to like, which says more about the strategy than the card; restricting to a single policy removes that circularity. ▲ marks ≥1.15, ▼ marks ≤0.85, both on the greedy figure.

Avg swing — average value the card added to or removed from the score it was attached to, counting only rounds where it reached resolution. A curse near 0.00 is being played around successfully — which is the design intent, but a curse that never lands is just a wasted Ghost turn.
Victim damage rate — how often the haunted player took the round's damage.

Two things this table cannot show. Cards that never touch a value — Shadow Shackles, Phantom Weight, Spectral Shift, Grave Robbery, Grave Gift — read 0.00 swing by definition; judge those on victim damage rate alone. Shackles in particular suppresses printed text rather than subtracting from a score, so the value it denies is never computed and cannot be measured here.
Times-played gaps between curses and blessings are an artefact of the Ghost AI, which curses the leader roughly three times out of four. Compare cards within a group, not across.

The rulebook prices going last by making it lose ties for lowest score. These numbers say whether that price is set correctly.
Read the index, not the raw rate. Exactly one player takes the round damage, so a seat's fair share is 1/(players in the round) and 1.00 means it took exactly its share. Raw rates are not comparable between buckets: a 2-player round has no middle seat, so middle is sampled only from larger games where every share is mechanically smaller.