Life LogMy stuff

Card Eval

Grade a new set's cards, then see how your grades line up with real win rates.

Opening the sets…

The rubric v1

The scale

  1. 5.0Best cards in the set. Take at any cost.
  2. 4.0Strong rare or premium uncommon. Nearly always a first pick.
  3. 3.0You basically never cut it.
  4. 2.5Solid playable.
  5. 2.0Filler. Sometimes cut.
  6. 1.5Below filler. Right deck only.
  7. 1.0Near unplayable.

The scoring machine

  1. Every card starts at 2.0.
  2. Body check. Stats against cost. Above rate +0.5, below rate −0.5.
  3. Text check. Real evasion +0.5. One shot value +0.5. A repeatable engine +1.0.
  4. Quadrant check. Good developing, at parity, ahead, and behind? +0.5. Only good while winning? −0.5.
  5. Job check. Can't name the deck that wants it in one breath? −0.5.
Overrides. Clean removal at a fair cost never grades below 3.0. Cards that need setup are graded at their floor, never their dream.

Downloads

Grade the whole sheet before you read or watch any set review. The .xlsx opens in Excel or Google Sheets. To use the template in Google Sheets, choose File, then Import.

How the numbers are made, and where they come from

How the numbers are made

The scoring machine

Rubric v1: every card starts at 2.0, then the body check (stats against cost, plus or minus 0.5), the text check (evasion +0.5, one-shot value +0.5, a repeatable engine +1.0), the quadrant check (good in all four spots, developing, at parity, ahead and behind, +0.5; good only while winning -0.5), and the job check (no clear deck wants it -0.5). Clean removal at a fair cost never goes below 3.0.

Checking yourself

You bring a table of real results (17Lands' card data export works; any table with names and numbers does). The score is the rank correlation between your grades and the data: both lists are turned into ranks, then correlated (Spearman's rho).

rho = Pearson correlation of your ranks and the data's ranks (1.00 is the same order, 0 is no relationship)

The page shows the number with those two anchors and the same number for every other set you've graded and checked in this browser; it puts no label on it. Grade groups (A is 4.0 to 5.0, B 3.0 to 3.5, C 2.0 to 2.5, D 1.5 and under) show the data's average for each, and say where a group's average sits past the group above it. A miss is how far your place and the data's place differ, as a share of the cards graded, so a grade scale and a win-rate scale sit on the same 0 to 100 scale. The two miss lists hold only misses in their own direction, largest first, up to a quarter of the cards checked (3 at least, 10 at most); a card at the data's own place is in neither. Each other set's number names the measure it was checked against (GIH WR, ALSA and so on). Clusters average the misses by color, rarity and type.

Sources

  • Set lists and card text from Scryfall.
  • Real results: whatever table you bring. The site fetches nothing from 17Lands.

The built-in check

For sets 17Lands has published, the check reads 17Lands' public game and draft data (CC BY 4.0) as summarized by the site: win rates in hand, in the opening hand, drawn later and in the deck for everyone and for players at 60% or more, improvement when drawn, and the average pick. Win rates count cards with 200 or more games in hand; the average pick counts cards taken 50 times or more.

About Card Eval

What it does

  • Grade the newest sets on the site: Reality Fracture, The Hobbit, Marvel Super Heroes, Secrets of Strixhaven and more, every card with its rules text, in grading order.
  • A fixed rubric: every card starts at 2.0 and moves by rule. The scoring machine does the arithmetic as you tick boxes, and you can always overrule it.
  • Keyboard fast: number keys grade, Shift adds the half, Enter moves on. Notes on every card.
  • Saved in your browser and in My Stuff, with a sheet view and a CSV export.
  • Check yourself: bring a card-ratings table (17Lands exports one) and see your rank correlation beside your other sets', the data's average for each of your grade groups, where your misses cluster by color, rarity and type, and your biggest misses in each direction (up to ten each, fewer for a small sheet) with room for an autopsy.
  • Still offered as a spreadsheet for people who like grading in Excel or Google Sheets.

Why

Card evaluation is a measurable skill. The only way to improve it is to make predictions, write them down, and grade them against reality. Round one isn't about accuracy. It's a baseline.

Asked for by players

  • A calibration score: once 17Lands data is in, see whether your A grades actually beat your B and C grades, and where your misses cluster by color, rarity, and card type. shipped

On the list because players asked for them in public feature requests and forums, September 2026.

Changelog

  • Check yourself against 17Lands' public data for the set, built in (win rates in hand, opening hand, drawn, improvement when drawn, average pick), with no file to bring.
  • The calibration check shows its number with its anchors (1.00 is the same order, 0 is no relationship) and your number on every other set you've checked, instead of a one-word label. Grade groups and misses are stated in averages and places.
  • Set icons on the set chooser.
  • Grade on the site: seven sets, the scoring machine, keyboard grading, and a calibration check against real win rates.
  • Added features players asked for in public feature requests.
  • Live. Rubric v1 and the Reality Fracture sheet published.