Card quality
How strong the card is on its own before the rest of the seat is taken into account.
How Final40 Works
How Final40 is trained
Final40 AI is a neural-network trained on 17Lands Premier Draft data. Its training pool is filtered toward strong finishes from Diamond and Mythic ranked players, and trained to identify the picks and patterns that lead to stronger final decks.
The exact filters can move a bit by set depending on data volume, but the goal stays the same: model strong, repeatable drafting rather than one-off edge cases.
What it is optimizing for
Not the flashiest pick. The best draft from this seat.
Final40 is not trying to maximize one isolated card grade. It is trying to keep you in the strongest lane the seat is offering, then turn that seat into the best final deck it can build from the draft.
That is why the model can prefer a card with a slightly lower public win rate when that card keeps your colors cleaner, preserves synergy, or matches the path the draft has already started to take.
What goes into a pick
The score is additive, but the draft context changes what matters most.
Color fit
Whether the card keeps the draft in the colors your pool is actually supporting.
Synergy
How well the card works with the cards and archetype pieces you already have.
Draft-path fit
Whether the pick matches the direction your recent picks have been taking you.
Public draft data
Signals like GIH win rate, ALSA, and play rate. These matter, but they are only one part of the call.
How to read the report
The report is built to answer three questions fast.
What lane did the seat seem to want?
Which picks moved the draft away from that lane?
What deck would Final40 have tried to register from the same seat?
The pick viewer shows Final40's top choices in the pack, always keeps your actual pick visible, and uses the alternate route to show how a different early branch would have changed the rest of the draft.
Worked example
What a disagreement looks like in practice.
Exemplar draft: P1P2
In the exemplar report, the human pick was Frog Butler. Final40 wanted Brilliance Unleashed, with Cool but Rude still ahead of the human pick.
The point of that example is not that Frog Butler was unplayable. The point is that Final40 read the seat as asking for a different branch early, and the rest of the report shows what that branch would have turned into by the end of the draft.
Limitations
Where Final40 can still be wrong.
Final40 is better at steady, repeatable good drafting than at spotting rare edge-case exploits.
It learns from many drafts, so it tends to behave more like a strong average drafter than a once-in-a-blue-moon metagame gambler.
Public 17Lands signals can lag behind niche table dynamics, especially when a seat goes strange early.
Training filters can shift by set depending on how much strong-finish data is available, so the exact pool is not identical every release.
Ready to use it on a real draft?
Analyze a draft