Set archive
SOS Pack One archive
Original analysis of 300 SOS replay seats: consensus confidence, close calls, recurring opening-pick leaders, and model context.
What is in the SOS sample?
This archive uses 300 historical replay seats. The consensus model for this set was trained from 5,000 drafts and 210,000 picks drawn from an experienced cohort of 70,209 drafts. The recorded win-rate cutoff for the cohort is 60%.
How decisive were the opening packs?
Across these replay seats, the model's average gap between its first and second cards was 28.8%. 26% of packs had a gap of ten support points or less, while 45% had a gap of thirty points or more. That split is useful when reviewing a miss: a narrow disagreement is not the same signal as passing a card the model strongly separated from the field.
Cards that most often led the opening pack
These are not a set ranking or a pick-order list. They are the cards that appeared as the model's most-supported card most often in this particular 300-seat replay sample. Frequency depends on which cards were opened.
Conciliator's DuelistTCGplayer
Snarl SongTCGplayer
Traumatic CritiqueTCGplayer
Postmortem ProfessorTCGplayer
Practiced OffenseTCGplayerAffiliate disclosure: TCGplayer links are marked sponsored. After Pack One's Impact partnership is activated, eligible purchases may generate a commission at no added cost to the buyer.
Three of the closest opening decisions in the archive
These packs produced the smallest support gaps between the top two cards in the sample. They are especially useful challenge packs because the model itself expresses little separation.
| Model #1 | Model #2 | #1 support | #2 support | Gap |
|---|---|---|---|---|
| Dreamroot Cascade | Paradox Surveyor | 23.2% | 23.2% | 0.0% |
| Charging Strifeknight | Stirring Honormancer | 28.5% | 28.5% | 0.0% |
| Additive Evolution | Inkshape Demonstrator | 24.8% | 24.8% | 0.1% |
How to use this page
Use the statistics to calibrate your expectations before treating a score as a verdict. A set with many close opening decisions should naturally produce more reasonable disagreement. The archive is descriptive of the stored replay sample and model version; it is not a complete ranking of every card or every possible pack.