The model's ranking of the actual pick:

Pack    |  1   2   3
---------------------
Pick  1 |  1   1   2
Pick  2 |  2   1   1
Pick  3 |  2   2   1
Pick  4 |  2   1   1
Pick  5 |  2   1   1
Pick  6 |  1   1   1
Pick  7 |  1   2   1
Pick  8 |  1   1   1
Pick  9 |  2   1   1
Pick 10 |  1   1   1
Pick 11 |  1   1   1
Pick 12 |  1   1   1
Pick 13 |  1   2   1
Pick 14 |  1   1   1

Notes:
- p1p10: At this point, the players move the three blue cards to the sideboard, estimating that they are unlikely to play them. The model agrees, ranking them at -0.52, -1.73 and -1.76. However, the model also has Arahbo at -1.05, which does not make that much sense. Perhaps it does not expect us to stay in white, but it does have Squad Rallier at 0.11. These rates are also somewhat fluctuating pick to pick. After the end of pack 1, the rates look quite sensible.
- p2p1: Notably, while the actual pick was correctly predicted, the other cards in discussion differ: the players are deciding between the two white cards Dazzling Angel and Guarded Heir, while the model prefers the two green cards (Bite Down and Beast-Kin Ranger) as well as Fiery Annihilation (a red card) over the Guarded Heir.
- p3p8: They pick Ashroot Animist, a green-red card, which would be a splash, noting that this is a speculative pick, but it might be played. The model agrees, rating the card at -0.05 to be played, so roughly 50%.
- deckbuilding: the model predicts most of the deck correctly, but it wants to splash the Elenda (and include the two Evolving Wilds). It also prefers Vanguard Seraph to Apothecary Stomper. So overall it plays Evolving Wilds (x2) and Elenda, Saint of Dusk instead of Elfsworn Giant and Apothecary Stomper.