Play WAYBOUNDWhich Companion is strongest?
WAYBOUND 0.35 · 17,920 simulations · rules and decks unchanged.
Best standard-AI overall result: Skara (67.9%).Prepared final-form leader: Nima / Starweave Sage (66.0%). These answer different questions: normal matches measure the full deck-and-Companion package; the prepared test starts with upgrades already granted.
Overall rankings include faction power. Same-faction comparisons below help isolate the Companion choice. These are heuristic AI outcomes, not expert human win probabilities.
Normal-start tournament
7,168 matches; each Companion appears 1,792 times against the seven others. Full faction decks, Guard 5, Journey progression, recovery, and +2/+2 Ascension enabled. Both seat orders, starts and all four scoring/aggressive policy combinations.
First-player win rate: 50.1%. Mean numbered final round: 6.33. Rankings use win + half a draw. Draws: 0.
Which Companion within each faction?
Direct duels use identical faction decks. Common-opponent rates compare the two Companions against the same six out-of-faction opponents, with matched seeds, starts, seats and policies.
Exploratory intervals resample whole seed blocks 10,000 times, preserving games within each seed. They reflect variation in these AI samples and do not establish human balance. An interval spanning zero does not show a clear choice.
Normal head-to-head matchups
Each cell is the row Companion’s score rate against the column Companion. Each pair has 256 games; no self-mirror games are needed for ranking.
Prepared Ascension tournament
3,584 constructed late-game matches; 896 appearances per Companion. Both Companions start fully ascended, with all four combinations of their retained first paths evenly tested. Round 6, six Journey marks, six Charge, no damage, otherwise empty board and opening hands/decks. This is not a naturally reached game state or a normal-game win forecast. The large first-player advantage illustrates the fixture’s limitations.
First-player win rate: 74.4%. Mean numbered final round: 9.57. Rankings use win + half a draw. Draws: 5.
Realm-strategy sensitivity
The main AI never explored the Netherway with a Companion in the normal tournament. To check that limitation, 3,584 additional normal-start games used the original policy and 3,584 used a separate realm-planning wrapper on the same eight new seeds. The wrapper plans legal open → ready entry → Explore combinations using only the player’s known hand and public board. It can wait at Lantern Crossing for an affordable next-turn combo or return there when blocked. This wrapper is an offline sensitivity test; the published game AI is unchanged.
Realm-aware leader: Skara (71.0%). This is one attempted realm strategy, not proof of optimal Veilborn play. Different seeds mean its control rates should be compared with the matched control column, not directly with the 16-seed main ranking.
Progression and limitations
Ascension occurred in 28 of 7,168 normal-start games (0.4%). Skara won the main ranking without ever ascending in this sample. Final-form strength therefore cannot explain that lead. The ordinary AI chose paths according to its current heuristic; the prepared test explicitly crossed both first-path choices.
28 distinct Companion pairings including same-faction sibling duels; canonical full 30-card decks, no mulligans, both seat orders and starts, all four shipped scoring/aggressive policy pairings. Normal: 16 fresh seeds, 7,168 games, 1,792 appearances per Companion. Ascended: two separate seeds, all four first-path combinations, 3,584 games, 896 appearances per Companion. Prepared ascended scenarios start in round 6 with each Companion fully ascended at six marks, no damage, otherwise empty board, opening hands/decks and six Charge; not naturally reached late-game states. Policies and engine unchanged. Rates are model outcomes, not human win probabilities. Seed blocks and mirrored games are dependent; no independent-binomial confidence claims. Faction/deck effects confound overall ranks; sibling direct and common-opponent cross-faction rates help compare Companions within a faction. Natural AI chooses paths itself; it may underuse realm, healing and setup strategies. Ascension samples are reported explicitly.
Eight additional fresh seeds 51000–51007; identical full round-robin schedules and styles in control/realm-aware conditions. Realm-aware wrapper plans a legal open + ready entry + Explore sequence using only own known hand/public board; can hold a Companion at Lantern Crossing for an affordable next-turn sequence or return there when blocked. No engine or published policy changes, no statistical retuning. This tests sensitivity to one realm strategy, not optimal human play.
Small policy sets, repeated seeds, unmodeled human plans and empty-board late-game fixtures limit the results. Veilborn, healing, and setup tactics may need more capable agents and human playtests. No balance changes were made from these outcomes. All matches completed with legal actions, nonnegative Charge and exact 30-card conservation for each player.
Tested gameplay source: 754354c677d6b7cb7c92f453f266d5e466da8559. Full engine/card hashes are included in the data download.
Download all 17,920 match records · Ascended forms and rules