An auto-battle system still needs to express player intent. Source of Thought uses hate-based target selection so placement influences who gets attacked without requiring manual commands every turn.
A multiplicative target score
The current target model combines column weight, depth weight, resonance weight, skill modifiers, and a small random factor. Multiplication is deliberate: formation and card state can reinforce or suppress one another instead of becoming unrelated points in a flat sum.
Front positions create more offensive pressure and attract more hate. Rear positions reduce exposure. Threat, distance, current health, and role can influence the target score while the small random term prevents every battle from replaying identically.
Hate is the bridge between where the player places a card and what the enemy decides to do about it.
Why not simply attack the weakest card?
Always selecting the lowest HP target would make formation mostly cosmetic. Always selecting the nearest target would reduce the system to geometry. Hate allows a durable card to protect a fragile card, a high-threat attacker to draw attention, and a skill to modify priority.
Resonance affects risk as well as power
Adjacent works can change more than attack and defense. Because resonance participates in hate weighting, a tightly connected group may become more effective while also changing which unit becomes the visible threat.
Controlled unpredictability
The random component remains intentionally small. The player should recognize why a target was likely chosen even when the result is not perfectly deterministic. Future Battle UI will expose enough targeting feedback to make that causality readable.