Battlefield Lab

Lab 6 · Random opponent · 24 compositions · 8 combat seeds · both AI assignments. Same tiers, 35,000 Value per stack, heroes ATK/DEF 5/5. Historical fixed benchmark, independent of the armies above.

Discovery and holdout each contain 12 different compositions. More depth is not guaranteed to improve the limited search. These timings are for full headless battles on the benchmark computer, not phone playback. Results against Random alone do not establish strength against Tactical or humans.

Value budget rounds up per stack. Ground creatures cannot cross living allies or enemies. Flying and blink can cross occupied cells but must land on an empty cell. Empty board: no obstacles or traps yet.

Shooters fire from their current cell. Adjacent enemies force melee unless a source trait permits shooting. Range penalty starts at distance 3 (−10%, capped at −50%). Long Reach uses a straight two-hex attack, without ordinary retaliation; adjacent attacks use melee. Lab assumption: the intermediate hex does not block reach.

Melee attacks and retaliations support Area Strike, Sweeping Strike and Dragonbreath with the same footprint. Breath can hit allies. Area strikes use separate damage rolls per target and only the primary target can counter. Retaliations never trigger another retaliation. Jousting uses the actual travelled distance.

Random chooses a reachable enemy or approaches one. Tactical scores damage, counters, protection, blocking and exposure using fixed rules. Search-2/3/4/5 forecast that many stack decisions, including the current action, with midpoint damage. They consider eight Tactical root candidates and three ranked continuations per node, preserving WAIT as an additional candidate. It does not predict morale or future random rolls and is not an exhaustive search.

Tactical and Search compare WAIT with moves, attacks and SKIP. WAIT defers the action to the reversed-initiative phase; it cannot repeat. All Search depths follow the shared queue across round boundaries, including retaliation reset. One morale roll is cached per stack and round, so waiting cannot reroll it. After three rounds with unchanged HP totals, a seeded random side must attack or approach until HP changes; blocked battles can still reach the safety limit. This stall rule is specific to the lab.

Vampirism restores HP and can revive losses; Phantom Vampirism can grow a stack above its initial size. The log shows restored HP. Corpse Eater clears all corpses in the destination hex and restores up to 120 HP once. Embodiments leave no edible body. Hero ATK/DEF apply to damage and AI evaluations; both default to 5.

Muted red cells show threatened destinations considered for the current decision. The explanation lists its largest scoring factors; numbers are Value-equivalent scores, not win probabilities. No hero spells, active abilities or manual movement yet. Passives affecting spells, mana, focus or the adventure map have no effect here.

Selecting a stack previews movement only; highlighting does not issue commands. Replay is available within this page session. The original Autobattler uses its separate rules and saved history.