From Earth combos to Water performance questions

Day 2 moved individual Earth abilities into a playable sandbox. Day 3 asks whether they can feel connected without turning a measured animation pass into a new choreography project. E drives a stone along the ground, Q launches its pair of rocks, and R finishes with a large left-handed aerial throw.

The comparison below places assembled source excerpts beside one continuous E → Q → R Unreal capture. It is a review edit, not a continuous three-move source sequence or proof of native Unreal audio. The synchronized soundtrack retains the corresponding source audio, including overlapping tails, so timing can be inspected without making a broader audio claim.

The useful change is small: the next valid ability can begin after the last projectile launches, rather than after the whole recovery finishes. That makes the remaining question specific. Do the short blends read as intentional footwork and upper-body continuity during a real playtest?

Each ability keeps its own brace, movement, release events, cooldown, and projectiles. The shared layer adds a one-ability input buffer: a later valid key replaces the pending choice, and the next cast begins after final release plus a 60 ms contact interval. Unreal blends into the incoming montage over 140 ms.

The boundary matters. Rocks already in flight continue independently, the outgoing casting sound can finish, and cancellation clears the queued input. An airborne or unsupported exit cannot become a grounded attack. The existing mannequin, locomotion, vaulting setup, and frozen Blender animations remain in place, while R retains the model’s anatomical left arm.

The previous-versus-current comparison isolates that policy change. In this capture, the buffered sequence finishes about 0.87 seconds sooner without speeding up the attacks themselves. It is a timing observation from one test, not a universal responsiveness target.

The test ran all six E/Q/R permutations: 18 casts, 24 projectile launches, and 12 transitions. Observed handoffs began 61–78 ms after the final release. The run ended with no remaining projectile actors and no movement lock.

A separate cancellation check queued Q during E, cancelled E, and confirmed that Q did not fire afterward. Those results establish event ordering and cleanup for this checkpoint. They do not establish that the blended motion looks convincing. The Earth combo remains pending Vincent’s playtest review.

This was a runtime integration pass with no new annotation cost. Its checkpoint preserves the request, recipe snapshot, traces, hashes, and review videos, which keeps a later visual decision tied to the exact implementation that was tested.

Earth has a stable visual anchor: a rock stays legible while it moves. Water has to change shape while preserving a believable connection between the hands, the moving mass, its wake, and the impact. A correct trajectory can still leave a stiff tube, a visible hinge, or a break at release.

That is why Water currently feels harder to shape. Earlier review records show a straight release tube replacing a curved source jet, foam strips reading as checker patterns, and material handling that rendered intermediate versions black or washed out. Adding more geometry did not automatically make the water believable.

The current aimed overhand Water comparison has technical checks for limb direction, hand guidance, downward flight, cancellation, and shared input. Its visual and interactive approval are still pending. The Earth combo deliberately uses the earlier calibrated animation graph so it does not overwrite that parallel work.

Water also introduces changing surface geometry, spray, and an experimental Niagara FLIP impact. Those are plausible sources of extra work compared with rigid Earth projectiles, but they are reasons to profile, not proof of a particular bottleneck. A fixed-rate capture cannot establish live frame time.

The next useful experiment is a repeatable scene and camera with warm shaders and the same resolution: idle baseline, Earth casts, Water with fluid impact disabled, and Water with it enabled. The record should include CPU and GPU frame times, median and high-percentile frame times, effect counts, and cleanup while leaving capture encoding outside the timed session.

The direction remains stylized and readable. Blender supplies editable motion and assets; Unreal handles aiming, effects, collision, and presentation. Day 3 does not settle Water’s cost or visual language. It gives that work a fairer question, while Earth now offers a concrete way to review combat rhythm.