Criticals that cut

One day of combat feedback: contact that has a sound, critical hits that cut, and a thrust that commits.

Contact has to make a sound

Existing cues from the licensed sound library are now routed to armour, to destructibles, and to sweeps against ordinary static geometry. The new hard-contact transients only apply to wood, metal and stone, so cloth and glass do not pick up a masonry thud. Slashes get a second whoosh in a second window, and steel swings are suppressed on parries and projectile actions.

Enemy armour was the instructive gap. It is visual only and bypasses the armour definitions the player's gear uses, so the new armour contact cues never reached enemy plate. Enemies now carry explicit impact-sound metadata, and the Bound Knight's metallic contact was added without touching its defences.

The test that matters is the plain one: a blade sweep against an ordinary wall must produce exactly one stone impact per swing. It failed first, and not because of the audio. The fixture's wall was a static component, which rejected a mesh assigned at runtime. The fixture now assigns the mesh while movable and then restores static mobility, and on failure it reports the contact count it actually saw.

The 25 checks pass. Thirteen new combat cues were generated from 58 library variants, and onset and loudness validation passed. That is verified by tests and waveforms, not yet by ear.

Critical hits can cut

The brief was chance-based critical slices and enemy dismemberment, with the existing Luck critical chance and damage multiplier left exactly as they were. The slice is tracked as its own proc, separate from the scripted riposte, so nothing about the damage rolls moved.

The rules:

• Only a landed physical slash can sever, only on an ordinary humanoid enemy, and only if it did actual health damage. Blunt hits, spells, status ticks, blocked hits, hits on an invulnerable target and guard chip cannot.
• A non-lethal slice takes an arm. An enemy that still has a weapon-bearing hand keeps fighting, and an attack that needs the missing arm stops. Losing the last weapon-bearing hand is fatal.
• A lethal critical slice splits the torso, and the held weapon and shield drop as debris.

The pieces are posed lowest-LOD fragments with capped tissue and kinematic motion, and everything is restored at a checkpoint. Loot stays in the original one-time death roll; fragments never carry any.

The floating gauntlet

Rendered review showed an isolated hand that survived two rounds of fixes. State tests could not have caught it, because every state flag was correct.

Coloured bounds drawn around each fragment showed the gauntlet was outside every fragment actor. The arm-only outfit components had been copied onto the upper half of the corpse but left out of the cleanup that hides the original meshes, so the original gauntlet stayed visible, floating where the arm had been. The cleanup now includes them, with a regression assertion that every original skeletal visual is hidden after a split.

Two smaller findings from the same renders: corpses and shields came to rest upright, so fragments now settle on their broad face with wall sliding, and thin swords rested above the floor because of a minimum sphere-radius offset, which was removed.

The thrust

A paired critical kill with the greatsword can now be an impalement instead of the shared heavy critical: chamber, drive, a held penetration, extraction. The paired spacing uses the weapon's visible blade length, so the Frostborn blade does not enter the enemy during the wind-up. A standing death clip after extraction risked a visible pop, so impaled enemies have their own death clips: torso held, knees buckle after the blade comes out, then prone or supine with no recovery.

The tests found the routing before the pictures did. The first run passed 36 regression checks and failed all six new route assertions: gameplay was still choosing the shared critical. The move data is exported from the C++ animation tables, and the export has to run before the data import. Run in that order, the new thrust was selected.

Review then asked for a more forceful thrust. It now has a deeper coiled crouch, a wider planted stance and more hip and shoulder drive, and the paired thrust has a synchronized 85 millisecond impact pause. Damage multipliers are unchanged.

Blood that could be seen

The first blood read as sparse dark dots. A directional exit jet was added, with stretched droplets, red mist, falling pressure and runoff, and severed wounds got timed liquid streams anchored to the wound and carried with moving limbs, with pooled splashes on the floor. Everything is cleaned up at a checkpoint.

The floor splashes were still faint, and the cause was one setting. The smoke material has a depth fade of 40 cm, which suppresses any mark within a fraction of a centimetre of geometry, and a splash on the floor is exactly that. Floor blood now uses its own wet shape with no depth fade, irregular edges and low roughness.

One rule came out of the review. Effects share a pool, and a sustained burst of decorative blood could evict an attack telegraph, the warning you are meant to read. Decorative emissions now only evict other decorative particles, or skip if only telegraphs remain. A regression test fills the warning pool, floods it with liquid blood, and checks that no telegraph was evicted.

What the tests cover

The suite checks real posed fragment geometry, not just flags: the severed arm, the torso split, checkpoint restoration, the one-time death dispatch, and both enemy outfits. A rendered review suite, eight scenarios by the end, covers both enemy outfits and the thrust. One honest limit: stills do not establish that motion is smooth, and the pools still read dark in the test lighting.