2026-10-11 · OVERWHELM II · BUILD 6.66.42

Thirty-six mini-boss encounters, and hands that close on the handle

Three releases in a day. The new mini bosses went live in all 44 guardian chambers, then in a refreshed roster of 36 encounters, and character hands were rebuilt so they close round their weapons. Phones also went back to normal graphics by default.

Three builds went live on the browser version today: 6.66.39, 6.66.40 and 6.66.42.

Mini bosses in every chamber (6.66.39)

The five newly sculpted mini bosses now cover all 44 guardian chambers, with a shared entrance and death presentation, and the new skeletal horse is on Gravetide. The player Knight is unchanged.

Entrance acting and pre-finisher hit reactions come from stock animation clips. Existing attack timing, phase changes, finishers and severing stay in charge, so the new bodies change how a boss looks, not how it fights.

One bug here is worth keeping. Some retargeted clips came out distorted, and the cause was scale baked into the source bind matrices. The checks: all five models and their lifecycle tests, 549,592 sampled posed vertices, 44 chamber mappings, five real entrance-and-finisher encounters with zero errors, and a live pass through the menu, character select and Level I.

A roster of 36 (6.66.40)

The temporary repeated roster in Levels 2 to 10 is gone. Each of those levels now has two new designs, 36 encounters in all: 18 newly sculpted mini bosses and 18 evolved versions of existing boss bodies, with changed silhouettes, materials and effects. The new cast includes drowned sailors, quarry creatures, forest hunters, foundry workers, corrupted scribes and spectral guardians. Names that were already approved stayed.

Every encounter has an entrance and a finishing sequence, plus attacks with a readable warning and a recovery. Two details:

  • Ninefold's serpent fan needed corrected torso weights and independent motion.
  • The Echo uses a copy of the player's own character.

Each of the 36 encounters passed a real game test covering its imported model, entrance, attack and lethal finisher, with zero reported game errors. Posed-point sampling covered 970,688 points across the roster, and the original five bosses passed their 549,592-sample regression again.

The 18 new models are about 241 MB together, so they load by level, and models from earlier levels are now released instead of piling up. The production build covers 364 modules.

Hands that close on the handle (6.66.42)

The imported character bodies came with flat, open hands and no finger bones, so nothing could close round a weapon. A first quick fix in 6.66.41 turned the right hands the wrong way round.

Now the hands are rebuilt when each body loads. The game finds the thumb and palm side from the mesh itself, turns the hand so the fingers hang down with the palm toward the body, and curls the fingers round the point where every weapon is held. The Butcher's hands close on the axe handle, the Archer's bow hand wraps the bow, the Tank has both hands on the hilt, the Wizard grips his staff, and an empty hand hangs as a loose fist. All nine imported bodies were checked close up from the front and both sides.

Phones look sharp again

Phones had been defaulted to the low graphics setting, which draws at half to 70 percent size and turns off light shafts and contact shadows, so everything looked fuzzy. They now start on normal quality like every other device. The memory cuts that keep iPhone Safari alive, a smaller crowd, smaller textures and lower-rate audio, stay.

The earlier removal of dithering was suspected, but it is not phone-specific and was not the cause. Reading the live game on a desktop browser showed full quality, with scale 1.0 and ambient occlusion and volumetric light on. Why phones also lagged is not proven. It needs a real phone to measure.

A score after every round

The game is now scored in 15 areas, each out of 10, including combat, bosses, visuals, audio, phones, co-op and balance, with a written rubric. No area can score above 8 until the studio has played and approved it. The first round puts the game at 64 percent. The lowest areas are co-op and online (3.5, because local co-op is built but online play exists only as a write-up), balance (5.0, since the Watch AI almost never clears Level 1) and phones (5.5, never tried on a real phone). These scores come from the logs, not from play-testing.