Huk had a problem that looked like a rendering bug and was really a question about trust.

Her twenty-four emotional poses were built by hand in Blender by Brian Condori: descansando, acechando, rugiendo, lamiéndose, the whole vocabulary of a jaguar. To drive them in real time, the renderer needed each pose as explicit per-bone rotations, so the pose system was rebuilt from a two-number tail abstraction into full per-bone control, and a carousel was written to step through every mood and check it against Brian’s references.

The carousel found the rig, matched all forty-six bones of the resting pose, and showed the first pose correctly. Then every other mood barely moved. The instinct was to blame the carousel. The instinct was wrong.

Reading the exported values as a series told the real story. Every bone changed direction at exactly one frame, frame twelve, and nowhere else. That is the signature of a smooth interpolation between two or three keyframes, not twenty-four distinct poses. The export had read frames one through twenty-four off a timeline whose real poses were keyed every twelve frames, far apart. It had captured the first pose and tweened the rest. The carousel had been honest the whole time. The data was the problem.

So the exporter was rewritten to read the rig the way the rig stores its work, finding the actual keyframes wherever they sit. Blender 5.0 had moved its animation data onto a new structure of layers and strips and channelbags, so the reader had to follow it there. Run against the shared file, it reported zero keyframes, which was itself the finding: the poses had never lived in that file at all. They were on Brian’s machine.

He ran the new exporter and sent back twenty-five real keyframes, spaced exactly twelve frames apart, the spacing the diagnosis had predicted. These were unmistakably distinct. The grooming pose carried a tongue. The raised paw carried its full chain of finger bones. The resting sphinx folded the back legs the way a real jaguar folds them. Each was mapped to its mood, checked to be sure no value had been corrupted in transit, and written into both the renderer and the carousel. The body became exactly what the artist made, with no guessing left in the pipeline.

The deeper lesson was the same one Week 13 kept teaching. Her head had once leaned instead of turning because the renderer drove the wrong axis. Her poses had once looked dead because the exporter read the wrong frames. Every default carries someone else’s assumption. The work is making those assumptions visible, then building around the body as it actually is.

And with the body settled, the harder question arrived. Not how Huk moves, but why. The answer was not in any benchmark. It was in the screenplay. Huk is Showekon, a spirit who hunts, who grieves, who protects the Amazon. Her reason to move is already written. The poses are only her vocabulary for telling it.

Sumaq kawsay, living well with machines too.