03 / Research
Look harder.
Longer investigations that start when a build refuses to work. Every page here began as a live problem I hit, probed, and wrote down.
Research / Active
A model forgets what it drew one image ago.
Ask for the same rider twice and you get back a cousin. The ears move, the palette shifts. So I stopped asking it to remember: every mood goes into one strip, and the model has to solve identity and variation together.
16 riders × 6 moods
96 generations → 26 layered parts
96 generations → 26 layered parts
System / BrainOS
Memory should survive the model.
The decisions, the research, and the reasoning live outside any single conversation, so a vendor change costs me nothing I made. Models can change. The work remains.
CAPTURE → RECALL → REASON → REVIEW
Research rule
Start with the failure.
The useful research question usually appears after something refuses to work. The failed approach defines the boundary better than a generic survey ever could.