Reinforcement Learning
Every review decision trains the agent. Track progress and alignment.
Current stage
Creative judgment is becoming more consistent.
80
out of 100
Maturity cap 80 — unlocks with more evaluated samples
Stage progression
Evaluated
48
Approved
33
Rejected
15
Caption rate
86%
Image rate
56%
First-pass rate
71%
Raw score
100.0
Regen penalty
-6.0
Approval rate by channel
Common rejection themes
Extracted from rejection notes. Longer notes give better signal.
Training Injection
Inject references — images, PDFs, captions, ideas — so SOMA can extract reusable creative intelligence. The original is discarded after extraction by default.