.omo/evidence/20260722-senpi-task-model-status/README.md
ultrabrain category.Before:
task category:ultrabrain (GPT-5.6 Sol reasoning:xhigh) ...
After, plain result:
task category:ultrabrain (openai GPT-5.6 Sol reasoning:xhigh) ...
After, 72-column component:
task category:ultrabrain (openai GPT-5.6 Sol xhigh) background pending
The compact row keeps all requested identity at 72 columns by rendering the reasoning level as xhigh; the expanded/plain result retains the explicit reasoning:xhigh label.
The failing tests pin both data plumbing and output grammar: the planner must preserve the friendly name, and both plain and width-aware renderers must combine it with provider and reasoning. Package-wide and adapter-wide gates cover neighboring state, persistence, sanitization, error, and task-tool behavior. The xterm artifact proves the exact user-visible terminal row at the width used by the existing task renderer tests.
The broad live Senpi probe's three follow-up/output-sequencing failures were reproduced on unmodified origin/dev@499b52358; see live-baseline.txt.
verification.txt; they are not presented as a pass for those behaviors.