0.148.0: Participants run gpt-6.1-sol

humanish 0.148.0 runs a computer-use participant on gpt-6.1-sol at medium effort when its study
names no actor.model, in place of gpt-5.6-sol. On the Taskly benchmark its analyses found as
many planted defects, its participants cost about a third as much ($0.071 against $0.216 median),
and its steps were no slower. Studies that set actor.model are unchanged, and the terminal
route's Codex participant keeps gpt-5.6-sol.

Added

• pnpm bench --storage local runs the Taskly benchmark on builds that keep their tasks in
  localStorage, so a page reload keeps the list. The five planted defects are unchanged. Rubric
  version 3 counts a lost-on-reload claim as invented on those builds and scores the in-memory
  builds as version 2 did.

Changed

• A computer-use participant that names no actor.model now runs gpt-6.1-sol at medium effort,
  in place of gpt-5.6-sol. On the Taskly builds that keep tasks across a reload, 15 planted and
  15 clean runs each, analysis recall read by hand was 52/75 (95% interval 0.58 to 0.79) against
  53/70 (0.65 to 0.84) for gpt-5.6-sol, and 42/75 against 48/70 by the rubric. It made no false
  assurance and no invented claim on the clean build read by hand, against one false assurance
  for gpt-5.6-sol. Its median participant cost was $0.071 against $0.216. It never reached
  Clear completed on the planted build, there or in the 2026-10-24 runs. Studies that set
  actor.model are unaffected, and the terminal route's Codex participant keeps gpt-5.6-sol.
  The evidence is docs/evidence/benchmark/2026-10-25-participant-default.md.