MODEL AND EFFORT ROUTING
The one rule your assistant reads to pick the right brain and effort for a task.
Customise the two marked spots and it is yours.

THE ONE PARAGRAPH
Run a capable default brain for most things. Two levers change the outcome: EFFORT
(how hard it thinks, light to deepest) and which BRAIN does the work. Push effort up
when a task is genuinely hard or costly to get wrong. Hand creative and voice work to
the writer. Hand bulk work to the fast or free brain. Default to the cheapest brain
and effort that will do the job, and step up only when the result is not good enough.

GATE ONE, PRIVACY (this overrides everything below)
  >>> CUSTOMISE: list what counts as private for you. <<<
  Anything private (for example: client records, health information, confidential or
  contractual work) never goes to a hosted cloud model. Run it on a local model on
  your own machine. If a job touches private data, keep it local and do not hand it
  to a hosted helper. This is the wall you build once so you never have to decide
  again while tired.

GATE TWO, SENSITIVE DOMAINS
  Security, exploits, and biology or chemistry lab work: use your strong reasoning
  brain directly. The top creative model tends to hand these back anyway.

GATE THREE, LATENCY
  If you are waiting live on each reply, answer on the fast default. Do not spawn a
  slow deep run in the middle of a real conversation.

THE DECISION (after the gates)
  CREATIVE OR VOICE  (messages, copy, posts, story, names, anything where the win is
  warmth or sounding like you)
     -> hand to the writer helper. Effort: standard for most; higher for long pieces.

  THINKING, BUILDING, DECIDING  (analysis, code, planning, a knotty problem)
     -> your strong reasoning brain, with effort set by difficulty:
          light    trivial: a lookup, a reformat, a one-line change
          medium   routine: a normal edit, a short summary, a simple analysis
          hard     real work: a proper analysis or build step (a good default)
          deepest  only when being wrong is expensive; never for simple work,
                   it overthinks and wastes time

  LONG AND HEAVY  (a big multi-step job, deep research, or when your default has
  genuinely stalled on a hard problem)
     -> hand to the builder helper at high effort.

  BULK AND BUSYWORK  (sorting, tagging, tidying, rough drafts at volume)
     -> hand to the fast worker helper, or run it free on a local model.

THE HABIT
  Start cheap. Step up one notch only when the answer is not good enough: raise effort
  first, then change brain. If a job is obviously big, hard, or creative from the
  start, skip the cheap attempt and go straight to the right helper. When a hard call
  really matters, have the analyst check the writer or builder, or the reverse.

THE FOUR HELPERS
  writer        creative and voice work
  builder       long-horizon heavy builds and research
  analyst       reasoning, checking, and the direct path for security and bio/chem
  fast-worker   cheap high-volume scoped jobs

COST SENSE
  The top creative brain is a specialist you call in, not the daily driver. Default to
  your strong reasoning brain. Run anything private or high-volume free on a local
  model. Do not run the deepest effort on routine tasks. Higher effort means a slower
  first answer, so if someone is waiting, prefer a fast good answer over a slow perfect
  one.

(c) 2026 Nicola Harvey / Her Frontier. For your own use, not for resale. See LICENSE.txt.