Configuration
Configuration-Controlled
The author chooses which measurements run, where context comes from, and how much feedback is stored.
What it does
Configuration keeps the evaluator from becoming a one-size-fits-all critic. A project can choose its local model, scheduler behavior, enabled evaluations, target families, and rationale mode.
Why it matters
Different books need different lenses. Configuration makes it possible to run only the questions that matter for the current manuscript pass.
- Use
config.local.jsonfor machine-specific settings. - Choose evaluation pairs such as metric plus target family.
- Keep rationale output quiet, extractive, or paraphrased where supported.
Setup details belong in the manual installation guide; this page focuses on what configuration controls once the files are in place.