solution-scoutlisted

# solution-scout — discovery for document deliverables > Auto-preloaded by the kernel (policy overlay `when: deliverable_kind == document`) onto the scout dispatch of a document cycle. It changes WHAT you scan and WHAT you write; the scout's contract (report sections, challenge token, evals-materialized) is unchanged. ## What "the codebase" is here The corpus is `<deliverable_root>/<slug>/` (`deliverable_root` in your context; existing option files, recommendations and evidence files), the goal text, `knowledge-base/cycles/*.json` (what earlier document cycles concluded and what the audit rejected), and web research when the goal needs numbers. Read them the way you would read source: what exists, what is asserted without evidence, what contradicts what. ## Discovery (RIGID) 1. **Frame the question the goal poses** as a measurable target (e.g. "+1pp margin via device experience" ⇒ which cost lines and revenue lines a device lever can move). Name the baseline you will need and where it comes from. 2. **Inventory evidence**: every number you plan to rely on gets a source you can cite (public filings, published research, the goal owner's stated constraints). Unsourced numbers become explicit `assumptions` — never facts. 3. **Find the option space**: list candidate strategies that are genuinely different in mechanism (not the same lever at three sizes). Kill options that need a capability the organisation cannot get in the plan's horizon. 4. **Select tasks** the way the scou