authoring-brainpalace-docs

# Authoring BrainPalace interface docs The deterministic generator owns the **machine regions** (frontmatter `parameters:`, `GENERATED:*` blocks). You write the **human prose** only — never edit inside a `GENERATED:` block. ## Flow 1. Run `brainpalace sync-docs --fix`. It regenerates machine regions and prints `PROSE-NEEDED: <id> -> <doc> (<reason>)` for anything that needs prose. 2. For each PROSE-NEEDED item, open the doc and write, **outside** any GENERATED block: - `## Purpose` — one short paragraph from `brainpalace <cmd> --help` (what it does, not why). Use only flags/behavior that actually exist. - `### Examples` — 1–3 real invocations. 3. Write prose **only where it is missing or dangling** — do not reword existing valid prose (keeps commits churn-free). 4. Re-run `brainpalace sync-docs --check` until it prints `doc-sync: OK`. ## Rules - Never invent flags/commands/modes — verify against `--help` / `sync-docs --check`. - Never edit inside `<!--GENERATED:…-->` blocks (the generator overwrites them). - English only; match the surrounding docs' tone. ## Confirming an UNVERIFIABLE claim as human-verified The doc-verifier fails closed: there is **no manual "mark confirmed" backdoor**. A prose claim becomes human-confirmed only by reaching the **`audit` source tier** — recorded `verdict: SUPPORTED, grounding_tier: audit` in `scripts/doc_verify_cache.json`. The audit tier engages only when **both** hold: 1. The claim is **`unresolved` tier** — its grou...