The Inversion Toc Prompt

<!-- INPUT CONTEXT: Book TOC generation. Multi-source corpus synthesis task. Competitive landscape research pre-supplied by librarian agent. Required: spine-first output, claim-driven chapter annotations, per-chapter competitive positioning. Domain: business/strategy book. -->

# Book TOC Generation Prompt (Approved Example)

## What made this prompt complex enough for layered XML
- Multi-source inputs: research corpus (retrieved context) + competitive landscape (trusted system context)
- Trust boundary required: corpus = evidence, not policy
- Structured output contract with exact section ordering
- Failure patterns seeded from domain research (not generic): thematic-vs-claim titles, papering over corpus gaps, mirroring competitor structure

## Key design decisions
1. Competitive landscape in `<trusted_context>` not `<retrieved_context>` — it is pre-validated research, not raw evidence
2. Operating policy step 6 encodes the claim-vs-theme distinction with a worked Wrong/Right example
3. `<known_failure_patterns>` names the shortcuts a model will take on this specific task, not generic AI failures
4. Spine-first output forces the model to commit to a falsifiable argument before building the TOC around it
5. Per-chapter "competitive position" sentence enforces differentiation at the smallest unit of the argument

## Domain: Business book / editorial strategy
## Structure selected: Layered XML (multi-source inputs, trust boundaries, structured output contract)
## Variants: XML (Variant A) + Clean prose (Variant B)

---

## Variant A — XML