| name | boc-narrative-layers |
| description | Generate multi-lens narratives for Bride of Charlie from structured data (episodes, claims, graph, social threads); support claim-constraint relaxation for alternate tellings; align book-app chat with meta-layer sources. Use when authoring BOC stories, book-app copy, or enabling others to narrate from the same evidence base. |
Bride of Charlie — narrative layers from data
Goal
Turn the evidence base (transcripts, inscriptions, graph, timestamps, social posts) into explicit narrative products. Each product names its lens (who tells, what is primary, what is background) so readers never confuse one telling with the only possible truth.
Canonical lenses (examples)
Use these as first-class modes; add more as named lenses with the same metadata block.
Lens A — Candace arc + social reception
- Spine: How Candace told the story (sequence, emphasis, phrasing where evidenced).
- Braid in: Responses and patterns on X and LinkedIn (quotes, paraphrase with source links, aggregate themes — no fabricated engagement).
- Tone: Documentary / media-studies; separate primary narrative from reaction layer (e.g. section breaks or typographic distinction).
Lens B — Erika-centered biography (chronological)
- Opening: Ground in verified facts (e.g. birth, documented life events as present in your corpus). Expand only where the data supports detail; otherwise mark gaps explicitly (“not established in source set”).
- Spine: Detailed telling of her story as supported by episodes, claims, and cited materials — not the same ordering or emphasis as Lens A unless the data aligns.
- Tone: Long-form literary or biographical; still cite or tag claims that anchor each paragraph.
Required output frontmatter (every generated narrative)
At the top of the narrative (YAML or a fixed header block):
lens_id: e.g. candace_plus_social, erika_chronicle
sources: list of files, episode IDs, inscription refs, URLs scraped or exported
claim_policy: strict (only claims present in source) | relaxed (see below)
relaxed_claims: if any — list which assertions are softened, hypothetical, or reordered for story flow
generated_at / generator: for provenance when others reuse the pipeline
Claim relaxation (“what if” narratives)
When the user wants to experiment with narrative by relaxing constraints:
- Inventory the claims the strict version depends on (bullet list).
- Declare each change: e.g. “Assume X unproven,” “Reorder events for dramatic effect,” “Omit thread Y.”
- Regenerate the narrative; label the output
claim_policy: relaxed and keep relaxed_claims explicit.
- Never present relaxed output as canonical; prefer a subtitle like Alternate telling (relaxed constraints).
This supports comparing narratives side-by-side in a book app without corrupting the integrity view.
Enabling others to generate from the data
- Publish a narrative spec template: lens description, allowed sources, claim policy, output sections, word-count band, citation style.
- Expose exports (CSV/JSON of claims, episode manifest, optional graph excerpts) with stable IDs so third-party prompts reference the same rows.
- Document minimum viable inputs so external authors know what “grounded” means for your hub.
Metaweb book app + “talk to a meta-layer expert”
When building or using conversational UI on top of the Metaweb book:
- Ground answers in
knowledge/metaweb_book.md, knowledge/Metaweb Book/*.md, and knowledge/meta_layer_schema.md (retrieve-then-answer pattern).
- Cite chapter, section, or schema heading when giving definitions or recommendations.
- Scope: If the question is about BOC facts, pull from BOC evidence exports; if about the metalayer vision, prefer Metaweb chapters; do not merge the two without labeling the blend.
- Persona: “Meta-layer expert” = helpful explainer of your published concepts and schema — not a generic futurist; defer when the corpus is silent.
Agent workflow (summary)
- Confirm lens and claim_policy.
- Pull relevant structured slices (episodes, claims, social exports, graph summaries).
- Draft narrative with clear sections and inline anchors to source IDs where useful.
- Emit frontmatter and, if relaxed, the relaxed_claims ledger.
- For chat/RAG: retrieve from Metaweb + schema (+ BOC export if asked), then answer with citations.