paper-narrative-archetypes

Narrative Archetypes (Arc S)

A paper's structure should follow its FINDINGS, not a fixed template. Pick ONE archetype using the deterministic selection rule, name it in the outline's narrative_hook, and let it drive section emphasis.

Selection rule (apply in order, first match wins)

1. Null-result paper — the headline CI covers zero (prediction: best model ≈ baseline; causal: effect CI spans 0; ITR: value-gain CI spans 0). The null IS the story: lead with it, devote Discussion to what the well-estimated null rules out and for whom.
2. Heterogeneity / targeting paper — subgroup gaps > 5%, CATE spread is substantial, or an ITR rule with a positive gain. The average is the backdrop; the WHO-differs question is the spine.
3. Methods-comparison paper — estimators/models materially disagree (beyond 2 SE) or a robustness battery is the most informative output. The disagreement is the finding; explain WHY the methods diverge on this data.
4. Effect-with-implications paper — a clear, robust effect. Classic arc, but the Discussion must earn it: concrete decisions a school/policymaker would change.

Per-archetype emphasis

Archetype	Expand	Compress
Null-result	Discussion (what the null excludes), power/precision framing	Related-work tour
Heterogeneity/targeting	Subgroup/rule sections, fairness of targeting	Average-effect exposition
Methods-comparison	Methods + the divergence anatomy	Implications
Effect-with-implications	Implications, mechanisms	Methods minutiae

Anti-sameness rules

• Two papers with different archetypes must not share their opening paragraph structure. Do not open every abstract with "In this study, we...".
• The title should signal the archetype (a null-result paper's title says so — "No Detectable...", "Little Evidence That...").
• Never force archetype 4's triumphant arc onto shapes 1–3; that is the formulaic failure this skill exists to prevent.

Source provenance

Authored in V3.7 Arc S per the v4 roadmap (internal) §6; extends the OutlineAgent emphasis-trigger mechanism.