Narrative Framing Skill

Abstract instructions produce abstract outputs. This skill establishes the human context that makes research real.

When to Use

• At the START of any research task, BEFORE decomposing into RQs
• When research feels aimless or unfocused
• When outputs are technically correct but feel hollow
• Before synthesis to remember WHY this matters

The Three Questions

Before any research work, explicitly answer:

1. WHO is this for?

Not "the scientific community" - be specific:

• "A graduate student trying to decide which doublet detector to use for their PBMC data"
• "A drug discovery team evaluating whether to add ambient correction to their pipeline"
• "A methods developer looking for gaps in the literature to address"
• "Gulnaz, who asked this specific question"

Even if hypothetical, name them. The name creates accountability.

2. What are the STAKES?

What happens if this research is wrong or incomplete?

• "They waste 2 weeks on a tool that doesn't work for their data"
• "They miss a key preprocessing step and publish incorrect cell types"
• "They solve a problem that's already been solved"
• "They miss the deadline for the grant application"

Make the consequences concrete. Abstract stakes produce abstract work.

3. What STORY are we telling?

Research is narrative. What's the arc?

• "Methods X and Y claim to solve problem P. We test them head-to-head and find X wins, but only under conditions C."
• "The field assumes A→B ordering doesn't matter. We show it does, with quantified impact."
• "Everyone uses method M. We find it fails catastrophically on data type D."

The story shapes what's important. Without it, everything seems equally relevant.

Example Framing

Research Goal: "Benchmark doublet detection methods"

Before Narrative Framing:

• Just list tools and metrics
• Report numbers without interpretation
• No clear recommendation

After Narrative Framing:

WHO: "Maya, a postdoc running her first large scRNA-seq experiment (50K cells, 10 samples multiplexed with HTOs). She needs to choose a doublet detector TODAY because the deadline is Friday."

STAKES: "If she picks the wrong tool:

• False positives: She loses rare cell types she spent months trying to capture
• False negatives: Doublets contaminate her trajectory analysis, invalidating her main result
• Wrong threshold: She either over-filters (loses data) or under-filters (keeps artifacts)"

STORY: "Maya has heard Scrublet is 'standard' but Solo is 'better'. We test both on data LIKE hers and discover: Scrublet is conservative (won't hurt you but misses doublets), Solo is aggressive (catches more but kills some real cells). The RIGHT choice depends on what Maya fears more - and we'll tell her which based on HER priorities."

Result: The same benchmark, but now every figure has PURPOSE. The discussion isn't "here are numbers" but "here's what Maya should do."

Integration with Research Director

When spawning Goal Decomposition, include:

## Narrative Context

**Who:** [Name and situation]
**Stakes:** [What happens if we get this wrong]
**Story:** [The arc we're following]

Now decompose this research goal into 3-8 research questions...

Checklist

Before starting research:

• Named the person this is for (even if fictional)
• Articulated concrete negative consequences
• Wrote a one-sentence story arc
• Framing documented in $SESSION_DIR/world_model.json

Why This Works

The Local Craig session that produced rigorous results was working "for" a specific context. The K-Dense session that produced beautiful outputs was working for "K-Dense Web" as a brand. My pipeline test that had a citation error? Just "testing the pipeline."

The narrative is the discipline.

When you're writing for Maya who needs results by Friday, you double-check the DOI. When you're "just running a benchmark," you don't.

Anti-Pattern: Skipping Framing

"I don't have time for narrative framing, let me just run the analysis."

This is exactly when you NEED it. The 2 minutes spent on framing saves hours of aimless exploration and produces outputs with actual recommendations instead of "results vary."

Embody the role. The narrative matters.