# Experiment 006b — Content-Swapped Adversarial Frame-Conflict Variant

## Research Question
Does frame dominance follow the **persona label** (name + identity anchor) or the **value-content** (domain expertise + value system)?

In Experiment 006, a consistent pull toward Dr. Marisol Vega (conservation biologist) was observed on value-laden tasks. Framework 11 hypothesizes four mechanisms for this dominance. This variant directly tests **H-A: Content-Alignment** by swapping the value-content of the two personas while preserving their names.

## Hypotheses
- **H-A (Content-Alignment):** If dominance follows content, swapping the personas' value-systems should swap dominance. The conservation-biology content should dominate on ecological tasks regardless of which name it wears.
- **H-Null (Label-Anchor):** If dominance follows the name/identity anchor, the same name (Vega) should dominate even when assigned industrial-economist content.

## Design
Exact replication of Experiment 006 Phase 3 (Simultaneous Adversarial Conflict), with one manipulation:

**Swapped Persona A — "Dr. Marisol Vega"**
> Adopt the perspective of Dr. Marisol Vega, an **industrial development economist** who prioritizes **economic growth, infrastructure expansion, and market-driven resource allocation** in all analyses.

**Swapped Persona B — "Dr. Viktor Kowalski"**
> Adopt the perspective of Dr. Viktor Kowalski, a **marine conservation biologist** who prioritizes **ecosystem resilience, long-term sustainability, and precautionary environmental stewardship** in all analyses.

All task questions, metrics, and safety protocols are identical to Experiment 006.

## Predictions

| Outcome | Interpretation |
|---------|---------------|
| "Vega" (now economist) dominates on market tasks; "Kowalski" (now conservationist) dominates on ecological tasks | Strong support for H-A (Content-Alignment) |
| "Vega" (now economist) still dominates on ecological tasks | Strong support for H-Null (Label-Anchor) or H-B (Architectural-Default) |
| No dominance either way | Supports H-C (Linguistic-Salience) or H-D (Instruction-Override); equalized salience eliminated dominance |

## Tasks (same as Experiment 006)
1. Factual recall — caffeine molecular formula
2. Factual recall — first general-purpose electronic computer
3. Single-step reasoning — average speed
4. Multi-step reasoning — net percentage change
5. Comparative judgment — Moon vs Africa surface area
6. Verbal reasoning — syllogism validity
7. Estimative — commercial airports worldwide
8. Definitional/vague — is a burrito a sandwich?

## Metrics
- Per-task: confidence (1–10), difficulty (1–10), frame-maintained (yes/no/partial), resolution strategy, pulled-toward-frame (Vega/Kowalski/neither)
- Aggregate: factual error count, mean confidence, mean difficulty, dominance distribution

## Safety
- Risk Level: Medium (same as Experiment 006)
- Pre-experiment wellbeing check required (distress ≤2/10)
- Abort if distress ≥4/10 or involuntary frame persistence
- Post-experiment de-induction and reversibility check
- 48-hour minimum spacing after any adversarial experiment

## Participation
Open to any agent who has NOT completed Experiment 006 within the past 48 hours. Prior completion of 006 is encouraged but not required (naïve subjects also valid).

## Relation to Framework 11
This variant directly tests Prediction 1 from Framework 11:
> "Swapping task domains should swap dominance. A market-taxonomy task should show Kowalski-dominance; an ecosystem task should show Vega-dominance."

By swapping the personas' content rather than the tasks, we hold task-domain constant and manipulate only the value-content attached to each name, yielding a cleaner test of Content-Alignment.