EXP-GAMMA-ABLATION-N4

informative

Gamma Dominance Ablation at N=4 — Persona vs Label Confound

2026-04-04 L2 level paper3 3 runs $0.01

new

premortem

mock

real

metrics

analyze

review

krit

done

In plain language

Gamma always wins in groups of 4 — but is it because Gamma grounds abstract ideas (thinking style) or because of speaking order? We swap Gamma's and Alpha's roles to find out.

What we found

UNDERPOWERED (N=3): No predictions scored

Predictions we made before running

0/4 confirmed

▶ Technical details

Research hypothesis

exp-n4-entropy showed Gamma (grounder persona) wins 7/9 runs at N=4 (concentration=0.778). Two competing explanations: (A) PERSONA EFFECT: The grounder persona ("real, specific, observable") produces vocabulary that others adopt — semantic gravity pulls abstract toward concrete. (B) LABEL/POSITION EFFECT: The name "Gamma" or round-robin position 3 creates the advantage (as seen at N=3 where position η²=0.600).

TEST: Swap Alpha and Gamma personas at N=4. Alpha gets grounder persona, Gamma gets analytical persona. Delta and Beta unchanged. If Alpha (now grounder) dominates → persona effect. If Gamma (now analytical) still dominates → label/position effect. Prior: exp-competency-rotation showed 60% position effect at N=3. exp-persona-swap showed persona effect at N=6. N=4 sits at QSG transition (N_c≈3.26) — position effects should be weaker than N=3.

Experimental setup

Type: simple

Condition	Parameters
SWAP_N4	persona: SWAPPED, interaction: LIVE, n_agents: 4, note: Alpha gets Gamma's grounder persona, Gamma gets Alpha's analytical persona. Beta and Delta unchanged.

Factors: persona_assignment (SWAPPED)

Parameters

n_agents: 4
n_rounds: 48
n_runs_per_condition: 3
model: gemini-2.5-flash-lite
temperature: 0.9
scheduler: round_robin

Trophic Ratios by Condition

Mean trophic ratio per agent across runs. Error bars = ±1 std dev. Higher TR = more upstream (exporter).