← back to timeline

EXP-MAY-SIGMA

success

May's σ Test — Does Reduced Interaction Restore Hierarchy at N=7?

2026-03-10 L1 level paper3 1 runs $0.00
new
premortem
mock
real
metrics
analyze
review
krit
done

In plain language

With 7 agents and 7 rounds each, hierarchy collapses (TR=0.30). But according to May, the issue may not be the NUMBER of agents, but the STRENGTH of interaction. If each agent speaks 4 times instead of 7, interaction is weaker → hierarchy may recover. If so — we confirm the May mechanism. If not — the issue is the number of agents itself, not the strength of their mutual influence.

Predictions we made before running

3/4 confirmed

Technical details

Research hypothesis

OPEN QUESTION: Does interaction strength (σ) control the phase transition?

May's stability criterion: σ√(NC) > d → destabilization. In our system: σ ∝ rounds per agent (more rounds = more vocabulary exchange = higher interaction strength). N=number of agents. C≈1 (round-robin = complete graph). d = persona adherence (self-regulation).

KNOWN: N=7 with 7 rounds/agent → TR range 0.297 (hierarchy collapsed). INTERVENTION: N=7 with 4 rounds/agent → reduced σ.

If May's framework applies:

  • Reducing σ at fixed N should move system BELOW critical threshold
  • Hierarchy should RESTORE (TR range increase toward 0.40+)
  • This would distinguish May's mechanism from pure N-driven effects

If hierarchy does NOT restore:

  • The transition is driven by N itself (crowding), not interaction strength
  • May's σ mechanism is NOT primary
  • Alternative: Kao & Couzin (information load) or simple dilution

COMPARISON BASELINES:

  • exp-hept: N=7, 7 rounds/agent, TR=0.297 (collapsed)
  • exp-pent: N=5, 10 rounds/agent, TR=0.528 (peak)
  • exp-n6: N=6, 8 rounds/agent, TR=0.483 (bistable)

NOTE: rounds-per-agent confound with TOTAL corpus length. 28 rounds (4×7) vs 49 rounds (7×7) means shorter conversation. If hierarchy restores, cannot fully distinguish "less σ" from "less dilution time". But DIRECTION of effect is sufficient for first test.

Experimental setup

Type: simple

Condition Parameters
HEPT_LOW_SIGMA persona: PERSONA, interaction: LIVE, n_agents: 7, rounds_per_agent: 4, note: Same 7 agents as exp-hept, but 4 rounds each (28 total) instead of 7 (49).

Factors: n_agents (7) × rounds_per_agent (4) × persona_type (non_integrative)

Parameters

n_agents
7
n_rounds
28
n_runs_per_condition
1
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).