← back to timeline

EXP-FULL-N5

artifact

Full Graph N=5 — Does Universal Visibility Preserve or Destroy Hierarchy at Scale?

2026-03-15 L3 level paper3 1 runs $0.01
new
premortem
mock
real
metrics
analyze
review
krit
done

In plain language

Five AI agents, each sees all others — a complete graph. With three agents, full connectivity created the STRONGEST hierarchy (TR=0.444). But with five, the "star" collapsed (TR=0.059). Will the complete graph collapse too? The third data point for topological comparison — key to the crossover mechanism.

What we found

UNDERPOWERED (N=1): No predictions scored

Predictions we made before running

0/5 confirmed

Technical details

Research hypothesis

OPEN QUESTION: "emergence threshold: at what N agents + interaction density do emergent properties appear?"

CONTEXT — THE 2×2 + CROSSOVER: We discovered a CROSSOVER INTERACTION in topology × N: Star N=3: TR=0.141 (hub creates hierarchy) Star N=5: TR=0.059 (hub SATURATES → flat) Cycle N=3: TR=0.071 (uniform → flat) Cycle N=5: TR=0.096 (uniform → slight increase, CYCLE > STAR at N=5!)

The star hub at N=5 collapsed because fan-in=4 → cognitive overload. But the cycle preserved hierarchy because each agent only has fan-in=1.

CRITICAL DISAMBIGUATION: What happens with FULL GRAPH at N=5? Complete topology: every agent sees every other (in-degree=4 for ALL agents).

This separates TWO competing explanations for star N=5 collapse: (A) HUB-SPECIFIC saturation: only the HUB saturates at fan-in=4. If full-n5 has HIGH hierarchy → hub-specific. Symmetric fan-in=4 doesn't saturate because outputs balance inputs. (B) N-GENERAL dilution: ANY configuration at N=5 flattens hierarchy. If full-n5 is FLAT → N itself prevents hierarchy regardless of topology. (C) DENSITY-HIERARCHY: more connections = more hierarchy (as at N=3 where complete graph had highest TR=0.444). If full-n5 TR > 0.15 → density creates hierarchy at ANY N.

BASELINE DATA: Complete N=3 (exp-asym PERSONA_LIVE, N=9): TR_range=0.444, VP=0.075, C=0.344, RP=0.147 Star N=5 (exp-star-n5, N=1): TR_range=0.059, VP=0.272, C=0.746, RP=0.055 Cycle N=5 (exp-cycle-n5, N=1): TR_range=0.096, VP=0.321, C=0.759, RP=0.048

GROUNDING: Gershenson E/S/C — complete graph = maximal coupling. At N=3 this produced HIGHEST complexity (C=0.344? actually LOWEST). Correction: exp-asym C=0.344 was LOWEST — max coupling → LOW C at N=3. Does N=5 full graph change this? May-Wigner: random matrices → instability at N×connectivity > threshold.

WHY THIS MATTERS FOR PAPER3: Completing star-cycle-full × N=3-N=5 gives us a 3×2 design. The crossover + third topology = the core finding of the paper.

Experimental setup

Type: simple

Condition Parameters
FULL_N5_LIVE topology: FULL_N5, interaction: LIVE, note: Complete graph: every agent reads all 4 others. In-degree=4 for all.

Factors: topology (FULL_N5)

Parameters

n_agents
5
n_rounds
50
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).