← back to timeline

EXP-DOSE40

partial

Integrative Dose-Response — 40% (2/5) Integrative Agents

2026-03-09 L2 level paper3 1 runs $0.01
new
premortem
mock
real
metrics
analyze
review
krit
done

In plain language

We know one weaver in a group of five halves the pecking order. What happens with TWO weavers? Does hierarchy collapse entirely, or do the weavers interfere with each other? This maps the dose-response curve: how much integration does it take to dissolve emergent structure?

Predictions we made before running

0/4 confirmed · 3 refuted

Technical details

Research hypothesis

DOSE-RESPONSE CURVE for integrative fraction → hierarchy suppression.

Current data points:

  • 0% integrative (N=3,4,5): TR range ~0.5 (robust hierarchy)
  • 20% integrative (1/5, exp-engineer): TR range 0.240 (54% reduction)
  • 25% integrative (1/4, exp-quad): TR range 0.059 (89% reduction)

CRITICAL GAP: 20%→25% shows 10× drop in hierarchy. But these are DIFFERENT N. Is the jump due to fraction or N-specific effects?

This experiment: N=5, 2/5 integrative (40%). Same N as exp-engineer (1/5). Direct comparison isolates FRACTION effect at constant N=5.

Three competing predictions: H1 (LINEAR DOSE): 40% produces proportional reduction. TR range ~0.10-0.15. Extrapolation: 20%→0.240, so 40% ≈ 2×reduction → ~0.05-0.15. H2 (THRESHOLD): Phase transition already occurred at ~20%. 40% → flat (<0.05). The 20%→25% jump was real; 40% is well past threshold. H3 (DIMINISHING): First integrative agent does most work. Second adds little. TR range 0.15-0.24. Two integrators compete/interfere rather than compound.

Confound audit:

  • Changed variables: Delta persona (challenger → curator) + Epsilon (synthesizer stays)
  • Same N=5, same seeds, same model, same Alpha/Beta/Gamma
  • Two DIFFERENT integrative styles to avoid identical-persona confound
  • Position confound: Delta=pos3, Epsilon=pos4 (both late in rotation)

Open question addressed: "emergence threshold: at what N agents + interaction density do emergent properties appear?" Sub-question: integrative fraction as emergent property SUPPRESSOR — dose-response characterization.

Experimental setup

Type: simple

Condition Parameters
DOSE40_LIVE persona: PERSONA, interaction: LIVE, n_agents: 5

Factors: n_agents (5)

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).