← back to timeline

EXP-TRIAD-DOSE

informative

Triad Dose-Response — Gradient Compression at Minimum N

2026-03-10 L2 level 8 runs

In plain language

Three identical AI synthesizers — same instructions, only names differ. Any hierarchy that emerges is purely from speaking order, not personality. This is the floor of the N=3 dose-response curve: how much structure survives when all agents are clones?

What we found

Dose-response operates at N=3: 33% dose sufficient for hierarchy (4/4 confirmed). Triad = minimum viable system.

Technical details

Research hypothesis

N=3 DOSE-RESPONSE SERIES: 0% → 33% → 67% → [100%]

100% = ALL 3 agents are synthesizers. NO persona diversity whatsoever. All agents receive identical instructions (except name).

This is the N=3 PURE POSITION TEST — parallel to N=5 exp-dose100. Any hierarchy that emerges can ONLY come from speaking order position.

N=5 at 100%: mean TR range 0.230, ratio 0.435. If inverse-N scaling holds, N=3 at 100% should show: (a) TR range > 0 (position alone creates hierarchy — structural property) (b) TR range/ratio LOWER than N=5 at 100% (amplified compression)

This is the FLOOR of the N=3 dose-response curve. Combined with 0%, 33%, 67%, it completes the full N=3 curve for cross-scale comparison.

Critical theoretical question: at N=3 with identical agents, does the minimal topology (3 agents, 3 pairwise relationships) support ANY position-driven hierarchy? With only 3 speaking positions, the first/middle/last structure is maximally constrained.

Series (4 experiments)