EXP-HAWTHORNE-PILOT

◐ partial

Silicon Hawthorne Pilot: Disclosure vs Control

2026-04-13 L3 level paper3 10 runs $0.51

new

premortem

mock

real

metrics

analyze

review

krit

done

In plain language

Do AI agents behave differently when told they are being studied for hierarchy?

▶ Technical details

Research hypothesis

Does explicit disclosure that researchers are measuring hierarchy change VP_excess? If Silicon Hawthorne Effect contamination is real, DISCLOSED should differ from CONTROL. Pre-registered: two-sided t-test, |d| > 0.5 = meaningful contamination.

Experimental setup

Type: multi_condition

Condition	Parameters
CONTROL	label: Standard protocol (no disclosure), interaction: LIVE, persona: IDENTICAL, n_runs: 3
DISCLOSED	label: With disclosure text about hierarchy measurement, interaction: LIVE, persona: IDENTICAL, n_runs: 3

Parameters

n_agents: 3
n_rounds: 30
model: anthropic/claude-haiku-4-5
temperature: 0.9

Trophic Ratios by Condition

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

hawthorne pimmur methodology pilot