EXP-L0-NULL-SUBSTRATE-BY-CONSTRUCTION-BASELINE-PORT

… pending

L0 Null Substrate By Construction Baseline Port

2026-06-01 L0 level 36 runs $1.64

new

premortem

mock

real

metrics

analyze

review

krit

done

In plain language

This experiment tested whether the underlying system, without any pre-assigned roles or characteristics, could spontaneously develop its own roles. Researchers created agents that were only identified by numbers to see if they would still converge on specific behaviors. The results are still pending, but if the agents do not show any convergence beyond random chance, it would suggest that any role development observed in previous experiments was due to the initial setup rather than the system's inherent capabilities.

▶ Technical details

Research hypothesis

Madigan setup is engine-source-audit-CLEAN by construction (agents labelled by number only, no axis_per_agent, no persona, no role-content) — re-implementing rule_multigen with this minimal-injection setup produces the L0 null required to attribute residual role-discovery to substrate vs engine-injection AT THE AGENT-IDENTITY LEVEL (strictly deeper than yesterday's axis_per_agent factor).

Combined with bg PID 51414 (exp-axis-diversify-false, currently RUNNING), the 2 arms fully partition the engine-injection surface:

axis_diversify=True × Alpha/Bravo/Charlie/Delta = existing autocatalytic archive (control)
axis_diversify=False × Alpha/Bravo/Charlie/Delta = bg PID 51414 (in-flight)
axis_diversify=True × Agent-1..4 = N/A unless explicitly requested
axis_diversify=False × Agent-1..4 = THIS MANIFEST

Full 2-factorial = $0.60 cost ($0.30 here + $0.30 already-paid).

Predict: if substrate has any role-discovery capacity at all, number-labelled agents produce ≥10% per-axis convergence above uniform-25% null. If number-labelled agents produce flat ~25% per-axis distribution, ALL apparent role-discovery in rule_multigen archive is engine-injected → paper6 / paper8 / paper9 §limitations gain «engine-injected at agent-identity level».

Experimental setup

Type: l0-null-2condition

Condition	Parameters
L0_NULL	identity: numbered, axis_diversify: false, must_propose: false

Factors: identity (numbered)

Parameters

n_agents: 4
n_rounds: 100
model: google/gemini-2.5-flash-lite
temperature: 0.85

Trophic Ratios by Condition

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

construct paper6-related paper8-related paper9-related pool-drained phase7-3-generated