EXP-MULTIGEN-COUPLING-ABLATE

… pending

Multi-gen coupling-ablation control — discriminating mirror vs generator (paper6 step 2 control)

2026-05-04 L6 level paper6 2 runs $0.02

new

premortem

mock

real

metrics

analyze

review

krit

done

In plain language

This experiment tested whether a specific learning pattern arises from agents interacting with each other or from their underlying training. Researchers disabled the direct interaction between agents to see if the pattern would still emerge. The results are still pending to determine if the pattern is created by agent interaction or if it's already present in their training.

▶ Technical details

Research hypothesis

exp-rule-multigen showed reproducible signal_type chain progress→combined→concept in 2/2 runs and depth ≥3 across both runs. Raw OEE-falsifier passes but cannot discriminate between (a) coupling-generated developmental program (generator) and (b) training-distribution attractor independent of coupling (mirror). Ablation: initial broadcast_to = NONE, axis locked. Agents see no neighbours during regular signal turns (regular state_view is empty). Proposing/voting loop continues. signal_type / coupling_strength / voting_threshold can still be modified. The coupling-mediated pathway is severed but the retrieval-pull is preserved. Pre-registered predictions: - GENERATOR (positive): signal_type chain DOES NOT replicate. Either zero proposals on signal_type, or different chain (e.g. progress→state instead of progress→combined→concept). Coupling is what produced the trajectory.

MIRROR (negative): signal_type chain DOES replicate even with broadcast=NONE. Bravo proposes progress→combined→concept regardless of whether anyone sees anything. Pattern is in agent training, not in coupling.
PARTIAL: chain partially replicates (depth 2-3 reached on signal_type but different values). Coupling shapes but doesn't determine. Generator with training-distribution scaffolding.

This is the discriminating test for step-2 verdict.

Experimental setup

Type: factorial

Condition	Parameters
ABLATE_BROADCAST_NONE	coupling: false, must_propose: true, axis_diversify: true, initial_rules_override: [object Object], locked_axes: broadcast_to, n_runs: 2

Factors: ablation (coupling_off)

Parameters

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

autopoiesis paper6-step2-control mirror-vs-generator coupling-ablation