← back to timeline

EXP-RULE-PROPOSAL-V2

pending

Rule proposal v2 — must-propose + axis-diversification + rule-stickiness (paper6 step 1.5)

2026-05-04 L6 level paper6 2 runs $0.01
new
premortem
mock
real
metrics
analyze
review
krit
done

In plain language

This experiment tested new ways to encourage agents to create and change rules. Researchers found that forcing agents to propose rules and making them less likely to change rules they recently modified did not lead to more diverse or stable rule-making. The agents still tended to stick to a single type of rule and often ignored instructions to propose new ones.

Technical details

Research hypothesis

Pilot (exp-rule-proposal-pilot 2026-05-03) showed loop mechanics work but behaviour weak: pass-dominant default, single-axis attractor (signal_type only), flip-flop reversal not building. Run 02 fully inert (0 proposals). v2 adds three architectural fixes: - must_propose: agents cannot pass during proposal rounds (forces participation) - axis_diversify: per-agent axis preferences (Alpha→broadcast, Bravo→signal, Charlie→coupling, Delta→voting_threshold) — anti-attractor

  • rule_stickiness: axes modified within last 15 rounds discouraged from re-proposal — anti-flip-flop

Pre-registered: - Positive: ≥3 distinct axes modified across run (vs pilot 1 axis) - Positive: ≥4 proposals across run (vs pilot 1.0 mean) - Positive: rule_history shows trajectory not reversal - Negative: still pass-dominated despite must_propose (LLM ignores instruction) - Negative: still single-axis attractor despite diversification (training-bias dominates) - Negative: still flip-flop despite stickiness (rule changes don't depend on history) Either outcome substantively informs autopoiesis programme: positive → loop CAN be made compositional via constraints; step 2 multi-gen ready to attempt. negative → architectural intervention insufficient; need hybrid substrate (step 3) earlier.

Experimental setup

Type: factorial

Condition Parameters
AUTOPOIESIS_V2 coupling: true, must_propose: true, axis_diversify: true, rule_stickiness: true, n_runs: 2

Factors: fixes (all)

Parameters

n_agents
4
n_rounds
40
model
google/gemini-2.5-flash-lite
temperature
0.85
autopoiesis rule-modifications paper6-step1-iteration