EXP-TELEODYNAMIC-PILOT

◐ partial

Teleodynamic Closure — Convention Regeneration After Perturbation

2026-04-01 L4 level paper3 2 runs $0.01

new

premortem

mock

real

metrics

analyze

review

krit

done

In plain language

We shuffle personality instructions between AI agents mid-conversation, then check if the group recreates its old conversation patterns. The group always builds a pecking order (that is just what this AI does), but the SPECIFIC words and phrases that became group conventions partially survive the shuffle — 1.7x more than random. The group remembers its culture even when individual roles change.

Predictions we made before running

3/4 confirmed · 1 refuted

▶ Technical details

Research hypothesis

TELEODYNAMIC CLOSURE HYPOTHESIS (Deacon TIER 1)

Do STRUCTURAL PATTERNS (not specific agent ranks) regenerate after perturbation?

L4 homeostasis was REFUTED for rank preservation (ρ=0.025, N=4). BUT re-analysis shows hierarchy ALWAYS re-forms (TR_range, F₀, VP_excess all reconverge — 3/3 metrics CI includes 1.0).

This reframes L4 from homeostasis (same agents same roles) to teleodynamic closure (constraints reproduce constraint TYPES).

Deacon hierarchy:

thermodynamic: work (trivial)
morphodynamic: self-organization → constraints (L1-L3)
teleodynamic: constraints → conditions → SAME TYPE of constraints

KEY TEST: post-perturbation convergence RATE. If teleodynamic: post-perturbation convergence slope HIGHER than fresh-start (shared memory containing prior conventions accelerates re-formation). If morphodynamic only: post-perturbation ≈ fresh-start (no carryover).

Reanalysis signal (N=4 homeostasis): post-perturbation slope = 0.00524 (2x fresh = 0.00289). This experiment: replication with swap_extremes (stronger perturbation).

EXPLORATION TIER (N=1). Goal: does the phenomenon replicate with stronger perturbation?

Experimental setup

Type: two_phase_perturbation

Condition	Parameters
CONTROL	interaction: LIVE, persona: PERSONA, perturbation: none
SWAP_EXTREMES	interaction: LIVE, persona: PERSONA, perturbation: swap_extremes

Factors: perturbation (swap_extremes, none)

Parameters

model: gemini-2.5-flash-lite
n_agents: 5
n_runs_per_condition: 1
scheduler: round_robin
temperature: 0.9

Trophic Ratios by Condition

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