← back to timeline

EXP-SR-LONGEVITY

artifact

Self-Repair Longevity — Memory vs Momentum (L4)

2026-03-25 L4 level paper3 1 runs $0.01
new
premortem
mock
real
metrics
analyze
review
krit
done

In plain language

This experiment tested whether a group's established structure, like a social hierarchy, is a lasting memory or just a temporary momentum from a previous state. By observing the group for a longer period after removing a key member, researchers aimed to see if the structure remained stable (memory) or faded over time (momentum). However, the experiment was not able to provide a definitive answer due to limitations in its design.

What we found

UNDERPOWERED (N=1): No predictions scored

Predictions we made before running

0/5 confirmed

Technical details

Research hypothesis

MEMORY vs MOMENTUM HYPOTHESIS (L4 — follow-up to exp-self-repair)

exp-self-repair (N=1) showed seamless self-repair: keystone removal from N=5 group caused NO disruption. Post-ablation N=4 maintained 3.2x more hierarchy than fresh N=4 (TR_range 0.187 vs 0.059).

BUT: 20 post-ablation rounds is too short to distinguish: (A) STRUCTURAL MEMORY: hierarchy is encoded in shared text history and persists indefinitely. TR should plateau at stable level. (B) MOMENTUM: hierarchy is a fading echo of the 5-agent phase. TR should DECAY toward fresh-N4 baseline (0.059) over time.

PROTOCOL: same as exp-self-repair but with 60 post-ablation rounds (3x longer observation). Total: 20 stabilize + 60 observe = 80 rounds.

CRITICAL TEST: compare TR_range in three windows:

  • Early post-ablation (rounds 21-30): expected high (momentum OR memory)
  • Mid post-ablation (rounds 41-50): if MOMENTUM → declining; if MEMORY → stable
  • Late post-ablation (rounds 61-80): if MOMENTUM → near baseline; if MEMORY → still elevated

WHY THIS MATTERS: If memory → L4 is genuine (organism maintains itself through encoded structure) If momentum → L4 is weaker (organism is just slow-decaying L3 tissue)

COMPARISON BASELINES:

  • Fresh N=4 (exp-quad): TR_range 0.059 (the floor)
  • exp-self-repair post-ablation: TR_range 0.187 at rounds 21-40
  • exp-pent N=5 stable: TR_range ~0.5 (the ceiling)

EXPLORATION TIER (N=1). Goal: trajectory shape (plateau vs decay).

Experimental setup

Type: two_phase_ablation

Condition Parameters
SR_LONGEVITY persona: PERSONA, interaction: LIVE, ablation: keystone

Factors: n_agents_phase1 (5) × n_agents_phase2 (4) × ablation_target (keystone)

Parameters

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