EXP-TURNOVER-R1

STRUCTURAL PERSISTENCE THROUGH COMPONENT TURNOVER (L4 — organism-level)

BACKGROUND: L3 self-organization is ROBUST: N=5 groups always form hierarchy (4/4 runs, TR_range ~0.2-0.5). But L4 homeostasis as tested via persona-swap is REFUTED: groups CREATE hierarchy but don't preserve WHO occupies which position (pooled ρ≈0.025 for rank preservation after persona rotation).

KEY INSIGHT: self-organization ≠ homeostasis. The L3/L4 boundary is now sharp. But there's a THIRD possibility between "hierarchy always forms" (L3) and "specific role assignments persist" (L4-homeostasis, refuted):

ORGANISM-LEVEL STRUCTURAL PERSISTENCE: the GROUP-LEVEL PATTERN (hierarchy depth, number of distinct roles, interaction topology) persists even when individual AGENTS are replaced — like biological organisms where cells turn over but organ structure persists.

This is distinct from:

• L3 self-organization: hierarchy forms de novo each time (no memory)
• Refuted L4 homeostasis: specific agents stay in specific roles
• THIS hypothesis: the group's STRUCTURAL SIGNATURE survives member turnover

exp-sr-longevity showed structural memory 2.7x baseline at round 80 after keystone REMOVAL (N=5→N=4). But removal changes group SIZE, which confounds. Member REPLACEMENT keeps N constant — isolates structural memory from size effects.

PROTOCOL: Phase 1 (rounds 1-20): N=5 agents interact, establish hierarchy. Replacement (round 20): remove keystone (highest TR), ADD a fresh naive agent (Zeta, no shared memory of prior rounds, new persona). Group stays at N=5. Phase 2 (rounds 21-60): observe the reconstituted N=5 group for 40 rounds.

CONTROL: fresh N=5 group with same personas (including Zeta) starting from scratch. This isolates what the group's HISTORY contributes vs what emerges from composition.

PREDICTION LOGIC: If organism-level structural persistence exists:

• Post-replacement group should show FASTER hierarchy recovery than fresh group
• Post-replacement hierarchy depth and role count should MATCH pre-replacement
• The naive newcomer should be "absorbed" into the existing structure (assigned a role by the group, not dominating/being ignored)

If only L3 self-organization:

• Post-replacement group should look like any fresh N=5 group
• Structural metrics should show no advantage from shared history
• Newcomer integration should be indistinguishable from fresh-group formation

BIOLOGICAL ANALOGY: stem cell transplant. The organism's structural pattern (organ layout, tissue architecture) survives individual cell replacement. The new cell differentiates according to the organism's existing plan, not its intrinsic program. If the fresh agent adopts a role consistent with the GROUP's existing hierarchy (not just its persona)— that's L4.

WHY THIS IS BETTER THAN PERSONA-SWAP: Persona-swap tested: does shared memory override persona instructions? Answer: NO (ρ≈0). But that's testing the wrong thing. Persona-swap asks about INDIVIDUAL identity persistence. This experiment asks about GROUP pattern persistence. The organism isn't defined by which cell does what, but by the fact that the PATTERN of cell types is maintained.

EXPLORATION TIER (N=1 per condition). Goal: does the phenomenon exist?