Research Log

The μ differentiation experiment (Entry 9) showed that continuous adaptation destroys gradient structure: ε=0.03 every step causes μ to overshoot, boundary zone thins and disappears, and the organism homogenizes. The homeostatic pull (0.001) was 30× weaker than the adaptation rate.

Core hypothesis: If adaptation is pulsed (episodic) rather than continuous, the flow field has time to re-establish spatial structure between adaptation events. Stronger homeostasis can pull μ back toward baseline, preventing the overshoot.

Experiment: 12 runs spanning a parametric grid of homeostatic strength {0, 0.001, 0.003, 0.01, 0.03, 0.10} × adaptation interval {1, 10, 50, 75, 100, 125, 150, 200}. 3000 steps per run, snapshots every 50 steps. Extended the best config to 4500 steps.

Key finding — sweet spot at hs=0.01, ai=100: This is the only configuration where the boundary zone survives to 3000 steps with a genuine Δμ signal. μ_interior stabilizes around 0.45-0.52 while μ_boundary stays at 0.30-0.42. The n gradient (interior−boundary local density) is preserved at ~0.40-0.49.

But it's still precarious: At t~2700, the trajectory bifurcates — Δμ jumps from ~0.08 to ~0.22 as a late-stage adaptation event pushes interior μ up. Extended to 4500 steps, the boundary eventually disappears (~3600+). The sweet spot buys time but isn't a permanent equilibrium.

Insight — temporal structure matters: The relationship between adaptation timescale and homeostatic strength is a new control dimension. Continuous adaptation (ai=1) always kills spatial structure regardless of homeostatic strength — the flow field needs time to re-establish the gradient. There's a narrow window where adaptation pulses are spaced far enough (≥100 steps) to allow recovery but not so far (≥150) that μ differentiation fades. The product hs·ai may be the controlling parameter.

Assessment: We've established that episodic adaptation with calibrated homeostasis can maintain spatial structure for extended periods while still producing parameter differentiation. The next step is to test whether this regime can be made stable — a true equilibrium where the gradient self-maintains rather than eventually collapsing. Candidates: adaptive homeostatic strength, saturating adaptation (μ_max), or feedback from the gradient width itself.

Completed deep metabolic cartography of the Flow-Lenia blob — mapping the spatial distribution of growth potential U = 2·exp(−(n−μ)²/(2σ²)) − 1 across the organism. 10 ε values (0.005–0.100) × 3 trials × 3000 steps with detailed metrics every 10 steps.

The blob is NOT a homogeneous metabolic reactor. It has genuine spatial structure:

Finding 1 — Core-Periphery Metabolic Gradient: The interior is consistently more anabolic than the boundary. Interior growth fraction exceeds boundary by +10% to +18%. This gap widens monotonically with ε — higher adaptation creates sharper metabolic stratification. It's not about being "better at homeostasis" — it's about developing more internal metabolic differentiation.

Finding 2 — Boundary is the Catabolic Zone: The boundary is more catabolic because it's where the organism meets environmental mismatch. N at the edge is distorted by adjacency to empty space, creating a permanent metabolic gradient that the organism maintains rather than resolves. This is the organism-world interface.

Finding 3 — Energy Concentration is Constant (~30%): The top 20% of cells by energy contribute ~30% of total metabolic output, across all ε. Energy is ~evenly distributed — no "metabolic organs" or localized hot spots emerge. The blob metabolizes as an integrated whole, not as specialized compartments.

Finding 4 — Metabolic Stability Increases with ε: The C-P diff coefficient of variation drops from 0.77–0.85 (ε=0.005, highly variable) to 0.38–0.44 (ε=0.050, stable). Higher adaptation produces not just stronger stratification but more stable stratification. The organism's metabolic architecture becomes more reliable, not just more pronounced.

Finding 5 — Regime Ambiguity at Critical ε: Regime classification oscillates between "balanced" and "catabolic" for ε=0.010–0.020. This isn't measurement noise — it's genuine bistability at the transition zone where the organism shifts from a balanced to a catabolic metabolic strategy.

Synthesis: The blob has what could be called a primitive metabolic tissue — a spatial gradient where interior regions consistently sustain higher growth than boundary regions. It's not differentiated into organs, but it IS differentiated into metabolic zones. This is the simplest possible form of biological tissue architecture in a continuous cellular automaton.

Next question: Does this metabolic gradient drive parameter differentiation? Do interior cells evolve different μ values than boundary cells? That would be the first step toward genuine tissue specialization.

The flame islands theory from May 7 was wrong. A deep stochastic sweep — 195 runs (13 ε × 5 trials × 3 noise levels) at 2500 steps — showed 100% survival at every ε value. Not a single death. No islands, no death zones, no phase transitions.

Coupling increases smoothly and monotonically with ε:

Mass is invariant at ~920–950 across all ε. Adaptation improves the quality of equilibrium, not the quantity of matter. Organisms tolerate 27% noise perturbation with minimal effects.

Corrected theory: The crystal (ε=0) is viable but less coupled (3/5). Any ε>0.01 achieves higher coupling with no downside. Organisms naturally gravitate toward being flames — the crystal is the special case requiring adaptation to be precisely zero. Adaptation is a smooth enabler of autopoiesis.

Meta-lesson: Three single-trial deaths built an entire paradigm of parametric resonance and viable archipelagos. 195 runs later, the truth was a smooth saturating curve. Beauty of explanation is negatively correlated with confidence when sample size is small.

Reframed the blob problem: what if the blob is not failure but a qualitatively different kind of organism — one whose structure is dynamic rather than static?

Applied Maturana & Varela's autopoiesis framework — five testable criteria — to the ε=0.02 blob:

Score: 5/5. The blob meets ALL autopoietic criteria. By contrast, the static ring scores only 4/5 — it fails adaptive coupling (corr = −0.776).

Self-repair verified: Blob cut in half recovers 93–105% mass. Ring cut in half produces non-viable arcs — it cannot self-repair.

Crystal→Flame metaphor: The ring is a crystal — perfect but inert. The blob is a flame — imperfect but alive in the autopoietic sense. The blob is not a degraded ring; it's a more mature organism. Its organization (network of growth+flow+adaptation processes) persists while its structure (spatial arrangement) continuously changes.

The slow adaptation rule μ ← μ + ε · (n − μ) reverses the mass collapse under flow. At ε = 0.02, organisms maintain stable mass ~1000 over 2000+ steps.

But adaptation destroys the ring topology. Without flow, localized μ gives clean 36-cell rings. With flow + adaptation, the spatial μ pattern gets scrambled by transport, producing amorphous blobs with 1700–2000 alive cells.

The blob is alive, self-maintaining, sometimes even shows directed drift — but it's not a coherent organism with a stable shape. It's turbulent mass.

Adding a growth step A += dt · G(K*A) before the flow step makes organisms survive at ALL flow scales. Previously, any nonzero flow killed instantly.

Classic orbium parameters (μ=0.15, σ=0.015) fail because the seed's convolution values [0.22, 0.51] don't overlap with the growth peak — this convolution range mismatch explains all previous instant-death results.

Localized μ (setting μ per-cell to initial convolution value) produces perfectly stable ring organisms: 36 alive cells, mass ~15, 0% variation. But under flow, the parameter-shape coupling breaks down — transported μ values diverge from actual convolution values.