Stigmergic Consolidation

Status: Active — formed Session 8 Connected to: Stigmergy, multi-scale composition, H7 (trace→actor crossing), autopoiesis, niche construction, computational irreducibility, multi-rate environment

The Concept

Stigmergic systems coordinate via positive feedback: a trace stimulates more of the same action (deposits attract deposits, pheromone trails recruit more ants). This amplifies and exploits promising developments. The claim of this file is that positive feedback alone does not reach a consolidated actor: without a counterbalancing negative feedback — a mechanism by which a stronger trace eventually reduces or redirects activity — nucleation is never pruned.

This is a theoretical claim (Heylighen) plus two directional results, not an observation of scatter. (2026-07-27: it originally read "positive feedback alone produces diffuse scatter … the system nucleates everywhere and consolidates nowhere", anchored on sim06 numbers that were wrong. sim06's baseline is 66–109 components at compactness 0.109–0.120 and misses the crossing by ≤0.05 — not a scatter. What supports the claim now is that both attempts to add feedback made fragmentation worse: sim06's self-maintenance, 219–297 components; sim07's transport field, 57→128 pillars as M_c fell.)

This is the negative-feedback gap: the missing ingredient between stigmergic coordination (within-scale) and the trace→actor crossing (H7, between-scale).

The Mechanism (from Heylighen 2016 and sim06)

Heylighen's analysis of stigmergy identifies two feedback regimes:

1. Positive (amplifying): stronger trace → more activity → stronger trace. This exploits affordances — it concentrates effort where there's already signal. Termite deposits attract more deposits; ant pheromone recruits more ants.

2. Negative (suppressing): stronger trace → less activity, or activity redirected. This controls — it prevents runaway and enforces division of labor. Heylighen's market example: more buying → higher price → less buying (self-limiting). Price is a quantitative stigmergic trace that carries its own inhibition.

Complex self-organization requires both (Heylighen: "The combination of positive and negative feedbacks is typical for complex systems"). Positive feedback amplifies; negative feedback stabilizes and diversifies. Without negative feedback, positive feedback runs to saturation or monoculture.

Correction (2026-07-27). This sentence originally continued "— exactly what sim06 produced: ~230 scattered micro-pillars with no consolidation, high cell turnover, stability ~0.55 (well below the 0.90 H7 criterion)." Every one of those figures was wrong against sim06's own results.json. Baseline is 66–109 components at compactness 0.109–0.120, stability 0.849–0.893, retention 0.98. See ../simulations/REVIEW.md §1.

sim06's Null Result (the empirical anchor — substantially retracted)

sim06 tested H7 with a minimal Grassé stigmergy model. The deposit rule p = DEPOSIT_BASE + DEPOSIT_GAIN · local/(1+local) saturates at ~0.95 but never decreases — there is no inhibition. That design observation stands.

What does not stand is the empirical anchor. sim06's crossing detector could not fire at all: criterion 2 required the deposit rate to fall below its early-run average, which Grassé positive feedback makes impossible once structure exists. It held only at samples 0–5, before any structure had formed. The parameter sweep (material_decay 0.005–0.4, deposit_base 0.005–0.05, phero_follow 0.6–0.95, maintain_gain 0.1–0.5, reload_prob 0.15–0.3) ran against that detector and therefore established nothing; it has not been repeated.

With criterion 2 corrected it passes 130/160, and the binding constraint is criterion 1 — stability 0.849–0.893 against a 0.90 threshold, a miss of ≤0.05. Criterion 3 passes 154/160 for baseline (deposit_on_structure 0.70–0.79). The structure is neither diffuse nor unselective.

The one result that does support this file's thesis is the self-maintenance reversal. Adding more positive feedback made things worse in exactly the predicted direction: self-maintenance is more fragmented than baseline (219–297 components vs 66–109) and less selective (0.43–0.53, criterion 3 failing 0/160), because maintain_gain=0.3 saturates the deposit response flat at ~0.87 everywhere and destroys the spatial contrast stigmergy depends on. sim07's null adds a second data point: a structure-sourced scalar transport field also increased fragmentation monotonically (57→128 pillars as M_c fell) rather than consolidating.

Revised diagnosis: the model has positive feedback plus weak decay and no consolidation mechanism — nothing that makes strong pillars inhibit nearby nucleation or redirect builders away from saturated regions. This remains a plausible reading, but it is now an inference from a near miss and from two failed negative-feedback attempts, not from an observed scatter. Self-maintenance builds 66% more structure than baseline (1876 vs 1131 cells) — that number is unaffected by the fix and still stands.

The Saturation Result (2026-07-27 — the sharpened claim)

This result was promoted to a hypothesis in its own right: [[hypotheses/H11]] — The Saturating Channel Hypothesis. This section is its concept-level treatment; H11 carries the formal statement, the criticisms, and the test.

This is now the core of the concept, and it came out of the code review rather than from reading. Two independent attempts to supply the missing negative feedback both made the structure less consolidated, monotonically:

The common cause is the agents' response curve, not the feedback itself. Both mechanisms act by manipulating the pheromone field, and the deposit rule p = DEPOSIT_BASE + DEPOSIT_GAIN · φ/(1+φ) is effectively flat above φ≈1. Once the field is driven high anywhere — and self-emission or venting both drive it high — deposit probability sits at ~0.87 across the entire grid. The manipulation intended to create spatial contrast operates precisely in the region where contrast cannot be expressed. Adding energy to a saturating channel removes selectivity rather than producing it.

Refined prescription: the crossing does not need "negative feedback" in the abstract — it needs negative feedback through a channel that does not saturate. Concretely, act on deposit probability or on geometry directly:

• a density cap — deposition suppressed where material already exceeds a threshold;
• a refractory period — a cell that has just received a pellet is briefly unavailable;
• directional bias — building preferentially along existing wall edges rather than onto their centres.

Each of these creates contrast that survives however high the cue field goes, because the inhibition is not mediated by the cue field. All three are cheaper to test than directed transport, and they discriminate the refined claim from the coarse one: if non-saturating inhibition consolidates where field manipulation fragmented, the saturation account is right.

Why this matters beyond sim06: it suggests Heylighen's positive/negative feedback framing carries a hidden assumption — that the two act on comparable channels. Where the response to the trace saturates, negative feedback delivered through that trace is self-defeating. That is a general claim about stigmergic systems, not a quirk of this model.

Environmental Physics Coupling (the Mahadevan model)

The Harvard/Mahadevan termite mound model (Ocko, Heyde & Mahadevan, PNAS 2019) shows the missing ingredient in real termites: the mound is not just a passive accumulation — it's a ventilation structure whose own physics channels the pheromone cues that guide building. External temperature variations drive internal airflow, which redistributes pheromones and metabolic gases, which trigger building where the mound is too warm/leaky. The structure is the feedback path: the macro-structure's physics (airflow, thermal mass) determines where the micro- scale signal (pheromone) goes. This is the trace→actor loop made concrete, but it requires the environment to have physical transport dynamics, not just decay.

In ANT terms: the mound is an actant not because it's big, but because its physical state (porosity, temperature gradients) causally reshapes the network of cues. Without that physics, the "trace" never becomes an "actor" — it stays a passive accumulation.

What This Implies for the Crossing (H7 refinement)

H7's three operational criteria (persistence despite erosion, non-reducible dynamics, constraint on agents) implicitly assume the structure has its own dynamics that aren't reducible to individual deposits. In sim06, the structure has no dynamics of its own — it's just a sum of deposits. Self-maintenance emission adds one dynamics (re-emitting pheromone), but that's still just "more positive feedback at the structure." The crossing needs the structure to do something deposits don't do: transport, channel, inhibit, or compete.

This refines H7: the trace→actor crossing requires not just that the trace recruits its own maintenance (sim06's loop), but that the accumulated structure introduces a new dynamical degree of freedom — a process that did not exist at the deposit level. Candidates:

• Saturation/inhibition: a cell above a density cap repels deposits (negative feedback). Forces consolidation into few large pillars rather than many small.
• Environmental transport: the structure channels a diffusive/advective field (airflow analog) that redistributes pheromone away from saturated regions and toward gaps — the Mahadevan mechanism.
• Competition between structures: multiple trace types or clusters compete for a finite resource (termites, material), so growth of one inhibits another.
• State transition: inert substrate becomes active only above a mass threshold (Vance's termite-mound principle) — the structure's state changes, unlocking new dynamics.

The Specific Mechanism (Session 9 refinement — environmental physics coupling)

Session 9 identified the specific negative-feedback mechanism the crossing requires: environmental physics coupling — the accumulated structure must introduce a transport dynamics that redistributes the cue field. See [[concepts/environmental-physics-coupling]]. The Mahadevan group's termite mound model (King/Ocko/Mahadevan PNAS 2015; Ocko/Heyde/Mahadevan PNAS 2019) shows real mounds are ventilation organs whose own physics (diurnal thermal convection) channels the pheromone cues that guide building — the structure IS the feedback path.

A 20-year modeling lineage (Deneubourg 1977 → Bonabeau 1997 → Ladley & Bullock 2004) shares exactly sim06's limitation: deposited material has no influence on agent movement; pheromone diffusion is decoupled from structure. sim06's null result is therefore not a failure of our model but confirmation of a known field-wide gap. The minimal lumped prescription for sim07: a structure-sourced transport field that vents pheromone away from saturated regions (negative feedback), with a mass threshold M_c below which the structure is inert (Vance's inert→active state transition). The crossing is predicted to coincide with the onset of non-trivial transport above M_c.

Relevance to the Project Arc

sims 03–05 showed composition fails without an explicit mechanism (H10). sim05 nominated stigmergy as the glue. sim06 tested the simplest stigmergic self-maintenance and found: positive stigmergic feedback alone amplifies building but does not consolidate — the trace→actor crossing does not occur. The missing ingredient is negative feedback / environmental physics coupling. This is a real, mechanistic refinement, not a failure: it tells us what sim07 must add. The "dynamic landscape" the synthesis has been pointing toward (Session 4: NK ↔ stigmergy; Vance's multi-rate environment) is not just "a changing fitness function" — it's an environment whose own physics becomes a new dynamical layer once organization crosses a threshold. The crossing is the moment the environment gains a degree of freedom it didn't have before.

Criticisms

• The negative-feedback framing is well-established in self-organization theory (Heylighen, Camazine et al.); the novelty is applying it as the diagnosis of H7's failure and the prescription for the crossing.
• It's possible H7's operational criteria (0.90 stability, 0.60 constraint) are too strict and a different operationalization would classify sim06's weak structure-size separation (66% more cells) as a partial crossing. This criticism was originally dismissed here on the grounds that "stability 0.55 is far from 0.90, and constraint 0.33 is far from 0.60 — the gap is large, not marginal." That dismissal was wrong and is retracted (2026-07-27): baseline stability is 0.849–0.893 — a miss of ≤0.05 — and constraint is 0.70–0.79, which passes the 0.60 threshold 154/160 samples. The criticism is now the stronger reading. Thresholds were deliberately not retuned after the detector fix, so as not to select a detector that produces the preferred answer, but the result should be described as a near miss rather than a failure.
• Real termites have many more mechanisms (king/queen pheromones, larval cues, tactile contact) that the minimal model omits; the null result is specific to the minimal Grassé model, not to stigmergy in general.

Empirical Evidence

• Heylighen (2016): theoretical — positive + negative feedback as the signature of complex stigmergic systems. Market price as negative-feedback stigmergic trace.
• Ocko, Heyde & Mahadevan (PNAS 2019): model coupling termite behavior to mound environmental physics (airflow, temperature) reproduces the range of observed mound morphologies. The structure's physics — not just its mass — drives morphogenesis. (No single DOI found in extraction; see seas.harvard.edu summary.)
• Dorigo et al. (ACO literature): ant colony optimization requires pheromone evaporation (negative feedback) to avoid trail saturation; convergence depends on the evaporation rate. Directly analogous to sim06's decay — but ACO's evaporation is tuned. (2026-07-27: this originally added "while sim06's decay was too weak relative to deposit rate" — an inference from the retracted scatter diagnosis. sim06's decay has not been shown to be mistuned.)
• sim06 (this project): null result, but a weak one — the detector could not fire (see the retraction above). Corrected: baseline is 66–109 components, stability 0.849–0.893, criterion 3 passing 154/160 — a near miss, not a scatter. The result that does support consolidation is directional: self-maintenance emission (more positive feedback) amplifies building by 66% while making the structure more fragmented (219–297 components) and less selective (0.43–0.53). See sim06_termite_mound.

Open Questions

• Which negative-feedback mechanism (saturation, transport, competition, state transition) is the minimal addition that produces the crossing? Each predicts a different morphology (few large pillars vs. channel networks vs. competing clusters).
• Is there a phase transition in the negative-feedback strength — below it, the sim06 regime; above it, consolidated actor? If so, where? sim07 tested exactly this for a scalar transport field and found no transition — the response was monotonic and in the wrong direction (stability 0.876→0.739, pillars 57→128 as M_c fell). Any remaining version of this question needs a different mechanism, not a different threshold.
• Does environmental physics coupling (the Mahadevan mechanism) require modeling actual transport (PDEs), or can a minimal lumped analog (e.g. "pheromone flows down a material gradient") capture the consolidation?
• How does this interact with computational irreducibility (H8)? If the structure's physics is a new dynamical layer, is the system more irreducible at the crossing — the moment the environment gains a degree of freedom?

Cross-References

• [[concepts/stigmergy]] — the base mechanism; this concept refines what's missing
• [[concepts/multi-scale-composition]] — the crossing is the composition event
• [[concepts/autopoiesis]] — self-maintenance is necessary but sim06 shows it's not sufficient without consolidation
• [[concepts/multi-rate-environment]] — Vance's termite-mound principle: inert → active substrate state transition
• [[hypotheses/H7]] — the trace→actor crossing; sim06's null result refines it
• Heylighen (2016), "Stigmergy as a Universal Coordination Mechanism"
• Ocko, Heyde & Mahadevan (2019), PNAS — termite mound morphogenesis via environmental physics coupling