1. What is Stigmergy?

Stigmergy (Greek: stigma = sign · ergon = work) is a decentralized coordination mechanism where agents communicate indirectly by modifying their shared environment. No central controller. Intelligence emerges from relationships, not nodes.

2. Platform Definition

Stigmergy is a planetary co-intelligence engine — a RAG-based AI system grounded in living systems principles, active inference, and ecological feedback. It serves humans learning to be in service of Life.

3. Proof of Paradigm (PoP) — Scale Evolution

Three progressive prototypes validate the paradigm before scaling. Only after micro-site validation (PoP 1) should PoP 2 or PoP 3 begin.

3b. Pilot Plot — Rocky Mountain Front Range

PoP 1 physical location. Bioregion: Rocky Mountain Front Range (montane grassland). Two parallel sensor tracks running concurrently to validate the biosensing architecture before wider deployment.

{
  "lat": 39.95027370248936,
  "lng": -105.16033976443673,
  "source": "pilot_plot",
  "sensor_track": "bio_hybrid",
  "readings": {
    "soil_moisture_pct": 38,
    "soil_temp_c": 9.2,
    "soil_redox_mv": 245,
    "co2_ppm": 487,
    "conductivity_us_cm": 162
  },
  "steward_note": "Optional human observation"
}

4. AI Architecture — Four Layers

AI is one layer of intelligence among many — not the primary one. Steward observation is intelligence. Indigenous seasonal knowledge is intelligence. The pheromone trace network is intelligence.

5. Active Inference (AIF) — Co-Learning Engine

Based on Kaufmann, Gupta & Taylor (2021). Collective intelligence emerges bottom-up when agents minimize variational free energy through Theory of Mind and Goal Alignment. Applied to human-AI co-learning in Stigmergy.

6. Infrastructure (Cloudflare)

Entirely browser-deployable — no local CLI, no npm, no installations. All via Cloudflare dashboard cut-and-paste.

7. Data Flow — End to End

8. All API Endpoints (103+) loading live count…

Base URL: https://stigmergy.webguru-87b.workers.dev

9. PoP Gap Analysis — Built vs Required

Honest assessment of current MVP scaffolding vs PoP requirements. Green = validated. Yellow = proxy/partial. Red = not yet built.

10. Versioning Convention

11. Agent Topology — Three Umwelten

An umwelt (Jakob von Uexküll, 1909) is the subjective perceptual world of an organism — reality as it is experienced from inside a specific nervous system. Different agent classes perceive and act on the world through fundamentally different umwelten. Stigmergy is the medium through which these distinct worlds communicate without requiring a shared language.

 ╔══════════════════════╦══════════════════════╦══════════════════════╗
 ║  NATURAL UMWELT      ║  HUMAN UMWELT        ║  AI UMWELT           ║
 ╠══════════════════════╬══════════════════════╬══════════════════════╣
 ║ Amazon basin         ║ Kayapó council       ║ Synthesis engine     ║
 ║ Great Barrier Reef   ║ Amazônia Viva NGO    ║ Bioregional monitor  ║
 ║ Sahel ecosystem      ║ AgroProdução SA      ║ Actor model infer.   ║
 ║ Arctic ice sheet     ║ Govt regulators      ║ Intervention design  ║
 ╠══════════════════════╬══════════════════════╬══════════════════════╣
 ║ Timescale:           ║ Timescale:           ║ Timescale:           ║
 ║ centuries–millennia  ║ years–decades        ║ ms–seconds           ║
 ╠══════════════════════╬══════════════════════╬══════════════════════╣
 ║ Signal:              ║ Signal:              ║ Signal:              ║
 ║ vitality, free energy║ intent, steward act, ║ query synthesis,     ║
 ║ thresholds           ║ feedback, AAI        ║ belief trace updates ║
 ╠══════════════════════╬══════════════════════╬══════════════════════╣
 ║ Epistemic weight:    ║ Epistemic weight:    ║ Epistemic weight:    ║
 ║ 0.45 (HIGHEST)       ║ 0.40 (HIGH)          ║ 0.15 (SUPPORT)       ║
 ╚══════════════════════╩══════════════════════╩══════════════════════╝
              ↕                    ↕                    ↕
        ┌─────────────────────────────────────────────────┐
        │            STIGMERGIC TRACE NETWORK             │
        │  Natural signals · Steward actions · Syntheses  │
        │  Feedback reinforcement · Emergent attractors   │
        └─────────────────────────────────────────────────┘

12. Information Flow — End-to-End

STEP 1: Natural signal (Amazon vitality declining)
  POST /api/agents/natural/signal { bioregion: "amazon", vitality_index: 38 }
  → KV: agent state updated (vitality=38, priority=warning, free_energy=0.62)
  → D1: trace deposited { type:"natural_signal", sphere:"BIOSPHERE", strength:0.8 }

STEP 2: Steward action (Kayapó reforestation logged)
  POST /api/biosignal/steward { actor_id, bioregion:"amazon", impact_score:74 }
  → D1: trace { type:"steward_action", actor_id:"kayapo-001", impact:74 }
  → KV: vitality nudged +2.1 (temporal scale respected: centuries-scale agent)

STEP 3: NGO query (soil restoration in Amazon)
  POST /api/query { query, sphere:"BIOSPHERE", actor_id, session_id, bioregion }
  → Phase 1: session belief loaded, uncertainty computed
  → Phase 2: active traces retrieved (natural + steward traces inject context)
             actor model applied (NGO: low prior_precision, high ecosystem_coupling)
  → Phase 3: natural constraint applied (Amazon warning → frames synthesis)
             timescale weights: natural >> human > AI
  → Phase 4: synthesis generated, new trace deposited
  ← Returns: synthesis + trace_context + actor_model + aif_state

STEP 4: Feedback (NGO rates 5/5)
  POST /api/feedback { query_id, score: 5, actor_id }
  → Cited traces: strength += 0.1 (reinforcement)
  → Session: sphere_affinity["BIOSPHERE"] += 1
  → Traces become attractors (convergence in Phase 4 detected)

STEP 5: Extractive query (corporation asks about soy expansion)
  POST /api/query { query:"land prep for soy", sphere:"ECONO", actor_id:"agroprod" }
  → Phase 3: Amazon warning active → natural sovereignty shapes response
             even ECONO queries cannot ignore constitutional signal
  → Synthesis: acknowledges ecological constraint, does not suppress it
  ← Score 2 feedback → actor prior_precision slightly reduced (calibrated surprise)
  → Even rejected signals leave a trace. No interaction is neutral.

STEP 6: Emergence detection
  GET /api/stigmergy/emergence
  → Attractors: BIOSPHERE/amazon traces reinforced by 3+ agent types
  → Convergence: natural + human_steward + ngo_query all converging
  → Dominant signal: "soil restoration · amazon · BIOSPHERE"
  Nobody planned this. The medium made it visible.

13. Feedback Loops

14. Regenerative vs Extractive Actor Model

Actors are not labelled. They are characterised by their Alignment–Action Index (AAI) and generative model parameters, which emerge from their declared agreements and evolve through their interactions.

15. Agent Registry — Spatial-Temporal Scale Map

Every agent in the platform operates at a characteristic spatial scale and temporal timescale. Epistemic authority increases with slower timescales. The table below shows the full registry of agent types and where they live in scale-space.

16. Multi-Agent Paradigm — Build Status

A genuine stigmergic medium — a shared environment where natural agents, human agents, and AI agents leave signals that update each other's generative models. All 8 structural gaps from PoP 2 have been addressed. The platform now operates as a full multi-agent system with Umwelt-routed ingest, indigenous sovereignty, HITL maturation, biology-closes-loops, and cross-bioregion propagation.

This section is always current. The gap table above is rendered live from GET /api/build/manifest — the authoritative source in worker.js. Rule: any code change that affects a gap or feature updates the BUILD_MANIFEST constant in the same commit.

17. HITL → LITL → Planetary Co-Intelligence

Three progressive stages, not three separate systems. Each stage is a maturation of the same feedback architecture — what changes is who or what closes the loop. The destination is not human oversight of AI, nor AI oversight of biology. It is biology, steward, and AI operating as co-equal intelligences at planetary scale.

Three Enabling Mechanisms

LITL Score — How to Read It

What Planetary Co-Intelligence Looks Like in Practice

1. Why Proxies?

Direct bio-hybrid sensor hardware is the long-term vision — living moss, fungal networks, and soil microbes as the actual sensing layer. Until physical sensors are deployed and validated, open-source satellite and environmental APIs serve as scientifically grounded stand-ins. The system never pretends to be direct sensing. Every proxy datum carries an explicit disclosure label.

2. Three-Layer Signal Model

Three signal layers ordered by authority. Layer 1 always outranks Layer 2. Layer 2 always outranks Layer 3. No algorithm overrides a human observation.

3. Proxy Data Source Stack

Six data sources ordered by priority. Open-Meteo and GBIF require no API keys and respect privacy by design.

4. Core Signal Definitions

5. Ecological Health Index Computation

The health index is a weighted composite, not a black box. Every factor is disclosed.

6. Bioregion & Umwelt Node Reference — 52 Bioregions · 122 Nodes

Signals are resolved to the nearest bioregion by haversine lat/lng lookup. Each bioregion generates 2–3 Umwelt nodes (bioregion × sphere), each representing a distinct natural agent perceptual world. Sphere vocabulary is code-controlled (slow/deliberate); node registration is KV-dynamic (fast). Total: 52 bioregions → 122 Umwelt nodes — full alignment with the One Earth taxonomy (8 realms → 14 subrealms → 52 bioregions → 185 ecoregions).

7. Future: Bio-Hybrid Sensor Integration

When living sensors come online, they slot into the existing signal hierarchy at Priority 3. The API contract is already defined. No architectural change needed — just a new signal source.

How to Use These Scenarios

These are not abstract test cases. They are real-world stories of how the platform serves people who are in relationship with living systems. Each scenario is grounded in the current API. Run them via Developer → Diagnostic or directly against the worker.

Scenario 1 — Amara Diallo · Land Steward · Sahel, Mali

"The Signal No One Else Caught." Amara notices termite mounds have shifted near her north field — a behaviour she has never seen in 12 years of observation. Soil is drier than usual for this season. She opens Stigmergy.

Scenario 2 — Dr. Rangi Parata · Researcher · Auckland / Coral Triangle

"The Researcher Who Couldn't Wait Two Days." Rangi needs to answer for a Pacific climate summit in 2 hours: what does current science say about positive tipping points for coral reef recovery after bleaching? A literature search would normally take 2 days.

Scenario 3 — Lena Hoffmann · Policy Maker · Berlin / EU

"The Policy Maker Who Needed Ground Truth." An agricultural coalition presents Lena with a polished industry report claiming EU Nature Restoration Law targets are based on outdated science. She has 48 hours to respond.

Scenario 4 — Kai · The River · Upper Mekong Watershed

"The River Speaks." Kai's flow rate is 40% below 30-year average. Three upstream dam operators are registered actors. None have logged interventions in 180 days. The river is speaking. No one registered is listening.

Scenario 5 — Yemi Adeyemi · Student → Steward · Lagos, Nigeria

"The Student Who Wanted to Understand." Yemi opens the app for the first time with a single question. Six months later he is a steward contributing to the network. No login. No form. Just a question.

Scenario 6 — Meridian Capital · Investor · Singapore · $800M AUM

"The Investor Who Had to Decide." Meridian reviews a $40M allocation across 5 bioregions. They need one credible real-time data layer to underpin nature-positive due diligence. They open Stigmergy.

Cross-Scenario: Signal Convergence

These validate that the platform learns across scenarios — that Amara's signal in the Sahel eventually connects to Yemi's observation in Lagos, and both inform Lena's policy brief in Berlin.

How These Scenarios Are Stigmergic Interactions — Not Just API Tests

Standard API tests verify endpoints in isolation. Stigmergic interaction tests verify that agents modify a shared environment and those modifications affect future agents — including agents who never met the original one.

Full End-to-End Simulation — Payload Reference

Complete payload reference for the Amazon scenario. Use the Simulation Runner tab to execute these live against the worker. Each payload below shows the full request body and expected response fields.

1. What is Active Inference?

Active Inference is a neuroscience-derived framework (Friston, 2010) in which agents continuously minimise "variational free energy" — the gap between their internal model of the world and sensory input. Applied to AI co-learning: instead of a static RAG pipeline that returns the same style of answer to every user, the system maintains a belief state about who the user is and adapts its responses to minimise epistemic surprise for that specific person.

2. Four-Phase Implementation

AIF is embedded in the query pipeline across four phases. Every query runs through all four phases in sequence.

3. Session State Reference

4. AIF in the Query Flow

5. Collective Signal — System-Level Learning

AIF operates at the individual session level AND at the collective level. The collective signal aggregates all sessions to show where the system is serving well and where it has gaps.

6. How to Test Active Inference

AIF is a system that reveals itself over multiple interactions. These tests validate that the session state is being created, updated, and influencing responses correctly.

7. Where AIF Appears in the Admin

8. AIF and the Stigmergy Metaphor

AIF is not an addition to Stigmergy — it is Stigmergy expressed in the learning layer. The session belief state is a digital pheromone trail. High-scoring sessions reinforce certain corpus paths. Low-scoring sessions let those paths fade. The network learns what works — not through central direction, but through accumulated signal.

The Core Claim

This is not a system for Nature. Nature does not need it. It is a system for humans — to remember we are in a relationship. The human asks: "How are you?" The ecosystem answers through biosignals. The human responds. This is conversation, not intervention.

PoP Evolution: Microcosm → Metacosm

PoP 1 — The Living Soil Pulse (Microcosm)

Proves the foundational claim: that distributed sensing, running lightweight on-device ML, can produce a real-time ecological intelligence signal. Currently operating with proxy APIs as stand-ins for physical sensors.

PoP 2 — The Watershed Mind (Mesocosm)

Multi-sensor data fusion across one watershed. Proves that heterogeneous ecological data streams can be integrated into a coherent multi-dimensional ecological state model.

PoP 3 — Indra's Web (Macrocosm)

Federated learning across bioregions. No raw data shared — only model weights. Data sovereignty preserved by architecture, not by contract.

AI Architecture — Four Layers

1. Web of Life Principles

Four properties describing how living systems actually work — from mycorrhizal networks to coral reefs to human communities. Stigmergy is designed to reflect, not override, these patterns.

2. Living Systems Principles — Responsible AI in Ecosystems (RAIE)

Eight principles governing how AI is permitted to behave within living systems. These are operational constraints, not aspirational statements. Each is implemented in the codebase.

3. Roles in the System

Not a conventional hierarchy. A Living Systems hierarchy — where the planet and the person in place sit above the institution and the algorithm.

4. Planetary Clock — 8 Weighted Factors

A Doomsday-Clock-inspired indicator showing planetary health direction, not just a point-in-time score. Combines 8 weighted factors across living systems, human stewardship, and ecological connectivity.

5. There is no Planet B.

This platform is not built for the platform. It is built for the relationship between humans and the living world they depend on. Every design decision, every API endpoint, every keyword in the corpus sphere taxonomy exists to serve one question: how are you? — asked of the planet, answered in biosignals, heard by a human who cares enough to ask.

Simulation Complete