# Stigmergic Mesh

> Structural intelligence for financial networks. A self-organizing computational architecture (Ambient Structure Discovery, ASD) that detects structural anomalies in financial networks without labeled training data.

Stigmergic Mesh is the commercial site for Ambient Structure Discovery, a patent-pending anomaly-detection architecture grounded in Adaptive Resonance Theory and stigmergic coordination. It processes complement-coded continuous structural vectors over ownership and transaction graphs to surface topological patterns that current AML, KYC, and sanctions-screening tools are structurally unable to see.

USPTO Provisional Patent #63/981,369. Open-source core under MIT.

## Paper

- [Ambient Structure Discovery via Stigmergic Mesh (PDF)](https://stigmergicmesh.com/ambient_structure.pdf): Full technical paper. Theoretical foundations, signal construction, experimental results on Elliptic++ Bitcoin data, and deployment architecture.

## Key concepts

- **Ambient Structure Discovery (ASD)**: Detect structure without queries, labels, or predefined rules. The mesh self-organizes around complement-coded continuous structural vectors bounded in [0, 1].
- **Stigmergic Mesh**: Coordination through shared environment. Workers leave traces; other workers respond. No central coordinator. Based on Grasse (1959), Theraulaz and Bonabeau (1999).
- **Seven structural coefficients**: D_L (depth layering), F_o (fan-out), N_P (nominee proximity), J_S (jurisdiction switching), C_B (circular beneficiary), V_C (velocity compression), A_D (action decoupling).
- **Eigenvalue Collapse**: Loss of independent evaluation modes in the mesh correlation matrix. Signals mass-produced concealment.
- **Linguistic Compression**: Measurable shift from direct to hedged language in entity descriptions.
- **Action Decoupling**: Discussion frequency decouples from resolution activity — filing as a substitute for action.

## Use cases

- Shell company detection (post-CTA beneficial-ownership analytics)
- Normalized risk drift (slow-laundering detection across client behavior over years)
- Network contagion scoring (multi-hop OFAC/sanctions proximity)
- Sanctions evasion topology (detect evasion by graph shape, not entity name)
- Fraud ring detection (structurally linked accounts that never directly transact)

## Validated data sources

UK PSC Register (10M companies), ICIJ Offshore Leaks (810K entities), GLEIF Level 2 (469K relationships), Elliptic++ Bitcoin dataset (203K transactions), OFAC SDN, Companies House BasicCompanyData.

## Results

On Elliptic++: text-based controls detect 1 anomaly; complement-coded structural vectors detect 59 (59x). 2.2x licit enrichment in flagged clusters (p < 0.00002). Worker self-organization from 12 to 60 with genuine differentiation (vs. saturated/identical under text encoding).

## Contact

- Inventor: Jeremy McEntire
- Partnerships: partnerships@stigmergicmesh.com
- Code: https://github.com/jmcentire/stigmergy
- Docs: https://jmcentire.github.io/stigmergy/

## Optional

- [llms-full.txt](https://stigmergicmesh.com/llms-full.txt): Expanded version with full section text, market sizing, regulatory context, and theoretical lineage.
- [sitemap.xml](https://stigmergicmesh.com/sitemap.xml)
- [robots.txt](https://stigmergicmesh.com/robots.txt)