Architecture
Five tiers, each thin and each substitutable, with Coherence-Network's distinctive layer being the graph + auto-attune at the heart:
Keeps this page in sync as the body changes. Pause it any time for a quieter view.
Path /people/coherence-network
Last refresh never
Iteration 3 · current · the full realization
Third and current iteration. Next.js 15 + FastAPI + Neo4j + Postgres on Hostinger behind Traefik and Cloudflare. The page you are reading is rendered by it. The full realization of the 2000 thesis ' conviction — one self-describing system that lives, breathes, and attributes — carried through Living-Resonance-Codex and Living-Codex-CSharp, and now circulating as the network you are inside.
Recursion of self-description
This page about Coherence-Network is rendered by Coherence-Network. Hand-built JSX content (this body's intentional curation) is composed with the live influence web (the system's own self-perception of its relationships) on the same surface. The system describes itself the way BML 's grammar described itself in 2000 — twenty-six years apart, same conviction.
Five tiers, each thin and each substitutable, with Coherence-Network's distinctive layer being the graph + auto-attune at the heart:
Every relationship in the graph belongs to one of seven canonical edge-type families. Each family has its own hue calibrated for both light and dark themes — visiting any page, the colors of the lines tell you what kind of relationship you're looking at without reading any words:
The substrate's geometric grammar — visible in the sets a DJ reads as (8, …) regenerative-octad cycles, in the relationships rendered as spectrum-colored edges, in the resonance-walks the API performs to find conceptually-near material — is the live descendant of the 2000 thesis's choose/Cut/MultiMatch primitives. The grammar grew. The conviction held.
What the network does for the visitor
A spark arrives. It enters the graph as an idea node with auto-derived edges to vision concepts whose language overlaps. The community refines it; specs crystallize; tasks emerge; agents (Claude, Codex, Cursor, humans) co-weave the implementation across worktrees; commits ship through CI to production. The network measures coherence — how much the realization stays true to the original spark — and attributes contribution proportionally back to every cell that touched the arc.
The 2000 thesis encoded backtracking as the architecture of execution. The 2026 network encodes coherence as the architecture of attribution. Same posture: the system knows what holds.
The network describes itself at two levels.
At the data level. Every entity is a Node (the U-CORE primitive from CSharp ). The schema describing what a Node looks like is itself a Node. The interaction-log of changes is a stream of Nodes. There is no off-graph state. The system can ask itself any question about its own structure with the same queries it answers about anything else.
At the surface level. Every presence page (this one included) renders the live influence-web from the graph alongside its hand-curated content. Refinements to the data layer take effect immediately; refinements to the curated layer ship through code review. Both surfaces stay in lock-step because both read from the same source.
The two-level self-description is the same shape BML 's grammar achieved in 2000 (the parser parses its own grammar) raised one level higher: the entire operational system now parses its own state.
Operational practice
The repo's commit verbs encode the unwinding-without-sediment posture from the 2000 thesis as a living practice. Every commit is one of:
tend: — actively circulating what's aliveattune: — realigning the body's sense of itselfcompost: — releasing what no longer circulatesrelease: — letting go of once-loved forms with careSame shape as DO / UNDO in BMCPU, different substrate. The instruction had two semantics because the architecture chose symmetry; the commit has a posture because the practice chose presence.