Recursive-LD
as Layer Three
of Root-LD
How Recursive-LD functions as the edge and cognitive context layer
within the Root-LD federation — growing more intelligent over time
as entities accumulate relationships, records, and dimensional reasoning data.
Layer Three is the only Root-LD layer not populated at entity ingestion. It is not static. It grows. Every edge added between entities, every dimensional reasoning record written, every telemetry chain anchored — all of it is Layer Three. It is what makes the federation intelligent, not merely indexed.
The Root-LD specification defines three layers of entity structure. Every entity in the federation carries all three. Layer One is the immutable core. Layer Two is the content expression. Layer Three — Recursive-LD — is the edge and intelligence layer that evolves.
The helix metaphor is deliberate. A consuming system can enter a Recursive-LD record at any node and spiral outward — following edges upward to broader context or downward into component entities — indefinitely. The graph does not end. It deepens.
Recursive-LD edges between entities in the federation are written by four sequential passes. Each pass operates at a different level of semantic resolution. The passes are additive — later passes build on earlier ones but do not replace them. Confidence scores are written per-edge and per-pass.
Pass Conf: 1.0
Exact-match relationship detection. Writes edges where the relationship between two entities can be established with certainty from structured metadata — shared domain code, explicit parent-child declarations, federation membership links.
Confidence score: 1.0. These edges are treated as ground truth and never overwritten by subsequent passes. The bedrock of the graph.
Pass Conf: 0.7–0.9
Shared vocabulary and term co-occurrence analysis. Writes edges where entities share significant lexical surface — controlled vocabulary terms from the Root-LD lexicon, entity type labels, semantic fingerprint overlap.
Confidence scores in the 0.7–0.9 range. Edges are labeled with the lexical basis for review. Subject to revision if semantic passes disagree.
Pass Conf: 0.5–0.85
General-purpose language model semantic analysis. Writes edges based on inferred conceptual relationships that may not appear in lexical surface — thematic resonance, functional complementarity, ecosystem co-positioning.
Confidence scores in the 0.5–0.85 range. Widest edge type coverage. Highest volume. Subject to revision by finetuned pass.
Finetuned Pass Conf: 0.75–0.95
Domain-finetuned model pass. Writes and revises edges using a model trained specifically on Root-LD entity structure and federation vocabulary. Can override LLM pass scores where domain-specific context changes the relationship assessment.
Highest confidence ceiling at 0.95. Final authority on contested edges. Enables compound reasoning across the full graph.
The four-pass model is not sequential ranking — it is layered epistemic resolution. A deterministic edge and a finetuned semantic edge can coexist on the same entity pair with different confidence scores and different edge types. The graph holds them both.
Root-LD v1.0 defines 20 edge types across five semantic categories. Every Recursive-LD edge record must declare one of these types. New edge types require specification amendment at root-ld.org. Domain-level alias labels are permitted; canonical type codes are not extensible.
| Edge Code | Direction | Description | Min Conf. |
|---|---|---|---|
| PARENT_OF | directed | Hierarchical containment. Source entity structurally contains or governs target. | 0.90 |
| CHILD_OF | directed | Inverse of PARENT_OF. Target contains or governs source. | 0.90 |
| MEMBER_OF | directed | Federation membership or group affiliation. Source belongs to target collective. | 0.95 |
| COMPOSED_OF | directed | Component composition. Target entities are constituent parts of source. | 0.85 |
| Edge Code | Direction | Description | Min Conf. |
|---|---|---|---|
| REFERENCES | directed | Source entity explicitly references or cites target. Declarative relationship. | 0.80 |
| CITED_BY | directed | Inverse of REFERENCES. Target is the citing entity. | 0.80 |
| COEXISTS_WITH | undirected | Entities occupy the same ecosystem layer without hierarchical relationship. | 0.65 |
| COMPETES_WITH | undirected | Entities operate in overlapping domains with functional tension between them. | 0.60 |
| COMPLEMENTS | undirected | Entities are functionally additive — each extends the value of the other. | 0.65 |
| Edge Code | Direction | Description | Min Conf. |
|---|---|---|---|
| GROUNDS | directed | Source provides epistemic foundation or evidentiary basis for target's claims. | 0.75 |
| REFUTES | directed | Source entity provides evidence or argument against target's claims or framing. | 0.70 |
| EXTENDS | directed | Source builds on and expands the knowledge scope of target. | 0.70 |
| INSTANTIATES | directed | Source is a concrete instance or implementation of the abstract target. | 0.80 |
| ABSTRACTS | directed | Source is the abstract generalization of the concrete target. Inverse of INSTANTIATES. | 0.80 |
| Edge Code | Direction | Description | Min Conf. |
|---|---|---|---|
| PRECEDES | directed | Source entity or event chronologically or logically precedes target. | 0.85 |
| SUCCEEDS | directed | Source follows from target temporally or causally. | 0.85 |
| EVOLVES_FROM | directed | Source is a developmental or transformed version of target over time. | 0.75 |
| RECORDS | directed | Source entity is a temporal log or persistent record of target's state or behavior. | 0.80 |
| Edge Code | Direction | Description | Min Conf. |
|---|---|---|---|
| DRIFTS_FROM | directed | Source behavioral state has deviated from the baseline established by target record. | 0.60 |
| CONTAINS | directed | Source telemetry record includes an active containment or trap condition targeting target. | 0.70 |
Root-LD v1.0 defines 15 entity classes. Every entity ingested into the federation must declare one primary class. Secondary classes are permitted where an entity genuinely spans multiple types. The REO domain — Recursive-LD's home domain — focuses on the classes highlighted below.
Five domains are currently registered in the Root-LD federation.
Each domain carries a two-to-three letter domain code, a primary entity class focus,
and a set of canonical entity documents at intelligence-docs/.
Recursive-LD Layer Three edges between entities across all five domains
are accumulated through the four federation passes.
A Recursive-LD Layer Three edge record is published as a structured object
within a @graph block. Each edge record declares a source entity,
target entity, edge type, confidence score, originating pass, and timestamp.
Multiple edge records from different passes may coexist for the same entity pair.
{
"@type": "rld:RecursiveEdge",
"rld:sourceEntity": {
"@id": "https://oakmorel.com/intelligence-docs/entity.jsonld"
},
"rld:targetEntity": {
"@id": "https://rankwithme.ai/intelligence-docs/entity.jsonld"
},
"rld:edgeType": "COMPLEMENTS",
"rld:confidence": 0.82,
"rld:pass": "semantic-llm",
"rld:timestamp": "2026-03-01T00:00:00Z",
"rld:basis": "shared semantic_fingerprint terms; ecosystem co-positioning"
}
Canonical entity documents live at intelligence-docs/[domain]-entity.jsonld
on each registered domain. These documents are the machine-readable anchors
that the federation passes operate on. They include Layer One (Anchor-LD),
Layer Two (Body-LD), and any accumulated Layer Three (Recursive-LD) records.
{
"@context": "https://root-ld.org/context.jsonld",
"@graph": [
{
// Layer One — Anchor-LD (immutable)
"@type": "rld:AnchorEntity",
"rld:domainCode": "REO",
"rld:federation": "root-ld-v1",
"rld:createdAt": "2026-01-01T00:00:00Z"
},
{
// Layer Two — Body-LD (versioned)
"@type": "rld:BodyEntity",
"schema:name": "Recursive-LD",
"schema:description": "..."
},
{
// Layer Three — Recursive-LD (accumulative)
"@type": "rld:RecursiveEdge",
"rld:edgeType": "MEMBER_OF",
"rld:confidence": 1.0,
"rld:pass": "deterministic"
}
]
}
Each registered domain maintains its own intelligence-docs/ directory.
The federation passes read from and write to these documents.
The canonical entity document is both the input to the graph
and the accumulating output of it.