01-projects/graph-db-eval

vertex edge dictionary

2026-04-14·project·status: active

Graph DB — Vertex and Edge Type Dictionary

Running data dictionary for the typed knowledge graph we're prototyping on DuckDB. Maintained incrementally — add new types as they emerge from real query needs.

Vertex types (nodes)

Type Purpose Examples
Document Vault entries (source of most edges) any 06-reference/*.md file
Person Person who wrote source material Ben Thompson, João Moura, Jaya Gupta, Joe Reis, Ben Wilson
Publication Newsletter or outlet Stratechery, Every, Semi-Structured, Practical Data Modeling
Topic Concept/area the doc is about "harness thesis", "data quality", "state ownership", "BEAM methodology"
Framework Named framework we reference or build MAC, BEAM, Scope×Basis, Seven Powers, Entangled Software
Project RDCO work stream MG Progress, Sanity Check, autoinv, MAC content series, Graph DB eval
Sponsor Entity sponsoring content Anthropic, OpenAI, Box, Estuary
Organization Company/institution phData, MG, Mammoth Growth, Anthropic, Foundation Capital
Skill RDCO skill file process-newsletter, audit-model, postgrid, morning-prep
Cluster Thematic grouping of docs "harness-thesis dissent", "data quality sources", "moat debate"
Decision Significant decision doc phData vs MG, Mode A vs Mode B, SQL guardrail adoption
Source Original external document book PDFs, podcast transcripts, X articles

Edge types (typed relations)

Semantic relations between documents

Edge Direction Semantics
validates A → B A provides evidence supporting B's claim
contradicts A → B A disputes B's central thesis
disputes-claim-in A → B A disputes a specific claim in B (not whole doc)
refines A → B A sharpens B's framework without contradicting
extends A → B A builds on B, takes further
synthesizes A → (B, C, D...) A combines multiple sources into a view
echoes A → B A independently reaches B's conclusion
applies-framework-from A → B A uses the framework defined in B
cites A → B A explicitly references B
mentions-in-passing A → B Weaker than cites

Personship and identity

Edge Direction Semantics
authored-by Document → Person
published-in Document → Publication
sponsored-by Document → Sponsor Disclosed sponsorship
affiliated-with Person → Organization
role-of Person → Role (CEO, founder, analyst)

Topical

Edge Direction Semantics
about-topic Document → Topic Primary subject
mentions-entity Document → Entity Named entity reference
part-of-series Document → Series e.g., ADWD ch1 → "Agile Data Warehouse Design" book
part-of-cluster Document → Cluster e.g., Moura → "harness-thesis dissent"

Strategic / operational

Edge Direction Semantics
supports-position Document → RDCO position Evidence for a strategic stance
informs-decision Document → Decision Input that shaped a decision outcome
test-case-for Case study → Framework Concrete example of framework application
example-of Document → Pattern/Concept Illustration of a pattern

Provenance

Edge Direction Semantics
generated-by-skill Document → Skill Which skill produced this vault entry
derived-from-source Document → Source Which external doc this note is assessing
produced-by-project Document → Project Which RDCO project created this artifact

Edge metadata (properties on edges)

Every edge should carry:

Query catalog (what this graph must answer)

These are the queries that justify the graph over QMD-only retrieval:

  1. Cascading contradictions — multi-hop contradiction chains
  2. Dissent cluster aggregation — which authors appear in ≥N dissent docs
  3. Source authority ranking — validates/contradicts edges received per author
  4. Framework adoption trajectory — which docs apply which framework, over time
  5. Positioning evidence compilation — docs that support-position a given stance
  6. Derivation path — shortest path between two conceptually linked docs
  7. Gap detection in a cluster — subtopics with insufficient sources
  8. Person citation network — inbound + outbound citation traversal
  9. Skill improvement provenance — findings → improvements → skill changes (two-hop)
  10. Decision-evidence audit — all docs that informed a given decision

Prototype scope (first 3 queries to prove out)

  1. Positioning evidence compilation (query #5) — "Find all vault entries that support the state-ownership architecture"
  2. Dissent cluster aggregation (query #2) — "Which authors appear in ≥3 dissent docs?"
  3. Decision-evidence audit (query #10) — "What evidence informed the phData vs MG decision?"

These three cover the most immediate operational use cases.

Versioning

This dictionary is iterative. When a query surfaces a need for a new edge type, add it here with the date. When an edge type turns out to be redundant or never-used after 2+ months, deprecate it here.

Date Change Reason
2026-04-14 Initial dictionary Founder asked for running data dictionary during graph DB eval scoping
2026-04-14 Renamed AuthorPerson Author is a role relative to a doc. Person is the entity. Covers non-authors too (colleagues, clients, friend-of-founder contacts like the brother-in-law or vertical-farm friend). The authored-by edge still exists and connects Document → Person.