writing / cartography-and-graphs 8 public repos
← writing ↑ part of Data & semantic layer
retrievalmapping

System cartography — schema-first maps and typed graphs

The class of problem

Before the build, the map.

Before you can build in an unfamiliar domain — or trust a large corpus — you have to know its shape, and most teams jump straight to a tool or a vendor pick without ever mapping the surface. The class is decision-under-unfamiliarity: scoping a landscape, or making a messy corpus answerable, before committing to a build.

What it is

Three related artifacts: a cartography method that catalogs an unfamiliar domain before any build, and two typed graphs that make a corpus queryable. The class is schema-first structure: define the node and edge types before touching the data, parse a real corpus into exactly those types, then query across the result. Each artifact applies that method to a different kind of corpus, and at a different stage of maturity.

galaxy-map — an unfamiliar technology domainmechanism-graph — a career-evidence corpusdev-tel-graph — a telemetry vault
93
mechanisms in the typed graph
316
evidence nodes
67
COMPOUNDS_WITH edges
24 ms
for the cross-domain probe
What's honest about the scope. galaxy-map is in use, with multiple universe maps produced — the count is tracked elsewhere rather than asserted here. mechanism-graph has the full corpus ingested, its schema versioned, and the cross-domain probe gate met at 24 ms. dev-tel-graph is a spike — enough to exercise the method, not a finished artifact. A public retrieval demo is the candidate next step; it is not yet built.

Where it generalizes

All three are instances of the BI / knowledge layer — cataloging a domain or corpus so it is queryable: the discovery step before a data or vendor decision, the scoping pass over an unfamiliar stack, the retrieval substrate under a knowledge product.

The front of the pipeline

The map is the deliverable — not a recommendation.

The /mapit method produces a universe map: a wide-net triage catalog of every component, vendor, and technology in an area, written before any build decision is made. The output is a dated document per area.

01

Wide net, held wide

The catalog stays wide — nothing is filtered out prematurely — while still flagging what fits. The map exists to establish the full surface of options first.

02

Scored against real constraints

Each entry is scored against the constraints that actually apply to the work: the hardware on hand, the standing license posture, and the domains already in play.

03

The decision is a later step

Deciding which component to adopt is a separate, later step. The map makes that decision possible; it doesn't make it.

04

Structure in the document

The lightest-weight of the three artifacts and the only one with no graph behind it: the structure lives in the document's schema — component, vendor, technology, score — rather than in a database.

This is the front of the pipeline — a domain is mapped before any candidate is promoted into a build. Several maps exist; the count is tracked elsewhere rather than asserted here.

The corpus, typed

The schema is the design.

A typed knowledge graph over a career-evidence corpus, built on KuzuDB — and a sibling spike that tests whether the approach transfers to a different corpus and a different engine.

01

Typed nodes, versioned types

Nodes are typed — mechanisms, domains, career phases, through-lines, meta-patterns, and evidence — and the schema itself is versioned: the type definitions evolve under version control rather than drifting implicitly.

02

Disposition, apart from method

The central modeling choice separates the invariant from the method. A disposition — the underlying invariant a mechanism expresses — is modeled as a distinct thing from the method that expresses it, so the same disposition can be traced across different domains and across time.

03

Disposition-preserving ingestion

The parser reads the real corpus and lands each item as the node type the schema declares — without flattening the disposition/method distinction the schema is built to hold.

04

De-identified corpus

No client data is present in the graph; the mechanisms and evidence are abstracted from their original engagements.

What's in the graph

Engine
KuzuDB
Mechanisms
93 74 deep + 19 cluster
Context nodes
10 domains · 4 career phases · 6 through-lines · 5 meta-patterns
Evidence
316 nodes
Edges
67 COMPOUNDS_WITH cross-domain relationships as first-class edges

The query that justifies the structure

A cross-domain probe: find the cluster mechanisms that span three or more domains across both past and present. It returns 10 such mechanisms in 24 ms. Over a flat document set the same probe would require manual cross-referencing; here it is a single traversal, because the cross-domain relationships are first-class edges.

The sibling — dev-tel-graph

Sibling

Same method, different corpus, different engine

A graph layer over a personal telemetry vault, built on a different graph engine than mechanism-graph's KuzuDB. It exists partly as a test: does the typed-graph approach transfer across both the corpus and the underlying store? At spike stage — enough to exercise the method, not a finished artifact.

spike stage

What they share

Define the types. Parse the corpus. Query the result.

The schema-first method is the through-line. In each case the result is queryable, not a diagram.

ArtifactCorpusWhere the structure livesStage
galaxy-mapan unfamiliar technology domaina document schema — component, vendor, technology, scorein use; multiple universe maps produced
mechanism-grapha career-evidence corpusa typed KuzuDB storefull corpus ingested; schema versioned; probe gate met at 24 ms
dev-tel-grapha personal telemetry vaulta typed store on a different enginespike stage

The three sit at different points on one axis: galaxy-map maps an unfamiliar domain before a build; mechanism-graph and dev-tel-graph make a known corpus queryable after the fact.

What comes next

next · candidate

A public retrieval demo

mechanism-graph plus a vector-retrieval layer is the candidate next step — the typed graph as the retrieval substrate under a public demo. It is not yet built.

candidate · not yet built