account_treeProcesses & flows

pgRDF's operating model is a set of composable verbs you chain with arrows. Each verb is one or two SQL function calls. This section is the verb reference, four worked patterns, and a method for building your own chain.

The verbs

Verb	What it does	Functions
Import	Load RDF into a graph.	load_turtle, load_turtle_staged_run, parse_turtle, parse_trig, parse_nquads
Seal	Freeze a graph against writes and record its fingerprint.	lock_graph, graph_digest, graph_manifest
Carve	Copy a right-sized slice of a graph into a new graph.	carve_graph
Reason	Work out what follows from the data (OWL 2 RL or RDFS) and store it beside the asserted triples.	materialize
Validate	Check a graph against SHACL shapes and get a report.	validate
Query	Read and update with SPARQL 1.1.	sparql, construct, describe
Unload	Package a graph as verifiable files, then remove it.	export_graph, graph_manifest, drop_graph

All seven are available. When a chain uses every verb, they run in this order:

Import ─▶ Seal ─▶ Carve ─▶ Reason ─▶ Validate ─▶ Query ─▶ Unload

Most chains use three or four of them. Load → Query on its own is a complete process.

How each verb scales

Kind	Verbs	Behaviour
Parallel	Import	The staged loader spreads the work across background workers, so more cores load faster. It ingested the complete 8.2-billion-triple Wikidata dump into one instance (Scale & benchmarks).
One graph, one session	Reason, Validate	Runs in a single database session over one graph. That graph has to fit your hardware.
Per query	Query	Planned and run like any other SQL statement, using the indexes.
Graph operations	Seal, Carve, Unload	Single function calls that lock, fingerprint, copy, export or remove a graph.

This is the constraint that shapes every chain: ingest is parallel, reasoning is not. When the source is larger than one session can reason over, load all of it, carve the slice you need into its own graph, and reason over the slice.

The patterns

Four worked chains cover almost every process. Pick one with Choosing a process:

• search Load → Query — the simplest chain: load RDF and query it.
• psychology Load → Reason → Query — materialize what follows from the data, then query it.
• verified Load → Validate → Query — a SHACL conformance gate.
• hub Ingest → Carve → Reason — carve a slice of a large source and reason over the slice.

Build your own

• Building a chain — the method: start from the output, work backwards to the verbs you need, decide whether to carve, and decide what to keep.
• Choosing a process — a decision table from goal and graph size to a pattern.

See also

• query_stats Scale & benchmarks — the ingest numbers these chains build on.
• storage Managing graphs — create, inventory, carve, lock and drop graphs.
• description Identity and export — digests, canonical export and manifests.
• layers_clear The four pillars — the engines the verbs map onto.