Configuration

The config is a JSON file. A config file is one run. The root object holds input, output, and the run knobs directly. Load it with squashage-dag build --config squashage.config.json.

Schema source of truth: src/schemas/squashage-config.schema.json.

Copy squashage.config.example.json as a starting point.

Root shape

{
  input:      { basePath: string; format: 'json' | 'jsonl' };  // required
  output:     OutputConfig;                          // required — see Output config below
  concurrency: number;                               // optional, default 1
  graphs:     { [key: string]: string };             // optional — named-graph IRI overrides
  ontology:   OntologyConfig;                        // optional — pass-through; plugins build the ontology
  quarantine: QuarantineConfig;                      // optional — quarantine output overrides
  subjectIri: SubjectIriConfig;                      // optional — subject IRI minting policy
}

The DAG topology is authored as .dag.jsonld documents and loaded at construction. See DAG for the full topology.

input

Key	Type	Required	What it does
basePath	string	yes	Base directory for the run's input.
format	"json" | "jsonl"	yes	One JSON object per file (json), or one per line (jsonl).

Root knobs

Key	Type	Required	Default	Notes
input	object	yes		Input source — see above.
output	OutputConfig	yes		See Output config.
concurrency	integer ≥ 1	no	1	Maximum concurrent record-DAG clones in the native scatter.
graphs	object	no		Named-graph IRI overrides; keys map to services.graphs.
ontology	object	no		Pass-through config; plugins construct JsonTologyOntology in register(). services.ontology is always null at the framework level.
quarantine	object	no		Quarantine output directory overrides.
subjectIri	object	no		Subject IRI minting policy.

---

Plugin

Plugin-specific configuration (classifiers, ontology engine, enrichment predicates) is owned by the plugin, not the run config. Pass --plugin <namespace> on the CLI to load the plugin. The plugin reads its own config from its source directory. See Plugins for the authoring contract.

---

Output config

See Output for full details. The minimal shape:

{
  type:   'file';
  path:   string;
  format: 'turtle' | 'trig' | 'ntriples' | 'nquads' | 'jsonld';
  mode?:  'dataset' | 'stream';   // default 'dataset'; 'stream' streams quads to disk
}

type is the discriminant. rdfjs-finalize writes these files on every run:

File	What it carries
<output.path>	The success graph.
<output.path-stem>.prov.<ext>	PROV-O activity graph emitted by the dispatcher's lifecycle hooks.
<outDir>/<run>/quarantine/<bucket>/<id>.json	Failed records, grouped by bucket.

---

Example

{
  "input":  { "basePath": "./input", "format": "json" },
  "output": { "type": "file", "path": "./graphs/aonprd.trig", "format": "trig" },
  "concurrency": 4,
  "graphs": { "default": "https://example.org/graph/aonprd/default" }
}

For bounded-memory streaming, set output.mode to stream and write a quad format:

{
  "input":  { "basePath": "./input", "format": "json" },
  "output": { "type": "file", "path": "./graphs/aonprd.nq", "format": "nq", "mode": "stream" }
}