@studnicky/squashage

Graph reconstitution pipeline; classifies JSON records and squashes them into deterministic RDF.

Upstream: Ripperoni produces the JSON records Squashage consumes. Pair them to go from raw web pages to structured RDF.

Documentation · Architecture · Classifier engines · Demo · Releases

---

Requirements

• Node 24+
• TypeScript 5.7+

Install

git clone https://github.com/Studnicky/Squashage.git
cd Squashage
npm install
npm run build

Quickstart

# Run the full Pathfinder/AONPRD demo (pipeline + cytoscape render)
npm run viz:demo

# Build from a config
squashage build \
  --target aonprd \
  --config squashage.config.json \
  --in ./output/aonprd

# Override output path/format for a one-off run
squashage build --target aonprd --out ./graphs/aonprd.jsonld

# Render any squashage JSON-LD output as an offline HTML graph
squashage viz --in ./graphs/aonprd.jsonld --out aonprd.html --title "My Graph"

Copy squashage.config.example.json to squashage.config.json and edit. The unprefixed file is gitignored.

Scripts

npm run build         # compile TypeScript
npm run typecheck     # tsc --noEmit
npm run lint          # eslint src/
npm run check         # typecheck + lint + unit tests
npm run docs:build    # build VitePress docs
npm run viz:demo      # rebuild the Pathfinder/AONPRD demo

Goals

• Consume structured JSON input records from one file, a directory tree, or JSONL.
• Classify each input record into an ontology type with deterministic evidence.
• Normalize source-specific records into stable graph entities.
• Project records into RDF/JS quads using a thin wrapper over @rdfjs/*, n3, jsonld.
• Serialize the canonical dataset to a single RDF file per build run (v0.x: Turtle, TriG, N-Triples, N-Quads, JSON-LD; v1.x adds RDF/XML and N3).
• Run a deterministic, declaratively configured classifier cascade.

Non-Goals

• Squashage does not scrape web pages.
• Squashage does not run probabilistic models in the build path.
• Squashage does not load graph stores. Hand the file to your loader of choice.
• Squashage does not fan out across multiple outputs. One build, one file.

Config Sketch

{
  "targets": {
    "aonprd": {
      "input": "./output/aonprd",
      "pipeline": [
        "json:read",
        "classify:source",
        "classify:structural",
        "classify:rules",
        "classify:ontology",
        "classify:conflict",
        "aonprd:squash",
        "rdfjs:finalize"
      ],
      "output": {
        "kind": "file",
        "path": "./graphs/aonprd.jsonld"
      }
    }
  }
}

See the full config reference for all options.

Plugins

Custom squash tasks register themselves with TaskRegistry:

import { TaskRegistry } from 'squashage/registry/TaskRegistry';

TaskRegistry.register('aonprd:squash', async (next, state) => {
  if (state.classification?.type !== 'feat') { await next(); return; }

  const ctx      = state.context!;
  const prefixes = ctx.prefixes;
  const subject  = ctx.factory.namedNode(
    `${prefixes.instances.base}${String(state.input['url'] ?? 'unknown')}`
  );
  ctx.dataset.add(ctx.factory.quad(
    subject,
    ctx.factory.namedNode('http://www.w3.org/1999/02/22-rdf-syntax-ns#type'),
    ctx.factory.namedNode(`${prefixes.vocabulary.base}Feat`),
    ctx.factory.namedNode(`${prefixes.graphs.base}feat`),
  ));
  await next();
});

Demo

Open docs/public/examples/aonprd/aonprd.html in any browser to see the package's JSON-LD output rendered as an interactive graph. Nodes are coloured by RDF class, edges show object-property links, and clicking a node reveals its properties in the sidebar. The file runs entirely offline , no network access, no Node.js, no node_modules required at display time.

To rebuild the demo from the fixture data:

npm run viz:demo

License

UNLICENSED