tsprov

An idiomatic TypeScript implementation of the W3C PROV Data Model — a port of the Python prov library. Author provenance documents with a fully-typed fluent API, and round-trip them through PROV-JSON and PROV-N.

📖 New to PROV or want the full tour? See the tsprov guide — provenance concepts, a worked example, the data model in depth, and design notes.

• Value-equality that works. doc.equals(other) is content-based — validated against the full 398-file Python PROV-JSON conformance corpus (every fixture round-trips: deserialize → serialize → deserialize is .equals()-stable).
• Fluent, typed authoring. doc.entity("ex:report"), e.wasGeneratedBy(a).wasAttributedTo(agent), with the camelCase PROV vocabulary as the primary API.
• Dependency-light core. Only luxon (datetime fidelity). Browser- safe and tree-shakeable.
• Dual ESM + CJS with .d.ts declarations — works in Node, Bun, and bundlers.

Install

tsprov is published to the public npm registry with provenance attestations — every release is built and signed on GitHub Actions from this repository.

npm install @inflexa-ai/tsprov     # or: bun add @inflexa-ai/tsprov

Quick start

import { ProvDocument } from "@inflexa-ai/tsprov";

const doc = new ProvDocument();
doc.addNamespace("ex", "http://example.org/");

// Elements
const article = doc.entity("ex:article", { "prov:type": "ex:Article" });
const compile = doc.activity("ex:compile", "2024-01-01T09:00:00+00:00", "2024-01-01T09:05:00+00:00");
const author = doc.agent("ex:alice");

// Relations — container form
doc.wasGeneratedBy(article, compile, "2024-01-01T09:05:00+00:00");
doc.wasAttributedTo(article, author);

// …or the fluent record form (chainable; returns the record)
article.wasDerivedFrom(doc.entity("ex:draft")).wasAttributedTo(author);

console.log(doc.serialize("provn"));
// document
//   prefix ex <http://example.org/>
//   …
// endDocument

Serialization

const json = doc.serialize("json");            // PROV-JSON
const provn = doc.serialize("provn");          // PROV-N (text)

const parsed = ProvDocument.deserialize(json, "json");
doc.equals(parsed); // true

// read() auto-detects the format
import { read } from "@inflexa-ai/tsprov";
const back = read(json);

PROV-JSON supports both serialize and deserialize. PROV-N is serialize-only (matching the reference library — there is no standard PROV-N parser).

Bundles

const doc = new ProvDocument();
doc.addNamespace("ex", "http://example.org/");

const bundle = doc.bundle("ex:bundle1"); // a named sub-bundle (inherits the document's namespaces)
bundle.entity("ex:nested");

doc.hasBundles();   // true
doc.flattened();    // a new document with all bundle records lifted to the top level

Typed literals

JavaScript has a single number type, so to preserve the XSD datatype distinction, wrap typed values in a Literal:

import { Literal, XSD_INT } from "@inflexa-ai/tsprov";

doc.entity("ex:dataset", { "ex:rows": new Literal(10_000, XSD_INT) });

A bare number defaults to xsd:double on encode; a bare string is an xsd:string.

Graph & lineage

Provenance is a graph, so tsprov ships a graph view and a lineage walker under the optional @inflexa-ai/tsprov/graph subpath — zero extra dependencies (the core stays luxon-only). The headline: a lineage answer is itself a PROV document you can serialize and feed to any PROV tool.

import { ProvDocument } from "@inflexa-ai/tsprov";
import { ProvGraph, resolve, lineage, toProvDocument } from "@inflexa-ai/tsprov/graph";

const doc = new ProvDocument();
doc.addNamespace("ex", "http://example.org/");
const article = doc.entity("ex:article", { "prov:type": "ex:Article" });
const compile = doc.activity("ex:compile", "2024-01-01T09:00:00+00:00", "2024-01-01T09:05:00+00:00");
article.wasGeneratedBy(compile).wasDerivedFrom(doc.entity("ex:draft"));

const graph = ProvGraph.of(doc);                         // flattened().unified() multi-digraph
const found = resolve(graph, { localpart: "article" });  // git-style resolve → all matches
if (found.kind !== "matched") throw new Error("no such record");

const ancestry = lineage(graph, found.records, { direction: "backward" }); // "where did this come from?"
const { document } = toProvDocument(graph, ancestry);    // …the answer is itself a PROV document

console.log(document.serialize("provn"));
// document
//   prefix ex <http://example.org/>
//
//   entity(ex:article, [prov:type="ex:Article"])
//   activity(ex:compile, 2024-01-01T09:00:00+00:00, 2024-01-01T09:05:00+00:00)
//   entity(ex:draft)
//   wasGeneratedBy(ex:article, ex:compile, -)
//   wasDerivedFrom(ex:article, ex:draft, -, -, -)
// endDocument

The default backward/"dataflow" walk answers ancestry; direction: "forward" and "both" answer descendants, depth bounds the hops (every cutoff surfaces as explicit frontier data, never a silent truncation). toFlatGraph(result) gives a JSON-safe { nodes, edges } projection, and lineagePaths(graph, result, target) enumerates the connecting paths. See §8 of the guide for the full tour.

What's included

• The full PROV-DM in-memory model: Identifier / QualifiedName / Namespace / Literal, all 3 elements + 15 relation classes, NamespaceManager, ProvBundle, ProvDocument.
• The complete fluent authoring API (camelCase PROV vocabulary primary; descriptive aliases).
• PROV-JSON (serialize + deserialize) and PROV-N (serialize), with read() auto-detection.
• Content-based equals(), plus flattened() and sub-bundles.
• Graph & lineage under the optional @inflexa-ai/tsprov/graph subpath — a multi-digraph view, composable record resolve(), and a directional, bounded lineage() walk, with no extra dependencies.

PROV-XML, PROV-RDF, DOT (graph-visualisation) rendering, and the CLI are out of scope for now.

Develop

bun install
bun run bootstrap   # fetch the 398-file PROV-JSON conformance corpus into reference/
bun test            # run the test suite (incl. the corpus oracle)
bun run build       # emit dist/ (ESM + CJS + .d.ts)

The corpus is the Python reference implementation's, and is deliberately not vendored — see CONTRIBUTING.md for why. Without bootstrap, the two suites that read it fail with ENOENT.

Contributing

Contributions are welcome, and PROV is a standard we implement rather than own — so if you work on PROV elsewhere in the ecosystem, we would rather have you inside this project than outside it. Start with CONTRIBUTING.md; GOVERNANCE.md explains how decisions get made and what is openly missing (PROV-XML, PROV-RDF, a PROV-N parser, PROV-CONSTRAINTS validation). Vulnerabilities go through SECURITY.md, never a public issue.

License

Apache-2.0. tsprov is a port of the Python prov library, used under the MIT License — see NOTICE.