Meridian Synthetic Plugin

meridian-plugin-synthetic is the in-process, deterministic synthetic-data generation plugin for Meridian V1. It generates schema-valid Records, relation-aware fixtures, canonical datasets, and provider-neutral Events. Every write goes through released Meridian public contracts.

This repository contains exactly one Python distribution and is licensed under the Apache License, Version 2.0.

Install

The plugin requires the released Meridian V1 Core, Semantics, Query, Evidence, and Streaming distributions. After those artifacts are available:

python -m pip install meridian-plugin-synthetic

Use

from meridian_storage.plugins.synthetic import Generator, MemorySink, SyntheticSpec

spec = SyntheticSpec.load(spec_document)
sink = MemorySink(max_bytes=spec.bounds.max_memory_bytes)
run = Generator(spec).run(sink, workers=4)

assert run.validation.passed
records = sink.records(run.run_id)

SyntheticSpec pins the package coordinate, optional implementation digest, schema fingerprints, seed, locale set, time interval, counts, generators, relations, validation rules, output mode, and execution bounds. Random streams are derived from logical coordinates, so partitioning, worker count, and run ID do not change generated values.

Available sinks are:

• MemorySink for bounded tests and fixtures;
• MeridianRecordSink for public structured.put writes;
• MeridianStreamingSink for public streaming.publish_batch writes; and
• CanonicalDatasetSink for canonical NDJSON partitions followed by an explicit ArtifactPublisher hand-off.

Direct Record writes require a DirectWritePolicy. Production targets require an explicit approval flag from the owning process. Dataset publication never uses the Object Catalog directly; the caller supplies the Configuration & Artifact plugin bridge.

Source-backed generation accepts only an explicit, projected structured.query Expression and a digest-pinned deterministic transform. Run evidence includes source boundaries, query fingerprints, transform digests, and row counts, but never source values.

Contracts and support

Serialized schemas are in contracts, design alignment is in docs/architecture.md, and the compatibility pin is in compatibility.json. Python 3.12 through 3.14 is supported. Report vulnerabilities using SECURITY.md.