Meridian Storage Semantics

meridian-storage-semantics is the independently released Meridian V1 logical schema, resource, data, and query-facing metadata package. It provides portable contracts for Records, Relations, Objects, cache entries, dynamic Schema publication, compatibility classification, multilingual ICU metadata, and Core mapping-first Catalog surfaces.

The distribution contains one Python package, meridian_storage.semantics, and depends on the released meridian-storage-core==1.0.0 contract. The shared meridian_storage import root is packaging infrastructure; this repository does not contain Core source or any other Meridian package.

Install

python -m pip install meridian-storage-semantics==1.0.0

Python 3.12 or newer is required.

Define and publish a Schema

from meridian_storage.semantics import InMemoryMetadataRepository, SchemaAPI

schemas = SchemaAPI(InMemoryMetadataRepository())
published = schemas.publish(
    namespace="investigation",
    name="case",
    version="1.0.0",
    definition={
        "semanticKind": "relational",
        "fields": [
            {"name": "case_id", "logicalType": "uuid"},
            {"name": "title", "logicalType": "string"},
        ],
        "identity": ["case_id"],
    },
)

assert published.publication.document.format_version == "meridian.schema.v1"

Schema versions are immutable. Publishing an identical version and fingerprint is idempotent; a different document under the same version is rejected. New versions are ordered by SemVer and classified before activation.

Build mapping-first Expressions

from meridian_storage.semantics import StructuredCatalogProvider

provider = StructuredCatalogProvider()
expression = provider.create_surface().query(
    resource="investigation.cases",
    where={"status": {"eq": "open"}},
    select=("case_id", "title"),
    limit=50,
)
operation = provider.normalize(expression)

Consumers create serializable Core Expression values. Planning normalizes them to portable Operation values. Engine selection, credentials, provisioning, physical DDL, ACLs, migration execution, and lifecycle remain Adapter/Platform responsibilities and are not accepted by these APIs.

Catalog registry

The V1 registry is exactly:

• structured — owned here; relational, document, key-value, search, geospatial, time-series, and relation profiles.
• object — owned by meridian-storage-object-common; object, artifact, and media profiles. This package supplies their logical metadata values.
• cache — owned here; explicit, non-authoritative cache semantics.
• evidence — owned by meridian-storage-evidence; telemetry, audit, lineage, and provenance profiles.
• streaming — accepted-pending and owned by the downstream streaming package.

Query, projection, ontology, search, geospatial, time-series, relation, artifact, media, telemetry, audit, lineage, and provenance are not Catalogs. NativeQuery is post-V1. No Kafka or streaming-provider behavior is embedded.

Contracts and verification

Language-neutral JSON contracts live in contracts, including the canonical meridian.schema.v1 schema, wire documents, public API ledger, exact Catalog registry, and valid/invalid conformance fixtures. Compatibility with Core and the design revisions is pinned in compatibility.json.

Run all local gates after installing test dependencies:

python -m pip install '.[test]'
ruff format --check src tests scripts
ruff check src tests scripts
mypy src
python scripts/verify_contracts.py
pytest --cov=meridian_storage.semantics --cov-report=term-missing

Release builds are created twice with a fixed SOURCE_DATE_EPOCH, compared byte-for-byte, inspected for repository/package boundaries and license metadata, and accompanied by SPDX SBOM and GitHub build-provenance attestations.

Further detail is in docs/contracts.md and docs/architecture.md.

License

Copyright 2026 Meridian contributors. Licensed under Apache License 2.0; see LICENSE and NOTICE.