OpenStatSpec Python

The reference Python implementation of the OpenStatSpec specification.

This package implements the specification; it does not define or extend it. The normative model lives in the OpenStatSpec/specification repository.

Boundaries

For each supported import, one source dataset becomes one dedicated wide SQL table. Cases are rows and source variables are physical SQL columns. The singular UUID-keyed tables from the specification (dataset, variable, operation, fidelity_event, and related metadata tables) are the public catalog contract and the only catalog representation created or read by the adapter. Databases created with the former *_catalog compatibility schema must be remediated to the normative schema before use. The adapter does not reshape data, create EAV or long-form tables, or harmonize studies or waves.

Unsupported source features, SQL targets, or export paths fail explicitly. There is no silent truncation, type conversion, metadata loss, or partial import.

Package layout

• openstatspec.core: pure standard concepts, validation, versions, capabilities, and loss reports.
• openstatspec.sql: database connection and wide-table/catalog operations.
• openstatspec.spss: SAV/ZSAV adapter boundary.
• openstatspec.transform: canonical plans, frontend-neutral schema concepts, and plan validation.
• openstatspec.frontends.spss: the SPSS-like syntax frontend and convenience execution adapter.

Intended workflow

from openstatspec import export_sav, import_sav

import_sav("responses.sav", database_url="postgresql+psycopg://user:password@server/database", dataset_id="responses-2026")
export_sav(database_url="postgresql+psycopg://user:password@server/database", dataset_id="responses-2026", destination="responses-roundtrip.sav")

openstatspec import responses.sav --database-url postgresql+psycopg://... --dataset-id responses-2026
openstatspec export --database-url postgresql+psycopg://... --dataset-id responses-2026 --output responses-roundtrip.sav

Optional database-first SQL workflow

Imported datasets remain immutable source records. The optional SQL transformation profile can register versioned, parameterized SQLite SELECT queries, materialize results, record lineage and weights, and expose derived datasets through a public catalog API. It uses a separate profile catalog and never presents SQL output as an imported source dataset. Workflow operations support SQLite only in this milestone and fail closed on PostgreSQL/MySQL/MariaDB; core import/export database support is unchanged. The core SQLite import/export profile accepts SQLite >=3.24.0,<4.0.0; the optional transformation workflow deliberately has the narrower >=3.35.0,<4.0.0 runtime preflight. These independent tiers do not change the server-profile matrix. Microsoft SQL Server is not supported; its future dialect is scoped only in the specification's MSSQL roadmap.

See the SQL transformation workflow for Python and CLI examples, migration behavior, hashing, atomicity, and the exact implemented capability boundary.

SPSS-like transformation frontend

The bounded SPSS-like frontend lowers RECODE, sequential COMPUTE and IF, STRING declarations, and DELETE VARIABLES, VARIABLE LABELS, VALUE LABELS, numeric FORMATS, VARIABLE LEVEL, and EXECUTE into a language-neutral typed canonical plan. Conditions support parentheses, variable and numeric literal operands, comparisons, AND, and OR. String comparison and v0.2 string assignment currently fail closed. The in-place path applies the plan to the same logical dataset, physical wide table, and normative/compatibility metadata without a derived dataset, copied table, snapshot, or hidden history layer. Numeric targets may be created atomically on SQLite and PostgreSQL. MySQL, MariaDB, and Dolt fail closed on target_mode=create and require a separately provisioned physical and catalog target before assignment. Dolt requires the caller's exact clean branch/HEAD, leaves success as an inspectable working-set diff, and never calls DOLT_COMMIT.

See the dataset transformations manual for schema installation, Python and CLI surfaces, database invariants, audit provenance, package layout, and extension guidance. Stata and SAS are unimplemented placeholders.

Current support status

The adapter requires openstatspec-pyspssio==0.5.1.post2 as its sole SPSS engine. Its import module remains pyspssio; the exact source commit is recorded in operation metadata. There is no fallback reader or writer. It supports unencrypted SAV and ZSAV import and SAV/ZSAV export for the semantics exposed by that engine. SQLite is the local reference path. PostgreSQL, MySQL, MariaDB, and Dolt are each covered by separate service-backed CI conformance checks. Dolt support is an independent core profile for the canonical stable range >=2.2.2,<2.3.0; earlier patches, other families, noncanonical versions, and unknown MySQL-wire products fail closed.

The supported family claims are broader than the deliberately exact CI evidence points: PostgreSQL 17.x/18.x is exercised at 17.10/18.4, MySQL 8.4.x/9.7.x at 8.4.11/9.7.2, and MariaDB 11.4.x/11.8.x/12.3.x at 11.4.12/11.8.8/12.3.2. Each service job checks the normalized live server version against its exact matrix entry before that run can count as evidence. Dolt claims the conservative 2.2.x range >=2.2.2,<2.3.0; its full service suite is exercised independently at exact versions 2.2.2 and 2.2.3 using immutable container-image digests.

Microsoft SQL Server (MSSQL) remains roadmap-only and is not a supported runtime profile; see the specification's MSSQL roadmap.

Use these explicit SQLAlchemy URLs:

• SQLite: sqlite:///dataset.sqlite
• PostgreSQL: postgresql+psycopg://user:password@host/database
• MySQL/MariaDB: mysql+pymysql://user:password@host/database
• Dolt >=2.2.2,<2.3.0: mysql+pymysql://user:password@host/database (detected by server identity)

The Dolt core profile supports strict wide-table import, validation, and export; the separate Transformation Workflow is unsupported.

Run openstatspec capabilities before an integration to inspect the machine-readable feature matrix. Export is deliberately strict: if known dictionary semantics cannot be reproduced, it stops until you pass the exact diagnostic code with --allow-loss. This avoids silent loss while making an intentional lossy export auditable.

The matrix is also available to Python callers as openstatspec.capability_matrix(). It distinguishes supported semantics from unobservable and fail-closed paths; see the SAV profile for the exact the openstatspec-pyspssio boundary.

See the SAV profile for feature boundaries and release readiness for the pre-tag checklist. Read third-party notices before distributing a bundled application: the required engine includes IBM redistributables under separate terms.