Modelwright

modelwright is an early-stage project for turning spreadsheet workbooks into transparent, version-controlled, standalone Python models.

The intended direction is a generic workflow that can inspect workbook structure, extract formulas and dependencies, generate maintainable Python source, and validate the generated model against the original workbook outputs.

This repository is currently an early implementation skeleton. It defines minimal Python package and test scaffolding plus initial validation, extraction, graph, generation, oracle records, and thin JSON command-line wrappers, but does not yet provide a release stability guarantee, catalog schema, or full workbook conversion.

Current Focus

• Build the first package-backed validation/report, workbook extraction, generation, and CLI cores.
• Keep extraction, code generation, validation, diagnostics, and reporting responsibilities separate.
• Avoid committing private notes, source workbooks, generated clones, or large artifacts while the project shape is still being established.

Python API Boundary

The durable API is organized by module responsibility:

• modelwright.extraction: workbook extraction records and extract_workbook.
• modelwright.graph: dependency graph records and build_dependency_graph.
• modelwright.formulas: formula expression records, translation helpers, and reference-index helpers.
• modelwright.generation: generated-module records and generate_python_module.
• modelwright.validation: validation scenarios, scalar comparisons, and report records.
• modelwright.oracles, modelwright.formulas_oracle, and modelwright.oracle_validation: oracle request/result records, optional formulas oracle execution, and oracle-backed report assembly.

The package root modelwright exposes a curated convenience facade for those records and functions. Module-level imports remain preferred for implementation work because this project is still pre-release.

Command-Line Interface

Bootstrap the repo-local virtual environment before using the console script:

scripts/bootstrap_dev_env.sh

The current CLI prints JSON to stdout and stays close to the Python APIs:

modelwright workbook extract path/to/workbook.xlsx > tmp/extraction.json
modelwright workbook graph path/to/workbook.xlsx > tmp/dependency-graph.json
modelwright conversion plan path/to/workbook.xlsx > tmp/conversion-plan.json
modelwright model generate --contract tmp/contract.json --expressions tmp/expressions.json --constants tmp/constants.json --out tmp/generated_model.py > tmp/generation-result.json
modelwright validation report --scenario tests/fixtures/synthetic_model/baseline_scenario.json --generated-values tmp/generated-values.json --oracle-values tmp/oracle-values.json > tmp/validation-report.json

These commands do not provide a one-step workbook converter. conversion plan reports extraction, graphing, formula-translation, and residual-blocker status; model generate expects explicit generated-module and formula-expression JSON inputs; and validation report compares already-observed generated/oracle values. See planning/cli-json-workflows.md for JSON examples and workflow boundaries.

Local Development

Bootstrap a repo-local virtual environment:

scripts/bootstrap_dev_env.sh

This installs Modelwright with the dev extra:

.venv/bin/python -m pip install -e '.[dev]'

Run lint checks:

.venv/bin/python -m ruff check .

Run tests:

.venv/bin/python -m pytest

Build docs locally:

.venv/bin/sphinx-build -b html docs _build/html -W
.venv/bin/python scripts/verify_docs_theme.py _build/html

Restore the public external FABLE benchmark workbooks into ignored local paths:

scripts/bootstrap_dev_env.sh --benchmarks

modelwright is pre-release. The current alpha line is 0.1.0a9; alpha releases must not be described as full-workbook conversion guarantees.

Check release artifacts locally:

scripts/check_release_artifacts.sh

Release checks write build outputs under ignored tmp/release-checks/.

See docs/guides/release-deployment.rst for the release and deployment runbook.

Repository Conventions

• AGENTS.md is the working contract for AI coding agents.
• CONTRIBUTING.md is the contributor onboarding and development workflow guide.
• ROADMAP.md is the current plan and next-step tracker.
• CHANGE_LOG.md is the append-only project narrative.
• planning/ contains focused design notes and research records that are too detailed for the roadmap.
• benchmarks/ contains tracked metadata for official external benchmarks; large workbook binaries remain untracked and are restored locally under tmp/.
• src/modelwright/ contains the importable Python package.
• tests/ contains package-backed tests and tracked synthetic fixture helpers.
• tmp/ is ignored local working space for private notes, source workbooks, experiments, and generated scratch outputs.