[MaxGhenis]

This repository had no license, so copyright law reserved all rights in its contents to their authors. With this PR, PolicyEngine releases the code under the MIT License and the repository's original text and figures under CC BY 4.0, in LICENSE and a README section. Closes #197.

On 2026-09-17 we opened license PRs like this one across the PolicyEngine, Thesis Institute, UBI Center and Cosilico repositories, with MIT for code and CC BY 4.0 for original text and figures. We kept AGPL-3.0 for the model packages that derive from OpenFisca.

Neither uv.lock nor any other dependency manifest on the default branch names a package that PolicyEngine licenses under AGPL-3.0. The paper notebook paper/imputing-from-scf-to-cps.ipynb imports two such packages, policyengine-us and policyengine-core.

Before PolicyEngine archives microimpute, I added:

• changelog.d/license-mit.added.md, so merging publishes a minor version with the license to PyPI.
• License metadata in pyproject.toml (license = "MIT" and license-files = ["LICENSE"]), with setuptools 77 or later as the build requirement.
• README sentences noting that the optional matching extra installs rpy2, which its authors license under GPL-2.0-or-later, and that the paper notebook imports AGPL-3.0 packages.
• An absolute URL for the README's link to LICENSE, so the link also works on PyPI's project page.
• 2025-2026 as the copyright years in LICENSE.

The README's license section covers paper/ as well, so I left the manuscript without a license line of its own.

Contributors who object to these terms for their contributions can email hello@policyengine.org.

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
microimpute-dashboard	Ready	Preview	Nov 8, 2026 5:43am UTC

[vahid-ahmadi]

Reviewed this with a fresh pass, checking the licence-compatibility question independently rather than taking the scan on trust — it is the one thing here with consequences for the five downstream repos.

The core claim holds. MIT is correct. I scanned the whole dependency closure, not just the direct list: everything resolves to BSD / MIT / Apache-2.0 apart from rpy2 (GPL-2.0-or-later), and MPL-2.0 for certifi, tqdm and pathspec, which is file-level copyleft and imposes nothing here. No policyengine-* package appears in the lock file at all, so the AGPL claim is confirmed. pytorch-tabular is MIT and quantile-forest Apache-2.0, so the mdn extra raises nothing.

The LICENSE text is verbatim OSI MIT, and the CC BY 4.0 split in the README is clean — disjoint subject matter, MIT scoped to "the Software", with a sensible third-party carve-out.

To do

Before merging

• Add the licence to packaging metadata. pyproject.toml has no license field, no license-files and no classifiers, so PyPI will keep publishing license: None — which microimpute has no licence, so all rights are reserved #197 says has persisted across all 53 releases. The LICENSE file satisfies JOSS, but not downstream consumers resolving a licence from installed metadata. Add license = "MIT", license-files = ["LICENSE"], and the License :: OSI Approved :: MIT License classifier.
• Sequence against Declare subpackages instead of shipping the source tree as package data #216. Declare subpackages instead of shipping the source tree as package data #216 edits the same [project] block — it adds classifiers, two authors and [project.urls]. GitHub shows no conflict today only because this PR touches just LICENSE and README; the moment it adds classifiers the two collide. Merge Declare subpackages instead of shipping the source tree as package data #216 first, then rebase this and append the licence classifier to the list it introduces.
• Get an explicit ack from @baogorek. microimpute has no licence, so all rights are reserved #197 names the substantial copyright holders; three of the four are PolicyEngine staff, but baogorek is external (9 commits on main). The PR offers opt-out — "contributors who object can open an issue" — and silence is treated as consent. An affirmative sign-off is cheap now and very expensive to retrofit once downstream repos have relied on the grant.

Worth doing in the same PR

• Note the GPL implication of the matching extra in the README. microimpute neither bundles nor links rpy2, the import is guarded, and the default install contains no GPL code — so MIT is right and this is not a violation. But a user who installs microimpute[matching] and redistributes that environment is bound by GPL-2+ for the combined work. One sentence pre-empts a JOSS reviewer asking.
• Add changelog.d/197.added.md and merge normally. The premise in the description does not hold: main has no branch protection (gh api .../branches/main/protection → 404), so the red changelog check is advisory and the merge button stays enabled. has-functional-changes.sh also lists LICENSE* and README.md in IGNORE_DIFF_ON, so this is classified as no functional change. An added fragment gives a minor bump, 3.1.1 → 3.2.0, which is right — and a release is needed anyway to push the metadata above onto PyPI.

Optional

• Copyright year 2025-2026 rather than 2025 — first commit is 2025-02-25 and work continues into 2026. (microdf uses an individual holder, but "PolicyEngine" is the better choice here with four authors and org ownership.)
• A licence line in paper/ — grep -rni "licen\|copyright" paper/ returns nothing, so a reader of the manuscript alone has no signal.
• license: MIT in CITATION.cff once this lands (that file arrives in Add JOSS paper for submission #201).

Not caused by this PR

The red Lint check is pre-existing. I checked the base commit rather than assuming: ruff format --check reports the same five files on main as on this branch, and none of them are touched here. Worth not letting "CI is red anyway" obscure the changelog point above.