GeoStep 0.5 private-release verification

This record covers the 0.5.0 candidate prepared for private GitHub release assets. The repository was confirmed private. No push, tag, GitHub release, upload or documentation deployment forms part of these checks.

Software and numerical checks

  • The checkout suite passed: 169 tests on Python 3.12.11, with seven known warnings from malformed-date, constant-data diagnostic and third-party plotting cases.

  • Ruff passed for source, tests, runner, scripts, examples, benchmarks and the documentation-site builder.

  • The bounded parallel inference screen completed 1,600 replicates with zero failures. Its machine-readable result records the seed, source hashes and environment.

  • The frozen 0.5 release screen completed 3,000 panels and 12,000 analysis calls with zero failures in 189 seconds. It retains source hashes, 48 cell summaries and every replicate.

Installed-wheel checks

The wheel was installed outside the checkout, then tested with a copied test, runner, script and example tree so imports resolved from the installed package.

Environment

Dependency resolution

Result

Python 3.11.15

Permitted package bounds

pip check and 169 tests passed; seven warnings

Python 3.12.11

Permitted package bounds

pip check and 169 tests passed; seven warnings

Python 3.12.11 with pytest 9

Updated test extra

pip check and 169 tests passed; seven warnings

The test extra now permits pytest 9 because the prior <9 bound excluded the available security fix. The runtime dependency audit found no known vulnerabilities. The audit cannot query the private geostep distribution on PyPI and reports that expected skip separately.

Packaging and release hygiene

  • Source and wheel distributions built successfully and passed twine check.

  • A tracked-file pattern scan found no committed credential, private-key or secret assignment. It is a release hygiene check, not a substitute for credential revocation or an external secret-scanning service.

  • The proprietary licence is included in package metadata and private distribution remains the required route.

  • The internal AI review is absent from public documentation source and its generated HTML; its record is retained only under .planning/.

Release boundary

The supported private release scope is complete or prespecified-stratified parallel allocation with DiD or Lift under the documented conditions. Stratified DiD, BCa, CRT and staircase inference remain experimental. CRT and staircase power are unsupported. The retained screens do not establish causal validity under missingness, interference, non-adherence or outcome-driven allocation.