ReproMin v0.1.0.dev9

ReproMin v0.1.0.dev9 (pre-release)

ReproMin reduces a failing repository while continuously checking that the
configured failure still occurs. This pre-alpha release adds a read-only way
to compare validated reduction evidence from multiple runs.

Highlights

• Added repomin report compare REPORT.json REPORT.json ... with text, JSON,
  and Markdown output.
• Comparisons expose aggregate reduction evidence only: sizes, retention,
  reduction counts, attempts, cache hits, budget and holdout state, phase
  coverage, and adjacent numeric deltas.
• Context drift is called out for versions, backends, execution settings,
  budgets, holdout controls, phase definitions, sampling, input selection, and
  oracle identity. Private values are compared internally through opaque
  digests and are never emitted.
• Added Bash, Zsh, Fish, and PowerShell completion for the compare workflow.

Boundaries

report compare validates report structure only. It does not read payloads,
execute commands, access the network, or claim performance, correctness,
causality, or production reliability. ReproMin remains a pre-alpha feasibility
build; the default host backend is not a sandbox and the optional Docker
backend is not a complete security boundary.

Verification

• 586 unit tests passed (2 platform tests skipped).
• 16 offline benchmark cases passed.
• Ruff, documentation checks, package build, twine check, wheel/sdist
  installation checks, and pip check passed.

Artifacts

File	SHA-256
repomin-0.1.0.dev9-py3-none-any.whl	b8b6d4ec15bfe46a5cef3c75ae6e6b1f5d945ac4902d104f9e61640c7fca90a4
repomin-0.1.0.dev9.tar.gz	77ece1fe5dc5bbf4d6703944b97a39df51e64ff3107edcdeb30a7c26ebb3b716

Install from the wheel or source archive below, then run python -m repomin --version. Please report real-world pilot results, sanitized fixtures, and
workflow friction in Discussion #16
or the issue templates.