SectionCheck 0.2.1.dev0 — review corrections

Corrects four errors found in review of 0.2.0.dev0:

• Non-regular raster inputs, including named pipes, now reject without blocking. File reads enforce byte limits on the opened descriptor.
• Unrepresentable calibration values return a structured rejection instead of an uncaught overflow.
• Affine matrices require the exact last row [0, 0, 1]; projective terms are not silently ignored.
• ROIs partly or wholly outside preview crops now produce specific visibility warnings that require acknowledgment.

Hosted CI passes 90 tests and a fresh installed-wheel protocol exercise. The six-panel report passes offline desktop and 390-pixel checks. See CHANGELOG.md, VALIDATION.json and SHA256SUMS.

Generate a new report and decision after upgrading. The tool version and source hashes change, so 0.2.0 decisions are stale and cannot authorize export.

Live synthetic example · Workflow

This remains an independent, AI-assisted synthetic engineering prototype. Researcher acceptance, VALIS integration and external-viewer interoperability remain unvalidated. The earlier release artifacts are retained unchanged.

SectionCheck 0.2.0.dev0 — synthetic review prototype

Superseded by 0.2.1.dev0, which fixes input handling and preview visibility errors. Generate a fresh report and decision after upgrading. These older assets remain unchanged.

CPU-only synthetic affine and ROI review prototype. Developed with AI assistance; independent of Neuralink, VALIS and QuPath.

• Strict coordinate, manifest and Polygon/MultiPolygon schemas, including holes.
• Six-panel report; explicit decisions bind inputs, tool source, schemas and geometry engine versions.
• Approved export and local profile re-import; stale decisions and edited output are rejected.
• 81 tests pass with zero failures/errors/skips in GitHub CI. A fresh installed wheel passes the positive and negative protocol exercise.
• Local and hosted wheel builds have the same SHA-256. See VALIDATION.json and SHA256SUMS.

Live synthetic example · Workflow and limitations

Only authored synthetic data is accepted. Automated fixture decisions are not researcher approval. VALIS integration, external-viewer interoperability and researcher usefulness remain unvalidated. This prerelease does not claim completion of the research delivery gates.

Install the wheel with Python 3.12 or later. requirements.lock records the exact tested Python 3.12 environment; source and examples are in the GitHub source archive.