Fullbleed 2.5.3: validate reproducibility baselines

Fullbleed 2.5.3 fixes reproducibility checks that could report a pass for a baseline with missing hashes, such as {}. Invalid records now fail with the structured REPRO_RECORD_INVALID diagnostic while retaining the rendered PDF and previews for review.

• Detect asset-lock additions, removals, and content changes.
• Reject record/check paths that refer to the same file, preserving the existing baseline.
• Expose the shared baseline schema through fullbleed --schema repro-record.
• Continue accepting valid older v1 records and additive metadata.

Install with python -m pip install --upgrade fullbleed==2.5.3 or use the fullbleed = "2.5.3" Rust crate. This patch does not change rendering algorithms or add Python runtime dependencies.

Validation: 491 Python tests, 22 real CLI regression scenarios on Linux and Windows, 15 wheel builds, 19 installed-wheel checks, and fresh public-package installation checks. A baseline generated with the public 2.5.2 wheel still passes with identical PDF bytes for the retained fixture. The attached verification record, engineering evidence, and SHA-256 sums document the tested scope; they do not assert independent PDF/VT certification.

Change details · Source tag