v0.5.2

Merged Pull Requests

• Add bounded polynomial determinants and release benchmarks #318
• Add outward-rounded interval division and solve enclosures #315
• Add exact dot products and rectangular column selection #314
• Export verified CI evidence through shared tooling #312
• Bump the github-actions group with 3 updates #310
• Unify Python tooling and move consumer tests #309

Added

• Add exact dot products and rectangular column selection #314
  979041e

feat: add exact dot products and rectangular column selection

• Add borrowed RationalVector dot products and a shape-checked rational_dot
  slice adapter without a runtime dimension cap or binary64 intermediates.

• Select rectangular column bases with explicit absolute tolerances,
  deterministic partial pivoting, and provisional numerical rank.

• Document exact certification boundaries and Delaunay workflows through 6D,
  with representative benchmarks and independent rational identities.

• Validate with 1,078 Rust tests, 281 Python tests, both doctest configurations,
  and 109 diagnostic benchmark cases; complete all CI validation stages.

• Add outward-rounded interval division and solve enclosures #315
  1328f9d

  • Add const reciprocal and direct division bounds for zero-free intervals,
    including subnormal quotients and typed finite-range failures.
  • Enclose fixed-dimension linear solutions with row pivoting and explicit
    inconclusive outcomes that preserve the caller's robust fallback.
  • Document the all-admitted-systems contract and add diagnostic solve corpora
    without changing the retained release benchmark catalog.
  • Refresh rustix and the contributor uv and cargo-nextest pins.

• Add bounded polynomial determinants and release benchmarks #318
  d87b10b

  • Add canonical sparse rational polynomials and exact determinant evaluation
    through order eight with caller-selected work budgets and typed failures.
  • Capture all 1,403 public benchmark cases in parallel default/exact release
    suites, then verify and archive their raw samples and original provenance.
  • Adopt research-repo-tools v0.2.4 for shared updater and release commands;
    keep Semgrep setup and fixture checks in the standard Python 3.15
    environment until the tracked v0.2.5 cleanup.
  • Validate with the full Rust CI pass, final check gate, and 303 Python
    consumer tests after shared-package adoption.

Changed

• Export verified CI evidence through shared tooling #312
  7320d6c

refactor(release): export verified CI evidence through shared tooling

• Append deterministic identity JSON with research-repo-tools v0.2.3 so
  normal and quiet Just execution preserve the same release evidence.
• Reject missing output destinations and export failures while retaining
  signed evidence checks and exact identity revalidation after approval.
• Forward verifier options through Just; pinned verification now uses
  just release-evidence --identity JSON instead of a positional identity.
• Validate with just ci on macOS: 269 Python tests, 1,031 Rust tests,
  doctests, examples, benchmark compilation, and repository checks pass.

fix(tests): preserve Windows release fixture paths and diagnostics

• Pass the fixture import path through the environment instead of embedding
  Windows backslashes in Python source across Bash and native argv parsing.

• Cover backslashes, apostrophes, and CRLF diagnostics through the real Just
  boundary while retaining exact LF command-file assertions.

• Require the verifier's actual result and rejection reason, including in
  quiet mode, so bootstrap failures cannot satisfy negative test cases.

• Document portable subprocess fixtures and validate with 271 Python tests,
  strict types, Markdown, spelling, and documentation checks on macOS.

• Adopt shared release orchestration dffae7a

  Replace the three release adapters with published research-repo-tools 0.2.4 commands and declarative policies. Run evidence and asset commands in isolated
  Python outside the observed checkout, preserving frozen post-approval revalidation before OIDC.

  Make performance release consume saved preparation; tag fetching is explicit and README publication remains independently retryable. Preparation positional
  tags move to the shared declaration, and evidence --identity JSON becomes --expected PATH. Retain native policy, workflow transport and scientific publication
  regressions.

Dependencies

• Bump the github-actions group with 3 updates #310
  110b7e5

  Bumps the github-actions group with 3 updates: github/codeql-action/upload-sarif,
  github/codeql-action/init and github/codeql-action/analyze.

  Updates github/codeql-action/upload-sarif from 4.38.3 to 4.39.0

  • Release notes
  • Changelog
  • Commits

  Updates github/codeql-action/init from 4.38.3 to 4.39.0

  • Release notes
  • Changelog
  • Commits

  Updates github/codeql-action/analyze from 4.38.3 to 4.39.0

  • Release notes
  • Changelog
  • Commits

Fixed

• Preserve CI evidence for crates.io publishing 2702cb7

  • Capture release-evidence JSON without suppressing stdout.
  • Reject empty evidence identities before exporting job outputs.
  • Update ty to 0.0.91 and Semgrep's pycparser resolution to 3.12.

• Patch Semgrep telemetry dependencies and update Ruff 988dfc9

  • Resolve GHSA-xcq7-p3pf-p88h by requiring OpenTelemetry API 1.45.1.
  • Align the telemetry SDK, OTLP exporter, and instrumentation releases.
  • Update Ruff from 0.17.1 to 0.17.2.

Maintenance

• Unify Python tooling and move consumer tests #309
  8769cc4

build: unify Python tooling and move consumer tests

• Run Semgrep in the root Python 3.15 environment with native wheels,
  patched dependency overrides, and the Linux stack-limit workaround.
• Inherit pytest, Ruff, and Ty versions from research-repo-tools and
  place Python consumer tests under tests/tooling with preserved scans.
• Align CI, audits, dependency automation, and contributor guidance;
  refresh cc and dprint and clarify scientific regression requirements.

fix: make Semgrep consumer scans portable

• Cap stack limits in the Linux scanner child for direct pytest runs.
• Compare native report paths consistently across Windows and Unix.

Performance

• Focus release benchmark acquisition 0d15f45

  • Separate full public la-stack coverage from the 16 local peer LU-and-solve
    cases, and omit peer timings from hosted release acquisition.
  • Cover construction, storage, angles, Gram matrices, interval filters,
    certified linear forms, and exact conversions with dedicated native cases.
  • Keep retained LU publication retryable, preserve historical evidence, and
    expose case budgets and elapsed runtime without reducing sampling.
  • Validate with just ci and all 1,270 selected native smoke checks.