Reject broadcast process covariance in UKF prediction

[FlorianPfaff]

Bug

The low-level unscented Kalman filter added self.Q directly to the predicted covariance without validating its matrix shape.

For a two-dimensional state, a vector such as Q = [0.4, 0.2] was therefore broadcast across the covariance matrix by NumPy. After the existing symmetrization step, an identity prediction silently produced a spurious cross-covariance of 0.3 instead of rejecting the malformed process-noise input.

Fix

Convert the configured process covariance once, require the exact (dim_x, dim_x) shape, and reuse that validated matrix in the covariance update. Validation happens before filter-state mutation.

Regression coverage

The new public regression verifies that a vector-valued process covariance:

• raises a clear ValueError;
• does not mutate the state mean;
• does not mutate the state covariance.

Validation

• python -m py_compile passed for the modified implementation and regression module;
• isolated NumPy reproduction confirmed the previous vector broadcast introduced a 0.3 off-diagonal covariance;
• branch comparison: 2 commits ahead, 0 behind current main;
• final diff is limited to the UKF process-covariance shape guard and one focused regression module.

GitHub Actions is the authoritative full backend, lint, packaging, documentation, and integration validation.

[github-actions]

✅MegaLinter analysis: Success

Descriptor	Linter	Files	Fixed	Errors	Warnings	Elapsed time
✅ COPYPASTE	jscpd	yes		no	no	23.66s
✅ JSON	prettier	7	0	0	0	1.42s
✅ JSON	v8r	7		0	0	4.47s
✅ MARKDOWN	markdownlint	68	0	0	0	1.96s
✅ MARKDOWN	markdown-table-formatter	68	0	0	0	0.8s
✅ PYTHON	black	1866	87	0	0	90.97s
✅ PYTHON	isort	1866	139	0	0	2.75s
✅ REPOSITORY	betterleaks	yes		no	no	2.06s
✅ REPOSITORY	checkov	yes		no	no	52.35s
✅ REPOSITORY	gitleaks	yes		no	no	16.75s
✅ REPOSITORY	git_diff	yes		no	no	0.2s
✅ REPOSITORY	secretlint	yes		no	no	81.2s
✅ REPOSITORY	syft	yes		no	no	5.57s
✅ REPOSITORY	trivy-sbom	yes		no	no	5.62s
✅ REPOSITORY	trufflehog	yes		no	no	31.06s
✅ YAML	prettier	11	0	0	0	0.86s
✅ YAML	v8r	11		0	0	11.53s
✅ YAML	yamllint	11		0	0	0.63s

Notices

📣 MegaLinter 9.5.0 is out! Discover the new features and security recommendations in the release announcement. (Skip this info by defining SECURITY_SUGGESTIONS: false)

See detailed reports in MegaLinter artifacts

Your project could benefit from a custom flavor, which would allow you to run only the linters you need, and thus improve runtime performances. (Skip this info by defining FLAVOR_SUGGESTIONS: false)

• Documentation: Custom Flavors
• Command: npx mega-linter-runner@9.6.0 --custom-flavor-setup --custom-flavor-linters PYTHON_BLACK,PYTHON_ISORT,COPYPASTE_JSCPD,JSON_V8R,JSON_PRETTIER,MARKDOWN_MARKDOWNLINT,MARKDOWN_MARKDOWN_TABLE_FORMATTER,REPOSITORY_CHECKOV,REPOSITORY_GIT_DIFF,REPOSITORY_GITLEAKS,REPOSITORY_BETTERLEAKS,REPOSITORY_SECRETLINT,REPOSITORY_SYFT,REPOSITORY_TRIVY_SBOM,REPOSITORY_TRUFFLEHOG,YAML_PRETTIER,YAML_YAMLLINT,YAML_V8R

Show us your support by starring ⭐ the repository

[FlorianPfaff]

Closing as superseded by #5253. Both PRs address the same malformed multidimensional UKF process-noise covariance broadcasting bug; #5253 is the refreshed current-main version and explicitly retains the one-dimensional scalar/length-one shorthand plus atomic state preservation. Keeping both open would duplicate review and CI work.