Stabilize UKF covariance symmetrization

[FlorianPfaff]

Summary

• symmetrize UKF prediction and update covariances with a scale-normalized average
• avoid overflow for finite covariance entries near the float64 limit
• add a focused regression for maximum-finite process covariance

Root cause

src/pyrecest/filters/_ukf.py used 0.5 * (P + P.T). For a symmetric covariance containing np.finfo(np.float64).max, the intermediate sum overflows even though the mathematical average remains finite.

Impact

A valid finite process covariance could raise FloatingPointError under strict NumPy error handling or become non-finite during UKF prediction. The update path used the same vulnerable expression.

Fix

The new helper scales each transposed entry pair by its largest magnitude before averaging, then restores the scale. This keeps finite pairs within [-1, 1] during the intermediate addition and also preserves subnormal values.

Validation

• reproduced the original FloatingPointError: overflow encountered in add
• focused patched harness preserves np.finfo(np.float64).max exactly and regenerates finite sigma points
• helper edge cases cover same-sign maxima, opposite-sign extremes, zeros, and the minimum positive subnormal
• both modified Python files pass python -m py_compile
• full GitHub Actions matrix passes on NumPy 3.11–3.14, JAX 3.11–3.13, and PyTorch 3.11–3.13
• lint, packaging, minimal installation, documentation, MegaLinter, CodeQL, dependency review, security, and release-preview workflows pass
• comparison with the latest main remains limited to exactly two changed files

[github-actions]

✅MegaLinter analysis: Success

Descriptor	Linter	Files	Fixed	Errors	Warnings	Elapsed time
✅ COPYPASTE	jscpd	yes		no	no	26.52s
✅ JSON	prettier	7	0	0	0	1.29s
✅ JSON	v8r	7		0	0	4.13s
✅ MARKDOWN	markdownlint	68	0	0	0	2.16s
✅ MARKDOWN	markdown-table-formatter	68	0	0	0	0.83s
✅ PYTHON	black	1808	55	0	0	95.99s
✅ PYTHON	isort	1808	86	0	0	2.74s
✅ REPOSITORY	betterleaks	yes		no	no	2.58s
✅ REPOSITORY	checkov	yes		no	no	55.85s
✅ REPOSITORY	gitleaks	yes		no	no	17.39s
✅ REPOSITORY	git_diff	yes		no	no	0.31s
✅ REPOSITORY	secretlint	yes		no	no	84.22s
✅ REPOSITORY	syft	yes		no	no	6.89s
✅ REPOSITORY	trivy-sbom	yes		no	no	7.57s
✅ REPOSITORY	trufflehog	yes		no	no	32.97s
✅ YAML	prettier	11	0	0	0	0.8s
✅ YAML	v8r	11		0	0	13.36s
✅ YAML	yamllint	11		0	0	0.61s

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 low value despite the green matrix. It adds a specialized symmetrization algorithm solely to preserve covariance entries near the maximum finite float64 value. Such covariances are not a meaningful UKF operating regime, and the added numerical path would burden core filter code.