Fix Privacy Bugs in Odometer and Covariance

[joshua-oss]

This PR fixes:

• Odometer Formula Bug #621 : Odometer Formula Bug
• Covariance Clipping Bug #622 : Covariance Clipping Bug

#621 could result in reporting the wrong privacy consumption, and #622 could result in privacy leak

[Copilot]

Pull request overview

Fixes two privacy-accounting bugs affecting reported ε/δ consumption and covariance sensitivity bounds, aligning implementation with intended DP guarantees.

Changes:

• Correct homogeneous odometer “optimal_b” formula by adding the missing log(1/tol) term (Issue #621).
• Fix DP covariance release to compute covariance from clipped/cleaned data rather than the raw input (Issue #622).
• Add a unit test to ensure DPcovariance.release() uses clipped data; add a sklearn import fallback for LinearModel.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
sql/snsql/sql/odometer.py	Fixes the homogeneous composition bound by applying the missing log factor in optimal_b.
sql/tests/odometer/test_odometer.py	Updates the expected ε value to match the corrected odometer formula.
synth/snsynth/models/dp_covariance.py	Ensures covariance is computed from new_data (clipped + imputed) to preserve bounded sensitivity.
synth/tests/test_dp_covariance.py	Adds regression test validating release() uses clipped data (noise forced to zero).
synth/snsynth/models/linear_regression.py	Adds compatibility import path for LinearModel across sklearn versions.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.