Security: remove eval() from parameter breakdown homogenization

[MaxGhenis]

Summary

homogenize_parameters.py executes eval() on YAML-sourced breakdown metadata.

Severity

High

Impact

A malicious or compromised parameter package can execute arbitrary Python during parameter homogenization.

Affected code

• policyengine_core/parameters/operations/homogenize_parameters.py:57-75

Details

When first_breakdown is neither a list nor a known variable, the code falls back to:

possible_values = list(eval(first_breakdown))

That input originates from parameter metadata.

Expected behavior

Breakdown metadata should be treated as data, not executable code.

Suggested remediation

• Remove eval() entirely
• Support only explicit lists or known enum/bool-backed variables
• If string parsing is needed, use a strict parser with an allowlist grammar
• Add a regression test proving arbitrary expressions are rejected

[MaxGhenis]

Closed by #453: parameter breakdown homogenization now evaluates only safe allowlisted literal/range forms instead of using eval().