--proxy-hints (chi-squared proxy-variable detection, #61) is only wired into faircode profile in faircode/cli.py - faircode compare has no equivalent flag, even though faircode/proxy.py's proxy_hints() doesn't depend on anything profile-specific.
Comparing a baseline dataset against a current one is arguably a more natural place for this signal than a single-dataset profile: a protected-attribute proxy that's weak in the baseline but has strengthened in the current data is exactly the kind of drift faircode compare already exists to catch (see --fail-on-drift), and right now there's no way to ask it to also flag a new/strengthened proxy relationship.
Scope: add a --proxy-hints flag to the compare subparser in faircode/cli.py, call proxy_hints() on both datasets (or the diff, whichever makes more sense - worth discussing in the PR), and surface the result in faircode/report.py's compare output (terminal, --json, and --html) the same way profile --proxy-hints already does.
--proxy-hints(chi-squared proxy-variable detection, #61) is only wired intofaircode profileinfaircode/cli.py-faircode comparehas no equivalent flag, even thoughfaircode/proxy.py'sproxy_hints()doesn't depend on anything profile-specific.Comparing a baseline dataset against a current one is arguably a more natural place for this signal than a single-dataset profile: a protected-attribute proxy that's weak in the baseline but has strengthened in the current data is exactly the kind of drift
faircode comparealready exists to catch (see--fail-on-drift), and right now there's no way to ask it to also flag a new/strengthened proxy relationship.Scope: add a
--proxy-hintsflag to thecomparesubparser infaircode/cli.py, callproxy_hints()on both datasets (or the diff, whichever makes more sense - worth discussing in the PR), and surface the result infaircode/report.py's compare output (terminal,--json, and--html) the same wayprofile --proxy-hintsalready does.