No user-submitted related servers found.
Scored across 45 tools
Descriptions work hard to cross-reference each other (e.g., brief says 'Same output as brief_result'; honest_did points to honest_did_from_result), but several pairs genuinely overlap: brief/brief_result, audit/audit_result, sensitivity/sensitivity_from_result, iv vs ivreg, and honest_did vs honest_did_from_result. The advisory cluster (detect_design, check_identification, preflight, recommend, route_estimator, causal) is also hard to separate without reading ordering hints, though the descriptions do provide them.
Mostly snake_case, but two conventions coexist: verb_noun for plumbing (load_data, transform_data, search_functions, plot_from_result) and bare estimator/package names for methods (did, iv, dml, synth, match, sdid, feols, ivreg, rdrobust, ebalance). The result-accessor suffix is also inconsistent, mixing '_result' (brief_result, audit_result) with '_from_result' (honest_did_from_result, sensitivity_from_result, vcov_from_result, plot_from_result).
45 individually listed tools is heavy for any single server, and several slots are redundant duplicates (brief vs brief_result, audit vs audit_result, sensitivity vs sensitivity_from_result) rather than earned surface. The search_functions/describe_function/call_function escape hatch already covers the ~1,200-function registry, so the long tail of listed estimator tools is partly unnecessary.
Coverage of the causal-inference lifecycle is strong: data load/transform/profile, design detection and preflight, estimators across DID/IV/RD/matching/DML/synth, robustness and sensitivity, pipelines, plotting, interpretation, and citation lookup. The main gap is state management — there is no way to list existing data_ids or result_ids, or to delete/export them, which forces agents to carry handles out-of-band.
