Related Servers

Hundreds of tools are exact or near-exact duplicates: bjs is the same estimator as borusyak_jaravel_spiess and did_imputation; gardner_did duplicates did_2stage; frontdoor duplicates front_door; rosenbaum_bounds and rosenbaum_gamma share identical descriptions; postestimation_contract and postestimation_report are identical; and there are dozens of R-style or article-facing aliases (synthdid_estimate, sc_estimate, did_estimate, xlearner, psm). An agent cannot reliably distinguish the intended tool among so many overlapping aliases and variants.

Names are mostly snake_case, but the convention is inconsistent: some are concise verbs (did, regress, test, contrast), some are bare nouns (bridge, panel, rate), some are R-package aliases (synthdid_estimate, did_estimate), and some use different spellings of the same concept (frontdoor vs front_door, psm vs psmatch2 vs match). Abbreviated and opaque names like discos, megamma, sqreg, and rlassologit_effects break any predictable verb_noun pattern.

548 tools is an extreme mismatch for any server purpose, far beyond even the 50+ upper bound. The surface is a sprawling econometrics library rather than a curated decision-agent toolset, and an agent would face a prohibitive selection problem before doing any actual analysis.

For the causal-inference domain, coverage is effectively exhaustive: DiD (2x2, staggered, continuous, DDD, event studies), RD (sharp/fuzzy/multi-cutoff/bunching), IV (k-class, weak-instrument, shift-share, MR), synthetic control variants, matching and weighting, mediation, decompositions, sensitivity analysis, survival, time series, CATE/meta-learners, causal discovery, and offline policy learning are all represented. There are no obvious methodological gaps; the problem is surplus, not scarcity.