MICA is a Bayesian framework for imputing missing cells in a partial correlation matrix — the matrix you get when a meta-analysis pools effect sizes across studies but no single study measured every pair of variables. The framework combines a Cholesky-factor positive-definite parameterization, anchor-regression-bound priors, and per-cell identifiability triage to produce calibrated credible intervals where the data supports them and honestly wide intervals where it doesn't.
The R pipeline that powers this site lives in the project repo under mica/R/. The web companion (this site) is in mica/web/; the Plumber API is in mica/server/.
Matrices submitted to the API are processed in-memory and not logged beyond standard request metadata. The optional LLM parsing fallback (when configured by the operator) forwards the matrix text to Anthropic's API; if your data is sensitive, prefer the deterministic CSV/Excel paths.