research paper · design doc, pre-experiment

Composition: what running imputation and calibration together does

The imputation and calibration dossiers each validate one operator with the other held fixed. This paper asks what happens when both run on the same file — measured where neither single-operator study can look: realized landmines, calibration's power to repair a bad imputation, the target-vs-survey trade-off as an unscalarized Pareto frontier, and whether effective support quietly collapses under composition.

Repository Strategy page

status: Design doc only. Zero experiments have run; no results exist. The repository holds a README stating the four questions and nothing else.
blocked on: populace#302 (holdout-masked builds, not yet merged), the calibration-paper method-adapter registry (not yet built), and the popdgp harness extraction (in progress, unmerged).
position in the portfolio: Third leg after the operator dossiers (imputation-paper, calibration-paper, sparsity-paper) and the popdgp referee — composition is what the single-operator papers cannot see.

status

No manuscript yet

composition-paper · repository created 2026-07-04, design-doc stage The repository states four questions (realized landmines, repairability, the target-vs-survey Pareto frontier, effective-support collapse) and nothing further — no code, no data, no manuscript, no PDF. Every term used on the composition strategy page describes this paper's plan, not a finding. TODO: replace this block with the paper viewer once the blocked-on dependencies land and the first experiment runs.