microcosm / lineage

observability · US stacked pool

Where every variable comes from.

Every imputed variable in the pool is declared once, in a spec: which stage draws it, from which donor, with which predictors, under which model. This page reads that spec — nothing here is hand-typed, and a conformance test fails the build if the code drifts from it.

spec: loading… Read the spec
01 — the shape

Six lineage classes; two of them draw.

Measured, certified, and computed variables are carried or derived deterministically. Only the two transfer classes draw from a fitted model — early gap-fill (ASEC → ACS spine) and late transfer (ASEC clone → whole pool) — so those are the only places quality can drift from the donor, and the only places a predictor set or a calibration choice matters.

02 — the model

One conditional model, everywhere.

03 — predictor sets

What each family conditions on.

04 — every imputed variable

Search the lineage.

VariableEntityFamilyStageDrawPredictorsDonor
05 — computed, not drawn

Deterministic producers.

ProducerKindInputsOutputs