Cohorts

synthetic demo data

Saved cohorts · versioned

membership never changes silently
CohortDefinitionIncludedState
COH-NUE-2526nitrogen-use efficiency · stratified by soil series193 method incompatibilities flagged
COH-COVER-SB-v2overwinter cover vs stubble · spring barley · VESS + yield11current
COH-RN-METAreduced-N systems · yield held vs standard31current
COH-SOM-LOIsoil organic matter · LOI method only4below k=5 — aggregates withheld
COH-MT-TRANSmin-till transition years 1–3 · yield + fuel14current

COH-BST-WW-v4 · membership

Definition (canonical filters, exposed): crop = winter wheat · practice = biological seed treatment · reference arm present · metric = yield · unit = t/ha · basis = whole field · method = weighbridge (v4: + calibrated yield monitor) · seasons 2024–25

23
strictly compatible · poolable
105
excluded · every reason named
+0.2
median difference t/ha · n=23 · opens to sources
RecordDifferenceStandingIncluded?Reason

19 further included and 101 further excluded records · adverse and null results carry the same weight in membership as favourable ones.

Provenance graph · how the evidence connects

structure, not just totals
OBS DEC ACT CMP OUT EVD one artefact supports two facts — reuse made visible

Cohort diff · v3 → v4

what changed, exactly
ChangeRecordsWhy
+ admitted6amendment 14 Feb 2025: calibrated yield monitors accepted
− removed2addendum on PRF-UK-A-000188 changed N context — no longer same-management
± value changed1corrected rate propagated · median unchanged

v3 remains frozen and reproducible; every export cites the version it used. New outcomes and addenda re-run the diff, never silently change the result.

SnapshotCohortFrozen
SNAP-2026-06-30COH-BST-WW-v330 Jun 2026 · cited twice
Cohort rules: saved, versioned, snapshot-able; strictly-compatible vs related-context is explicit; every inclusion and exclusion carries its reason; aggregates only over poolable records with k=5 fail-closed; every number opens to sources.