Compare reproductions

Grouped by theme, each result shown against the reference it was measured on. A record with gated metrics but no single quotable figure shows how many of its gates it cleared. A record that declares floors nobody has run yet reads "not yet measured" with the count still waiting, which is a different thing from having no floors. A record assessed against a reference product in prose says that instead of inventing a number. "not-checkable" is reserved for its ADR-0058 sense: the paper reports nothing a rerun can be checked against. All four are honest outcomes, not failures.

Land cover & change

reproductionresultvs referencetierstatus
MapBiomas LULC, fire, water — Rondonia (2023) lulc overall accuracy 88.9% — human validated
LandTrendr temporal segmentation — PNW (Kennedy et al. 2010) assessed in prose 3 observations Detecting trends in forest recipe unverified
Solar-on-forest trajectory similarity — Massachusetts not yet measured 0 of 4 gates run synthetic inputs — recipe unverified

Hazards (fire, flood, vulnerability)

reproductionresultvs referencetierstatus
LandTrendr burn detection — 2021 Bootleg Fire (Kennedy et al. 2010) peak_to_trough disturbed recall producers accuracy 77.7% MTBS Bootleg-2021 perimeter (mtbs_perimeter_data.zip), rasterised to the run grid agent validated
Climate vulnerability model for Malawi (Malcomb et al. 2014) not yet measured 0 of 3 gates run validation/reference-fig4-ta-classes.geojson human unverified

Vegetation

reproductionresultvs referencetierstatus
Alpine Treeline Ecotone detection (Wei et al. 2020) spatial correlation 98.1% data/weiss_268_treeline.csv human validated

Mineral prospectivity

reproductionresultvs referencetierstatus
Embedding-analog mineral prospectivity — Mberengwa belt, Zimbabwe gates cleared 6 of 6 research/mineral-prospecting/occurrences/occurrences.geojson agent validated