◆ fingerprint
How the atlas decodes mixed soil.
MS² features that let us tell one phylum apart from another — collectively, the fingerprint that turns a community of spectra into phylum percentages.
Each biomarker below contributes to Bray–Curtis dissimilarity between phyla. We publish the top contributors per phylum; a full sample is decomposed by projecting its spectrum onto this basis. The current implementation is SIMPER v1 — swappable with a spectral foundation model (L1 Identity) when that layer ships.
◈ proof
analysis-19 ·
iter3a · floor 0.20
IS + RIE correction
ClimGrass soil, decomposed.
Projecting a 12-sample long-term warming × drought experiment onto the fingerprint, averaged across samples — mean phylum composition rolled up to kingdom. These are the numbers the manuscript carries.
Fungi
48.2%
Plantae
34.9%
Animalia
8.1%
Bacteria
7.6%
Protists
1.2%
Archaea
0.0%
The analysis-19 IS + RIE + ArchLips + floor=0.20 configuration lifts Bray-Curtis plausibility from 0.341 → 0.131 — a 62% improvement over raw quantification. Published treatment effects are preserved.
◆ per-phylum fingerprints
top 5 contributors shown per phylum
Bacteria
· 3 phylaActinomycetota
top 1Bacillota
top 4Archaea
· 3 phylaEuryarchaeota
top 1Methanobacteriota
top 1Thermoproteota
top 2Fungi
· 2 phylaAscomycota
top 1Basidiomycota
top 1Plantae
· 1 phylumMarchantiophyta
top 1Animalia
· 2 phylaArthropoda
top 1
method note ·
fingerprint scores are SIMPER contributions to Bray–Curtis dissimilarity (Clarke 1993),
normalized per-biomarker to 0–1 for display. Phylum resolution is the publication
scale; kingdom-level is a sum aggregate. The data contract is method-agnostic —
when the spectral foundation model (L1) ships, the same table gains a second row
per biomarker with method=foundation_v1 and the UI exposes a toggle.