The Earth is a building made from local material: one quarry, one recipe from the first course to the last, and a recipe no sample in any museum drawer matches. Paolo Sossi and Dan Bower at ETH Zurich’s Institute of Geochemistry and Petrology showed it in Nature Astronomy this spring; ScienceDaily’s September re-run has it circulating again. Material from beyond Jupiter is less than 2 percent of Earth’s mass, possibly none.

Meteorites come in two families: non-carbonaceous from the inner Solar System, carbonaceous from beyond Jupiter. Earlier models compared two isotope systems against a chosen inner reference and put the outer share at about 6 or about 40 percent, depending on whether enstatite chondrites or ureilites stood in. Without “multiple isotopic systems simultaneously”, the abstract notes, Earth’s provenance “remains equivocal”. This is science refining its own answer.

A data-science experiment

No new sample was measured. This is Zurich work on open data: Sossi and Bower re-read ten nucleosynthetic anomalies at once across 14 reservoirs, from ureilites to Mars, mostly from the public OriginsLab database. The tools, rare in geochemistry: principal component analysis with Bayesian latent factor analysis in PyMC, York regression carrying every uncertainty, Monte Carlo propagation over 10,000 iterations. “Our studies are actually data science experiments,” Sossi says in the ETH release.

Line up the inner Solar System bodies in any two of the ten anomalies and the line hits Earth within one standard deviation, every time; the mean Z-score is 0.65 ± 0.55. Calcium-48 alone caps outer material at under 2 ± 0.2 percent of the bulk silicate Earth. Molybdenum and ruthenium cap CI-type material at under 0.3 ± 0.1 percent.

The recipe was, on average, “unlike that of any chondrite”: it appears only when ten measurements are read together. And it held through construction. Molybdenum and ruthenium, pulled into the core, carry the same signature as titanium and calcium, which stayed in the mantle. Early and late deliveries drew on what Sossi calls “a single material reservoir”.

Headlines skip the people underneath: the geochemists who measured those anomalies sample by sample and published them openly. Like Andrew Tallon’s Notre-Dame point cloud in PAZ’s 3D Reconstruction concept panel, their work was ready before the question was.

←TODAY: Ten isotope systems read from public data cap Earth’s outer-Solar-System share below 2 percent.
→3012: Zurich certifies a reclaimed beam’s origin by reading every test together, never one.
Fulcrum: The same method traces a planet and a beam; provenance is a pattern, not a match.

The line to the building site is one of method: one pair of measurements can match more than one source; read many together and the ambiguity collapses. For a Swiss office tracing reclaimed stone or a reused steel section, that is the whole discipline.

Atelier: Offices adopting AI for material checks meet the same two-at-a-time ambiguity: hand a model one certificate and one reference and it will often find a match. Monday move: when the next test report lands, ask the lab for the full panel as a spreadsheet, not the summary PDF, and read every column together before anyone signs.

Hack: Measure how far each sample sits from a test panel’s main axis, reading every column at once. It is a bare-bones cousin of Sossi and Bower’s principal component analysis, with no uncertainties carried: a flag, not a verdict. The rows printed last sit furthest off the shared pattern; question those first.

import numpy as np, pandas as pd
X = pd.read_csv("panel.csv", index_col="sample").dropna()  # rows = samples, columns = tests
Z = ((X - X.mean()) / X.std()).to_numpy()
U, S, Vt = np.linalg.svd(Z, full_matrices=False)
print(pd.Series(np.linalg.norm(Z - np.outer(U[:, 0] * S[0], Vt[0]), axis=1), index=X.index).sort_values())

The water question

If the outer Solar System built almost none of the Earth, where did the oceans come from? The paper keeps the door open: CI-type additions of about 0.1, 0.3 and 1 percent by mass would supply Earth’s entire silicate budget of nitrogen, carbon and hydrogen respectively. A trace could carry the volatiles, never the walls. ETH’s next question is why the hot inner disc already held enough water.

Extend the array and Mercury and Venus come out “more extreme than for Earth”; sample return from the inner planets, the authors write, is “sorely needed to test these ideas”.

Run the Hack on the last test panel your office accepted and see what sits off the line.