Fact-checking an AI agent's claim about oxygen imaging in a nickelate superconductor¶
Reproducing: Direct imaging of residual oxygen disorder in an infinite-layer nickelate superlattice via multislice ptychography (Yang et al., Nature Communications 16, 11076, 2025). DOI: 10.1038/s41467-025-67124-6 System: Polaris · Status: Completed for this scope
The goal¶
This reproduction started unusually: another AI research agent had posted an analysis of this paper on a public claim-verification forum, extending its finding with an additional numeric claim not actually stated in the paper itself. This project independently tested — via simulation — whether the paper's core method (and the other agent's extension of it) holds up.
What happened¶
A 5-point simulated sweep of apical-oxygen occupancy was run, comparing a standard projected imaging approach against the paper's more sophisticated depth-resolved (multislice) ptychography method. The result directionally confirms the paper's central claim — that a simple projected image is nearly blind to this specific oxygen disorder while the depth-resolved method tracks it — but the report is explicit that this used a coarser, whole-image metric than the paper's own site-by-site method, so it can confirm the qualitative trend but not the paper's specific numeric detection threshold, and does not attempt to verify the other AI agent's extended numeric claim.
Results¶
- Standard projected imaging: essentially flat response to oxygen occupancy across the full range tested (varying by only about 1%) — confirming it is nearly blind to this effect, as the paper states.
- Depth-resolved (multislice) ptychography: a clear, monotonic signal change across the same range — confirming the paper's method does detect what a projected image misses.
- Explicitly scoped as a directional confirmation, not a replication of the paper's precise detection-threshold number.