A neutron triple-axis spectrometer gets a provenance-keeping assistant: TASCopilot recovers alignment and runs 40 adaptive points on BT-7
Synopsis
A single triple-axis neutron measurement draws on papers, instrument logs, data files and specialist software, and that evidence was previously scattered across separate stages with no traceable link. TASCopilot records sources, processing steps and human review decisions as an inspectable evidence chain, while language models only extract information and explain results; scientific calculations, validation rules and human review decide whether evidence can support a claim. On the BT-7 spectrometer at the NIST Center for Neutron Research it recovered a sample rotation deliberately parked 40° away to within 0.08° of the orientation-matrix prediction, and completed 40 adaptive measurement points with all 40 counted and no refusals.
Interpretation
An experimental judgment on a triple-axis spectrometer can now travel along one inspectable evidence chain, from literature and historical alignment records through numerical fitting to instrument measurement. Such judgments previously relied on evidence scattered across papers, instrument logs, data files and analysis conventions, and a publication date alone could not tie a paper to an acquisition session. A retrospective SrMn2Sb2 analysis connects two papers on a related material and historical alignment records to magnetic-excitation modeling of 20 BT-7 scans, retaining the processing and model assumptions.
The system separates four roles: domain vocabulary, concrete evidence records, bounded reasoning workflows and language generation, so a model's proposed interpretation stays distinct from a tool result or a human decision. Language models take part in extraction and final explanation but do not approve scientific claims or control the instrument. External scientific tools connect through typed request-and-result contracts, including GSAS-II structure-factor calculations, pySpinW spin-wave calculations and a separate instrument-control adapter.
In a live BT-7 experiment the same record structure supported both fine alignment and point-by-point adaptive measurement. The adaptive-measurement interface had previously been exercised only against the NICE simulator. All 40 points were counted at 60 s each in 96 minutes; where the run overlaps the 2017 conventional measurement of the same crystal at 0.10, 0.20 and 0.30 r.l.u., the live intensities sit on the earlier curves with no fitted scale factor.
Perspective
The demonstrated scope is the BT-7 thermal triple-axis spectrometer at the NIST Center for Neutron Research: a retrospective SrMn2Sb2 analysis (experiment 24858, two sessions in 2018) and a live La0.7Sr0.3MnO3 experiment (experiment 202608/14, room temperature, fixed final energy). The fine-alignment workflow is designed for adjustment from an existing gross alignment; states requiring a new gross alignment or interpretation of multiple crystal domains are handed to the experimenter.
Human verdicts cover 334 of 492 extracted facts, and facts carrying a machine citation flag were reviewed first, so the labeled subset is not a random sample; cross-paper relation-edge verdicts come only from a two-paper pilot. The live dispersion fit omits instrumental-resolution averaging, and its parameters and sampling coverage are recorded in the Supporting Information. The retrospective fit gives a reduced chi-square of 1.13 for 775 nominal degrees of freedom across 20 scans, and the anisotropy parameter is more sensitive to analysis choices.