Interpreting the Nexus

The Nexus does not produce conclusions. It produces structured, traceable and quantified results that the user interprets within their geochemical context. This page explains what these results mean, how to read them, where they are reliable and where they reach their limits.

What the Nexus Produces After Execution

After workflow execution, results are accessible from the analysis card and the side panel. They are organised into three levels.

The predicted output signature

This is the central result. The Nexus calculates the expected δ value at the output of the modelled processes, starting from the initial signature and the fractionation coefficients applied at each step. This predicted value is accompanied by its propagated uncertainty, which cumulates the analytical uncertainty of the source signature and the uncertainties on the ε parameters of each process.

This predicted signature is the value one would expect to measure in the sample if the modelled process scenario is correct. Comparing it with the actually measured signature is the workflow's primary consistency test.

Correspondence results

When a provenance search card is included in the workflow, the Nexus queries the IsoFind database and returns reference samples whose signature is compatible with the predicted signature. Each result is accompanied by an overall score calculated by Mahalanobis distance, which measures isotopic compatibility taking into account the analytical uncertainties on both sides.

Results are ranked by decreasing score and distinguish full matches (all analysed ratios compatible at 2σ), partial matches (fewer than 50% of ratios compatible), and incompatibilities with the detail of non-coinciding ratios. Each result indicates whether the reference sample is a source or a daughter in the database.

ML contextualisation results

The ML analyses panel displays the predicted speciation distribution for each trace element in the workflow, the associated confidence level, and the list of dominant chemical species. These results document the geochemical context in which the fractionation calculations were executed. They are not an interpretive result as such, but information about the robustness of the parameters used in the calculations.

Reading the Predicted Signature

The central question is: is the signature measured in the sample compatible with the signature predicted by the workflow?

If the two values overlap within their respective error bars (at 2σ), the modelled process scenario is consistent with the data. This does not prove it: other scenarios may produce the same predicted signature. Conversely, if the two values do not overlap, the modelled scenario is refuted for this data, and it is necessary either to revise the parameters (ε value, tracked phase, reacted fraction), or to reconsider the process scenario itself.

The role of propagated uncertainty

A large propagated uncertainty does not invalidate a result. It simply reflects that the workflow parameters are poorly constrained, generally because the ε values available in the literature show high dispersion for the modelled process. In this case, the prediction is compatible with a wide range of measured signatures, which reduces the discriminating power of the workflow but does not eliminate it.

To reduce propagated uncertainty, the two available levers are to use better-constrained ε parameters (experimental measurements specific to the system under study rather than generic literature values) and to work on multi-element systems, where the combination of several isotopic ratios reduces the space of compatible solutions.

Reading Correspondence Results

The correspondence score is between 0 and 1. A score above 0.75 indicates that the predicted signature is statistically compatible with the reference sample at 2σ. Below 0.75, the match is rejected. This threshold is conservative: it can be adjusted in the provenance search card parameters according to the required rigour.

It is important to distinguish two cases. When the match is on the predicted output signature of the workflow, it tests whether the transformed signature corresponds to a reference sample. When it is on the initial signature of the workflow, it tests whether the source signature corresponds to a reference origin in the database. These two uses have different geochemical meanings and must be configured explicitly in the workflow.

Strengths of the Nexus

+ Complete traceability of every calculation. Each predicted δ value is decomposable: initial signature, ε coefficient of each process, tracked phase, reacted fraction. There are no hidden parameters. An external geochemist can reproduce the calculation manually with the same inputs.

+ Tracing through transformations. The approach of explicitly modelling successive fractionations allows a chain of processes to be traced back, and tests whether a measured signature is compatible with a given origin, even after natural or industrial transformations that have modified the signature. This is the use case that was previously inaccessible without specialised expertise that was slow to mobilise.

+ Explicit assumptions. Each process scenario is formalised in the workflow. The user can compare multiple scenarios in parallel in separate tabs, and see which produces the predicted signature most compatible with the data. This scenario comparison is structured and documented, not informal.

+ Reproducibility. An exported .isofind file contains the complete workflow with all its parameters. Anyone who reimports it into IsoFind obtains exactly the same calculation result. This is a necessary condition for publication and for inter-laboratory comparison.

+ Accessibility. The Nexus makes accessible to non-specialists an analysis that previously required significant expert work for each study. It remains a tool on which experts can also rely, but it lowers the entry threshold for users less familiar with isotopic fractionation calculations.

Limitations of the Nexus

– Result quality depends on input parameter quality. If the ε values used in the cards are not representative of the system under study (temperature, pH, mineralogy different from the experimental conditions in the literature), predictions will be biased. The fractionation database distributed with IsoFind currently covers Sb, with still incomplete coverage of some process types. For other elements, the user must enter their own reference values.

– The Nexus models discrete processes. In reality, geochemical processes are often continuous, simultaneous or partially overlapping. The sequential card representation is a simplification. Results must be interpreted in light of this simplification, particularly in complex diagenetic or hydrothermal systems where mass balances are difficult to constrain.

– ML contextualisation models are limited to certain trace elements. They cover Sb, Fe, Pb, Zn, Cr, Cd, Cu, Sn, Sr, Se in the current version. For other elements, statistical contextualisation is unavailable and the Nexus will run in simulation mode, without parameter adjustment through speciation context.

– The correspondence database is only as good as its content. Provenance search can only identify what is represented in the reference database. A geographic origin absent from the database will be invisible. The relevance of correspondence results is directly proportional to the density and representativeness of the database used (local, archive, community).

– The Nexus does not test the exhaustiveness of scenarios. It tests the scenario the user has built. The quality of the interpretation therefore also depends on the relevance of the underlying geochemical assumptions. An incorrect but well-parameterised scenario can produce a predicted signature strongly compatible with the data if the processes have compensatory effects.

Best Practices for Interpretation

Building several competing scenarios in separate tabs and comparing their predicted signatures is the most robust way to use the Nexus. A scenario that produces a compatible prediction is not validated if another equally plausible geochemical scenario produces the same prediction.

Working on multi-element systems considerably strengthens discrimination. Agreement on both δ Sb, δ Fe and δ Zn for the same process scenario is far more constraining than agreement on a single ratio. Multi-element workflows fully exploit the power of coupled isotopic data.

Documenting the ε values used and their sources in the card notes is a recommended practice for any workflow intended to support a publication or report. The exported .isofind file preserves these notes with the workflow.