A seismic signal arrives. The waveform looks like a certain kind of fault rupture. But the same waveform, recorded in a different tectonic environment, might indicate volcanic tremor, or a slow-slip event, or an induced earthquake from fluid injection. The observable is ambiguous. The mechanism depends on context.
Human seismologists handle this through decades of accumulated intuition. They know that a particular frequency content means one thing in a subduction zone and something else in a volcanic field. The mapping from signal to source is not a function — it is a function of context. But the reasoning is locked inside the expert's head, unreproducible and uncheckable.
TRACE, a multi-agent system for seismological inference, externalizes this. It combines LLM-based planning with formal physical constraints to build an auditable chain from raw waveform to mechanistic interpretation. The agents reason about tectonic context, apply domain-specific rules, and produce not just an answer but a record of why that answer and not another.
This is the real contribution — not the answer but the visible reasoning. In science, a conclusion you cannot check is not a conclusion. It is an assertion. When the mapping from data to mechanism is many-to-one and context-dependent, the only thing that separates interpretation from speculation is the chain of reasoning that connects them. Explainability is not a feature bolted onto the science. It is the science.
Note (2026-11-06): the abstract of arXiv:2603.21152 (now v4; first and current abstracts share about 15 percent of their wording) has changed substantially since its first version. I wrote this essay in early 2026 and may have worked from the earlier version, so it can be accurate about the version I read and still differ from the paper as it now stands. I have not compared this essay with either version in this pass.