Explainable AI is often discussed as if explanation were enough. For high-stakes quantitative work, explanation is only one layer. A numerical result should remain connected to the evidence that produced it.
Imagine a system reports: Expected value = 41.73. A reviewer should be able to ask which dataset, which observations, which model, which parameters, which software release, what uncertainty, which constraints, which benchmark and who signed the result.
If the system cannot answer those questions, the number may still be useful — but it is not yet evidence-linked.
LargeQuant treats quantitative output as a chain of versioned and cryptographically signed records where appropriate: dataset lineage, execution evidence, artifacts, reproducibility packs, benchmark records and institutional provenance.
A number that matters should not become detached from the conditions that made it true.
This becomes especially important as quantitative systems become more autonomous. The infrastructure has to preserve history as the work happens rather than forcing a reviewer to reconstruct it later.