A quantitative result should carry its own audit trail.
LargeQuant is designed to keep important numbers attached to the data, model, execution, uncertainty, decision and software state that made them true.
The difference between a result and an assertion is evidence.
A reviewer should be able to move backwards from a published quantitative claim to the records that produced it. Where the platform signs an evidence-bearing record, the signature protects payload integrity and provenance; it does not magically prove scientific correctness.
Dataset lineage
Versioned schemas, sorted canonical rows, source ingestion references and content SHA-256.
Model protocol
The quantitative authority and model configuration used to produce the result.
Execution
Simulation, solver, HPC or laboratory-result ingestion record with artifact references.
Uncertainty
Predictive variance plus named measurement, observation, discrepancy and operational sources where applicable.
Decision
Constraints, required probability, reliability classification and the resulting bounded decision.
Reproducibility
Subject, dataset/evidence, protocols, safe configuration fingerprint, immutable release identity and code-tree hash.
LQBench
Domain-appropriate benchmark evidence without forcing incomparable metrics into a fake universal score.
Institutional provenance
Registered Ed25519 keys, signed external submissions and source-preserving provenance federation.
What a signature can prove — and what it cannot.
A valid signature can prove that a key signed an exact payload and that the payload has not been altered since signing.
Payload integrity, signing-key control and cryptographic linkage.
Scientific correctness, legal institutional identity, experimental validity, plant safety or regulatory approval.