Why citations are the product

Large language models are genuinely good at the expensive parts of digital forensics: recognizing unfamiliar formats, sequencing the right repair steps, and correlating events across dozens of sources.

They are also unacceptable in one specific way: they fabricate. An uncited claim in a forensic report can end a case — or a career.

Siftbench's bet is that both statements are true, and the product is the reconciliation:

1. Deterministic parsers produce ground truth.
2. Citations are mandatory and machine-resolvable.
3. A human is always in the loop.

That is not a marketing slogan. The API rejects findings whose pointers do not resolve. The UI never lets AI prose blend into evidence grids. Acceptance is recorded against a named analyst.

If we get one thing right, it is this: leverage without laundering.