What we have put to work.
Kettle has been used with institutional records and paid document-processing work. The verification tools are at an earlier stage.
Any published benchmark needs a collection, date, and method. A number without that context is not useful.
Four things we can point to.
Work inside an archive
The system has handled institutional records, researcher requests, collection restrictions, and human review.
Paid pipeline development
An outside customer has paid Kettle to build document-processing infrastructure for a specific problem.
Three input paths
We process professional scans, phone photos, and existing digital collections.
Claims with citations
The research system keeps the source passage and document reference with extracted claims.
Current, internal, or still to build.
| Capability | Status | Public description |
|---|---|---|
| Document and graph processing | Current | Deployed collection workflows produce clean text, records, citations, and graph data. |
| Research retrieval | Current | Search results carry supporting claims and source references. |
| Deterministic addressing | Internal | Stable object IDs are used inside the research system. |
| Public verification standard | In development | The specification, test vectors, and outside verifier still need to be published. |
| Institution-held source pilot | Next test | Run a pilot where signed IDs and approved data move between systems while the institution keeps the master source files. |
A node count is not proof.
A useful test starts with the source material, shows what the pipeline produced, checks the citations, and states what still failed.
See how we process a collection or send a representative sample.