Notes from the workbench

Why ShrikeDB?

Three practical arguments about hard queries: match the work to the answer, keep inference attached to evidence, and make source semantics explicit before the first large run.

01

Small answers should not require enormous searches

A fraud-ring count, a GraphRAG shortlist, and an evidence trail can share a pattern while requiring very different work. This is how to tell them apart.

For graph analytics, fraud, and GraphRAG 02

A relationship is a claim, not a line

Turn observations into candidate episodes without converting reporting frequency, public infrastructure, or missing data into false certainty.

For intelligence, movement, and financial analysis 03

No ETL still needs a contract

Direct Parquet queries work when identity, grain, version, and multiplicity remain visible. A worked order example shows where plausible totals go wrong.

For lakehouse and operational data teams