STRUCTURED INPUTS
CSV, Parquet, DuckDB and NDJSON are the current priority formats.
+Engine / DataAuthority
DATAAUTHORITY ENGINE: QUERY YOUR DATA WITHOUT LOSING ITS HISTORY.
The DataAuthority Engine reconciles and checks structured data while keeping the provenance, transformations, coverage and limits needed to interpret its results.
The engine is not open for self-service. We currently use it on selected cases and on datasets submitted on request.
01 / What the engine is for
FROM STRUCTURED DATA TO TRACEABLE FACT UNITS.
An atom = a fact + its provenance + its context + its limits. The engine starts where inputs, transformations and outputs can be controlled precisely. In our architecture, an atom is a unit of fact tied to a source, version, context and the conditions that constrain its use. This is our working vocabulary, not an industry standard.
ATOMS
The engine is designed to decompose results into traceable fact units instead of one opaque global verdict.
PROVENANCE + CHECKS
Source, version, transformations and checks remain connected to the data path.
LIMITS
Unknown, conflict, insufficient coverage and unsafe inference remain explicit.
02 / From file to result
STRUCTURE, ATOMIZE, CHECK, THEN CONCLUDE.
A missing observation is not automatically a measured zero. An absent source is not proof of absence. A conflict between two sources is not resolved merely because one value is easier to use.
CSV / ParquetDuckDBNDJSON
↓Atoms + provenance + checks↓Result + evidence + limits03 / Test on request
GIVE US A DATASET. WE WILL SHOW YOU WHAT THE ENGINE CAN ACTUALLY SUPPORT.
The engine is not open for self-service. We currently use it on selected cases and on datasets submitted on request. In practice: a structured sample, a concrete question, and a walkthrough of what the data supports — and what it does not support.
Current boundary: PDFs and documents are considered case by case when they contain native, verifiable text. Arbitrary scanned-document OCR is not presented as a guaranteed capability.