Lab / erate-explorer
erate-explorer
Public E-Rate prices that look odd vs peers, so a person can explain them
2026
It pulls USAC Open Data into DuckDB, scores line items against a stated cohort, then Jev decides whether a human should look. It is not a fraud model. The viewer is local, and there is no LLM.
Why I built it
I watched a video about people finding discrepancies in contracts using public government data, and I wanted to try that on E-Rate. I also wanted a real use case for Jev, and this felt like a decent one to start with.
The question is whether a price still makes sense next to its peers, not who cheated.
How it works
Ingest hits USAC Open Data. Build scores Category 1 line items in DuckDB by purpose, product, speed bin, applicant type, urban or rural, then state, then region. Odd rows go to Jev as a typed packet. Jev returns look, mapping, or leftover-puzzle. That is still not a verdict. The local viewer is the queue.
Current state
Open source at github.com/dezmathio/erate-explorer. Virginia plus neighbors, FY2023 to 2025. You need a TypeSafe key. There is no hosted instance.