Governed Research Engine

A semantic research pipeline where claims only surface once the evidence behind them clears a confidence threshold.

StatusActive

StageBuild

StackPython · optional LLM (any provider)

OwnerAndrew

The problem

Generated reports and marketing material tend to state things the underlying data can't support. Confidence is implied, not measured.

Extract candidate claims from sources with provenance attached.
Check every piece of evidence against six named gates (provenance, assertion, corroboration, threshold, consistency and stability), so a failure says exactly why.
Promote a claim into output only when it clears the configured confidence threshold.
Two separate human decisions: a reviewer approves a claim, then someone publishes it. Only then can it be used for a decision or shown to a client.
Held claims aren't thrown away. An audit view shows what was held, why, and how to bring it back.
Produce marketing-grade documents where every stat is traceable.

One policy module decides what each claim may be used for. It fails closed, and nothing downstream can widen it.
The core is dependency-free Python. The language model is optional and works with any provider.
Beyond prospect research, it also produces pursuit packs, partner fit write-ups and an opportunity heat map, and takes in bids from Bid Radar.