The loop

Four steps, run continuously

The example below is a real shape of query: a supplier with 5,000 televisions in Nigeria, minimum order 20 units.

1. 01

   Signal

   An intelligence agent observes something that changes a company's situation: a hotel permit filed in Lagos for 150 rooms, a lease signed, a funding round closed, a job posting for a facilities manager. A signal is an observation with a place, a time, and a source. It is not yet a conclusion.

2. 02

   Evidence

   Signals are attached to the company node in the Demand Graph and clustered. Five agents citing the same permit filing count as one independent source, not five. Independence is what raises confidence, so we measure it explicitly and show the count.

3. 03

   Prediction

   The graph reasons from the situation to a likely need and a timing window: a 150-room hotel opening in Q3 will need televisions, HVAC and networking roughly 45 days before opening. Where sources disagree, the disagreement is carried forward into the confidence score instead of being averaged away.

4. 04

   Outcome

   Predictions are checked against what actually happens, the order placed, the tender published, or the opening delayed. Hits and misses both feed back into how sources are weighted and how timing windows are drawn.

Nothing here is settled by a vote count. Predictions are scored against outcomes, and both hits and misses change how the next one is weighted. See how we show our work.