How and When We Publish New Predictions
Predictions are not published immediately when a signal appears. There is a deliberate process to ensure quality. Here is how it works.
Stage 1: Signal Detection
Our daily data collection flags any significant change — score crossing a threshold, velocity surge, department composition shift, or new role cluster appearing. This triggers a review.
Stage 2: Pattern Matching
We compare the flagged signal against our 18-month historical database of signal patterns. Does this pattern match a category (product launch, regulatory filing, IPO prep, M&A)? What did similar patterns precede?
Stage 3: Confidence Assessment
We assess signal strength on four dimensions: role specificity (generic vs highly specific), historical precedent (how many prior matches?), signal duration (is it sustained or one-week?), and supporting signals (are there corroborating data points?).
Stage 4: Publication Threshold
We publish predictions at 60%+ confidence. Below 60%, we add to the "watch list" for continued monitoring. The 60% threshold reflects genuine uncertainty — we do not publish predictions we do not believe in.
Stage 5: Monitoring
After publication, we update confidence scores if signal strength changes. Upgrades happen when new corroborating signals appear. Downgrades happen when signals weaken or reverse.