Every GNI claim is traceable to a specific pipeline run, specific articles, and a specific AI analysis chain. The Transparency Engine documents every algorithmic decision from 400 raw articles down to the final 3 selected for analysis.
GPVS scores every prediction against real SPY market movement after 3 and 7 days. Sources that led to correct predictions gain higher trust weights via Exponential Moving Average — 1.1x for correct, 0.9x for wrong.
Confidence intervals using t-distribution (t=4.303, n=3, alpha=0.05) provide IEEE-citable uncertainty quantification for every sentiment score. This is Novel Contribution 3 in the academic paper.
Every MAD debate produces a directional prediction (BULLISH or BEARISH) with a specific verify date at 3-day and 7-day horizons. The prediction is tied to exact articles and sources that drove the analysis.
After 3 and 7 days, actual SPY market movement is measured against the prediction. Binary outcome measurement — correct or wrong — eliminates ambiguity and enables statistical analysis.
Sources whose articles led to correct predictions have their trust weight multiplied by 1.1 via EMA. Wrong predictions multiply by 0.9. Weights are bounded between 0.5 (penalised) and 2.0 (highly trusted).
As predictions accumulate, statistical confidence of accuracy claims increases. The GPVS Prediction Scorecard provides real-time evidence for every accuracy claim in the IEEE paper — not theoretical, empirical.
Pattern Intelligence becomes more valuable over time. As GNI accumulates verified predictions, the statistical confidence of its source weights increases and the GPVS Scorecard becomes a genuine empirical accuracy record — not a theoretical claim.
By April 2026, the first GPVS verifications complete. By Q3 2026, Model Learning begins recalibrating from surprise outcomes. By Q4 2026, the Pattern Library identifies historical escalation sequences that predict future geopolitical events with statistically significant accuracy.