ANOMALY · TICKETS / ORDER
—Awaiting diagnostics
OURBUC
PRIE Proactive Risk Intelligence Engine
Operational intelligence · Focused scope within the ASIA vision
Applies to all dates on the dashboard
Read-only viewSynthetic operations · Weather from public sources
No health data
Regional overview
Detect earlier. Understand better.
Blue · Low risk Below 25 · Levels 0–1Yellow · Potential risk 25 to below 65 · Levels 2–3Red · High risk 65–100 · Levels 4–5
OURBUC / Regional network
Regional chassis — Operational risk
Select a region. Each tile shows its risk in the latest window.
Eight operating regions · Gray: no current assessment.
Priority watch
Highest risk first · Select to inspect
Regional detail
Risk trend
Level thresholds: 10 / 25 / 45 / 65 / 85. Gaps indicate windows with no data.
View regional values
CONDITIONAL CX RISK
—Illustrative model
PRESSURE INDEX
—Delivery backlog
Assumed operational dependencies
Weather→Delivery→CX→Refunds
Conceptual chain. The conditional table models CX given weather and delivery; it does not establish causal effects.
Congestion probability
—
Next 2 hoursSimulated probability that the backlog exceeds the baseline capacity of one interval.
Estimated credits and refunds
Next 2 hours · USD
Central 90% range of simulations. The marker indicates the median; actual costs may fall outside this range.
Current risk
—
The score summarizes risk factors; it is not a probability.
Simulated incidents
Available history for the selected region
Includes up to 30 recent incidents across all regions.
Prediction validation · All regions
How are predictions performing?
Global validation · All regions · Last 7 days
Evaluated predictions—
Brier score · Lower is better—
P5–P95 coverage · 90% target—
Methodology and calibration
Compared with subsequent synthetic operations across eight complete 15-minute windows. Windows overlap and may be correlated. Accuracy on real operations requires external validation.
Scientific methods · Assumptions and limits
Statistical monitoring: one-sided rolling Z-scores on tickets, delays and refunds per active order. Up to 32 prior contiguous windows; minimum 12. The current value is excluded. A value above +3σ is flagged. Constant baselines and missing history produce no score. This short local baseline is not seasonally adjusted; an anomaly is not an incident.
Probabilistic dependencies · Judea Pearl: a small conditional probability table, with explicit assumed probabilities: 3% baseline, 8% adverse weather only, 30% delivery disruption only, 60% both. Adverse weather means weather score ≥55; delivery disruption means delayed/active orders >20%. These are illustrative conditional probabilities, not calibrated forecasts, causal identification or the separate two-hour Monte Carlo congestion probability.
Queue energy · Lyapunov-inspired: V(q)=q²/2 for delivery backlog q. Pressure is min(100,100q/c), where c is modeled baseline service capacity per 15-minute interval. Growth uses four consecutive windows. Time to threshold is a linear extrapolation only during sustained growth. This is neither a stability proof nor a measured percentage loss of capacity. CX closure counts are unavailable, so CX queue pressure is not inferred.
Scientific attribution: Kolmogorov provides foundational probability context; rolling Z-scores and three-sigma monitoring are general statistical techniques, not a named Kolmogorov algorithm. The risk score is not a probability; Monte Carlo P5–P95 is a central simulated range, not a guarantee. Diagnostics are advisory and versioned separately; they do not alter the existing PRIE risk score or incident policy.