How this platform works
and what you can trust.
Every alert, score, and data point in this platform is labeled with its source, age, and calculation method. This page explains those labels — what they mean, where the data comes from, and where the limitations are.
We do not currently use satellite imagery. We use the Kinetic Disruption Proxy (KDP) and Commodity Price movements as a proxy for geopolitical risk. This is a Decision-Support tool, not a guarantee of future outcomes.
We believe transparency is a commercial requirement, not just an ethical one. A paying customer deserves to know exactly what is live, what is estimated, and what is model-derived.
Section 1
Freshness labels
Every article, alert, and intelligence card is stamped with a freshness label based on when it was published. Live surfaces enforce strict age limits.
≤ 12 hours old
Published within the last 12 hours and verified against at least one live news feed. These items appear at the top of the intelligence rail and trigger the LIVE indicator on the ticker.
12 – 72 hours old
Still within the active monitoring window. Shown on live surfaces with a RECENT badge. May reflect evolving situations rather than breaking news.
3 – 7 days old
Moved to background context. Not shown in active disruption surfaces. Remains accessible for situational awareness of developing or ongoing situations.
7 – 30 days old
Historical background. Useful for understanding the origin of a continuing situation, but not surfaced as current intelligence.
Older than 30 days
Archived. Not shown in any active dashboard surface.
Date missing or unparseable
Date could not be parsed from the source. Item is excluded from all live surfaces until a valid date is confirmed.
Model-derived / synthetic
Value derived from deterministic or probabilistic models (e.g. PCM, Synthetic Intelligence Narrative). Clearly labeled in the disruption panel with source and calibration reference.
Live surface policy: The headline ticker and active intelligence cards enforce a strict 72-hour freshness gate at the backend. Articles that cannot be dated are excluded from live surfaces entirely. If all available articles exceed the window, the ticker shows a degraded-feed notice rather than silently displaying stale content.
Section 2
Data sources & provenance
Where each data type comes from, how it is collected, and what its limitations are.
GDELT 2.0 Document API
LIVEWhat
Sub-hourly global news event tracking across 65+ languages.
How
Polled every 15 minutes via GDELT's public JSON API with geopolitical keywords (Hormuz, Iran, Red Sea, Taiwan, Suez, Arctic). Articles are filtered to ≤72 hours at ingest time.
GDELT can return articles from slightly earlier windows even with a 15-min timespan parameter. We enforce age filtering at the backend before any article reaches the UI.
Reuters / BBC / Al Jazeera / AP RSS
LIVE (fallback)What
Tier-1 news RSS feeds used as fallback when GDELT is rate-limited or unavailable.
How
Fetched in parallel via feedparser. Articles are keyword-filtered to crisis-relevant content, deduplicated by title, and age-gated using the same 72-hour rule as GDELT.
RSS feeds serve their full article backlog — some entries may be older. All articles are age-checked and filtered before display.
yfinance commodity pinger
LIVE (15-min delayed)What
Real-time (15-minute delayed) quotes for Brent Crude (BZ=F), WTI (CL=F), TTF Gas (TTF=F), Copper (HG=F), Gold (GC=F), Wheat (ZW=F).
How
Polled on every ingestion cycle (~15 min). Each commodity is stamped with `fetched_at` and marked `stale: true` on fetch failure.
Prices are 15 minutes delayed during market hours per yfinance terms. Pre/post-market prices may differ from official settlement.
Probabilistic Congestion Model (PCM) v1.2 — AIS-Correlated
OperationalWhat
Vessel density anomalies for Hormuz, Red Sea, Suez Canal, Taiwan Strait, Arctic NSR.
How
Probabilistically generated from current oil/commodity stress (Brent level and 1d volatility). AIS-correlated calibration; live AIS feed integration in roadmap.
Derived from commodity stress. Use alongside Kinetic Disruption Proxy (KDP) and actuary signals for full context.
Supply chain alert scenarios
MODELWhat
Disruption scenarios covering Hormuz, Red Sea, Taiwan Strait, Suez Canal, and Arctic NSR.
How
Schema-driven reference scenarios for alert structure and operational value. Labeled by source in the dashboard.
Not sourced from live feeds. Always labeled with calculation basis; combine with live KDP and commodity data.
AI risk synthesis
MODELWhat
Operational risk summaries in plain English linked to specific geopolitical events.
How
Synthetic Intelligence Narrative [v4.2]. Not investment advice; decision-support only.
Model-derived. All synthesis content is labeled with source and "decision-support only — not investment advice".
Section 3
Calculation basis labels
Every metric shown in the platform carries a calculationBasis field that tells you how the value was derived.
live_feedValue derived directly from a live data feed (GDELT, RSS, yfinance) within the freshness window.
rules_engineComputed by applying deterministic rules to live inputs (e.g. GTI formula applied to Kinetic Disruption Proxy (KDP) scores).
modeled_estimateDerived from a statistical or heuristic model rather than direct observation.
manual_overrideSet manually by an operator or analyst. Treated as authoritative until the next live update.
model_derivedValue from deterministic or synthetic models. Labeled by source; used for schema and reference scenarios.
Section 4
What is live vs. model-derived
A plain-English summary of what is live and what is model-derived as of this release.
- ✓GDELT 2.0 news articles (15-min cycle)
- ✓RSS fallback feeds (Reuters, BBC, Al Jazeera, AP)
- ✓Commodity prices via yfinance (15-min delayed)
- ✓Freshness classification & age filtering
- ✓Global Threat Index (derived from Kinetic Disruption Proxy (KDP) scores)
- —Supply chain alert scenarios (schema-driven)
- —Freight rate table (reference)
- —Region threat scores (reference)
- —Probabilistic Congestion Model (PCM) v1.2 — AIS-Correlated
- —Synthetic Intelligence Narrative [v4.2]
Section 5
Intended use & limitations
Intended use: This platform is designed for operational planning purposes by freight brokers, importers, procurement teams, and supply-chain operators. It is intended to help users get earlier visibility into geopolitical events that may affect shipping lanes, commodity costs, and lead times.
Not investment advice: Nothing on this platform constitutes financial, investment, or trading advice. The AI synthesis and commodity impact estimates are model-derived decision-support and should not be used for financial decisions.
Verify before acting: All alerts and disruption signals should be cross-referenced with your freight broker, logistics partner, or commodity supplier before making material sourcing or routing decisions.
Data sources: Some data types (PCM vessel density, Synthetic Intelligence Narrative, freight rate and region reference tables) are model-derived and clearly labeled. We are transparent about source and calculation basis at every level of the UI and this documentation.
Freshness guarantee: We guarantee that no article older than 72 hours will appear in a surface labeled LIVE or ACTIVE. If our live feeds are unavailable, the UI shows a clear degraded-mode notice rather than silently serving stale content.
Calibration Standards
Formula calibration
All formulas are calibrated against realized 2024–2025 Red Sea cost variance data to ensure surcharge recommendations (0.8%–2.0%) remain within standard contract tolerance.
Surcharge outputs from this platform are calibrated to historical Red Sea disruption cost variance (2024–2025). Recommended surcharge bands (0.8%–2.0%) are designed to fall within typical contract tolerance and dispute-resolution norms. Evidence packages generated from the dashboard carry a Digital Origin Signature (DOS) v1.1 for audit traceability.
Phase 2 Assurance & Controls
Assurance & compliance posture
ChokePoint AI is currently engineering toward SOC 2 Type 1 processing integrity standards. Our DOS v1.1 infrastructure ensures origin authentication and non-repudiation for every generated evidence package.
ChokePoint AI is currently engineering toward SOC 2 Type 1 processing integrity standards. Our Digital Origin Signature (DOS) v1.1 infrastructure ensures origin authentication and non-repudiation for every generated evidence package. Each recommendation carries an audit metadata block (ID, timestamp, auth scope, tamper-evident status) and a confidence score derived from corroborating signals.
Event Confidence Thresholds
To de-bias news loops and prevent “hallucinations of scale,” we apply event confidence thresholds: (1) Single-signal guard — if only the Kinetic Disruption Proxy (KDP) is live, the recommendation is flagged UNSTABLE and corroboration is required. (2) Multi-source weighting — surcharge confidence scales with the number of distinct live sources (KDP, PCM, commodity). (3) Freshness gates — signals older than 1 hour are degraded in the confidence score; beyond 1 hour they do not count toward “corroborating.” (4) Historical Acceptance Benchmark (HAB) — every evidence package states the historical acceptance rate for the recommended surcharge band in similar regional scenarios, so users can assess defensibility before client submission.
Section 6
Signal breakdown
The following table reflects the five platform signals, their current ingestion status, and their contribution to the Sentinel Score.
| Signal | Value (0–10) | Weight | Source | Last Ingest | Status |
|---|---|---|---|---|---|
| Geopolitical Threat | 7.2 | 35% | GDELT | Mar 11, 08:00 | CALIBRATED |
| Commodity Volatility | 6.8 | 25% | yFinance | Mar 11, 08:00 | CALIBRATED |
| Port Congestion Index | 5.1 | 20% | INTERNAL | Mar 11, 08:00 | CALIBRATED |
| Freight Rate Signal | 6.0 | 15% | INTERNAL | Mar 11, 08:00 | CALIBRATED |
| AI Composite | 6.5 | 5% | INTERNAL | Mar 11, 08:00 | CALIBRATED |
| Composite Sentinel Score | 6.47 | ||||
© 2026 ChokePoint AI · For operational planning purposes only