Home › Methodology

Methodology

World Tension Meter is an instrument, not an opinion. Every number on this site is produced by an automated engine on a fixed schedule — full analysis twice a day, market data every ten minutes — from named sources, with every calculation described on this page.

The media signal: coverage share, not keyword counting

The media factor follows the approach of the academic Geopolitical Risk (GPR) index used by central-bank researchers: measure what share of the world's news coverage is tension-related, then rank today against that query's own five-year history. A war reads as elevated because coverage genuinely is elevated versus calm years — not because a keyword list says so. Baselines refresh weekly and the ranking is soft-bounded, so the score can never mathematically pin to 0 or 100.

Five independent factors

Conflict — global conflict-coverage intensity (GDELT, 5-year baseline), event-level conflict data where available, and the share of conflict stories across world-news feeds.
Energy — Brent level, natural gas, oil volatility and the gold/oil ratio, each ranked against multi-year history.
Trade — real trade data: the Baltic Dry Index, chokepoint and tariff coverage share, and food-commodity stress.
Finance — VIX, the 10-year yield, dollar-index deviation and gold.
Media — the GPR coverage-share signal described above, plus headline fear density and coverage tone.

Weights: Conflict 25% · Energy 20% · Trade 20% · Finance 20% · Media 15%. The factors are computed from separate inputs, so one noisy source cannot move the whole index.

Fear physics

Real fear is asymmetric, and so is the index. When the composite rises, the score jumps the same run — 85% of the gap immediately. When it falls, the score can bleed at most three points per day. A major attack therefore spikes the meter in hours and holds for days, exactly as markets behave. On top of this, detected shock events (invasions, nuclear threats, market crashes) add a temporary overlay capped at +18 points that decays with a per-event half-life. Question headlines ("Will X happen?") are excluded from shock detection.

Honesty rules

If a data source fails, the engine reuses the last known good value and flags it "estimated" on the page. Nothing is ever fabricated.
Every daily score is appended to a permanent, append-only ledger and never edited afterward. The Backtesting page charts that raw record.
The status chip in the header turns to Delayed automatically if data stops flowing, instead of quietly showing stale numbers.

Sub-indexes

Food Security, Currency Stress, Shipping Stress, Sanctions Pressure, Energy Security, Gold & Silver Fear and Crypto Fear & Greed each follow the same principles — real market data, dynamic multi-year calibration, coverage-share news signals and estimated-data flags — with their exact recipes documented on their own pages.

Sources: GDELT Project (global news monitoring) · market data via public exchange feeds · event-level conflict records where available · major world and business news feeds. Updated twice daily (full) and every 10 minutes (markets).

Download the Methodology Whitepaper v3 (PDF) — this page in citable form, including the baseline-window research history. The full dataset behind this site is public and free — API documentation. The original v1 whitepaper (first-generation engine) remains archived here.