PolyPulse — Bilateral Threat Monitor
v2
Real-time signals from global news and event data, normalized and smoothed.
Legend
Low Moderate Elevated High Critical
Data indexed by H1DR4
Active Monitors
Click any pair to analyze
UNKNOWN
USA / RUS
Loading...
Index:
Click to analyze →
UNKNOWN
RUS / UKR
Loading...
Index:
Click to analyze →
UNKNOWN
USA / CHN
Loading...
Index:
Click to analyze →
UNKNOWN
CHN / TWN
Loading...
Index:
Click to analyze →
UNKNOWN
USA / IRN
Loading...
Index:
Click to analyze →
UNKNOWN
USA / VEN
Loading...
Index:
Click to analyze →
GDELT and GPR methodology
How the data is built, what the index means, and how to use it

What is GDELT?

GDELT is a near‑real‑time database of global events and news. It captures who did what to whom, when, and where, along with tone (sentiment), impact (Goldstein scale), and coverage. We aggregate a curated set of conflict/cooperation‑relevant signals between country pairs on a daily basis to track shifts in geopolitical pressure.

The standardized risk index (GPR‑style)

Our index is a smoothed and standardized signal derived from daily tone and coverage. After applying a 28‑day moving average, we normalize values to a baseline from 2017 onward so that the scale is interpretable across time.

  • 0: Baseline (average conditions since 2017).
  • +1: ~1 standard deviation above baseline (elevated).
  • +2: ~2 standard deviations (unusually high stress).
  • −1: ~1 standard deviation below baseline (quiet).

Why "hot war" pairs don't always score the highest

The scale is relative to each pair's own history. Pairs with sustained high tension establish elevated baselines, so a steady state conflict may read as moderate unless there is a fresh surge relative to recent norms. Conversely, a sudden flare‑up in a typically calmer pair can score higher because it deviates more from that pair's baseline.

How the signal is constructed

  • Aggregate daily events and article coverage relevant to bilateral conflict/cooperation.
  • Compute a raw risk score from tone (absolute sentiment) and coverage (log‑scaled counts).
  • Apply a 28‑day smoothing window to reduce noise.
  • Standardize to a z‑score using a 2017+ baseline.

Intraday "FAST" views are useful for monitoring but are not normalized until daily aggregation completes.

How to use this in practice

  • Watch for inflections — accelerations up or down from the baseline.
  • Contextualize prediction markets and resolution clusters with the timeline.
  • Compare pairs on a common scale, while remembering the pair‑relative nature of scores.
  • Flag outliers (≥ ±2σ) that may indicate unusual levels of risk or de‑escalation.

Caveats & best practices

  • Scores reflect media coverage; limited reporting can mask real‑world shifts.
  • Do not mix normalized outputs from different runs; smoothing depends on full history.
  • Use intraday views for alerts; rely on daily normalization for the standardized index.