Asia-Pacific
Every day, five kinds of pressure in this region are measured from a public archive of the world's news: geopolitical, energy, climate, economic and digital. Each is scored from 0 to 1 by how unusual it was compared with the rest of the year — so 0.5 is about a typical day and 0.9 is a day busier than nine out of ten.
No AI is involved on this page — it is plain arithmetic over that archive, which is what lets it serve as the answer key that the language models are marked against elsewhere on this site.
What 'unusual' means here, precisely
Each reading is a fuzzy membership in Charles Ragin's sense: a 0-to-1 degree of belonging to the category "a high-pressure day", not a probability and not a percentage. It is computed from three anchors — the 10th, 50th and 90th percentiles of the calibration window ( full_out, crossover, full_in).
Because the anchors are percentiles of this window, the scale is relative to the window and to nothing else. A 0.9 says the day sits in the top tenth of the last year for that vector. It does not say the day was dangerous, and it is not comparable to a 0.9 computed under a different anchor version.
Anchors are versioned and immutable: re-deriving different numbers under an existing label is rejected, so one label can never cover two calibrations. Choosing cut-points after seeing which answer they produce is the standard way to manufacture a finding, and freezing them is the defence.
What this site found
Across 18 days × 5 regions in June–July 2026, 5 frontier language models were handed identical evidence and asked how tense the next 7 days would be. None of them forecast detectably better than chance (0 of 5 cleared the bar). And no combination of the five pressure readings reliably preceded escalation: the best was followed by it 53% of the time, against a bar of 80%.
The readings on this page refresh every day. The model test is a closed pilot — Leaderboard, Model Arena, Evidence — and every page of it says so.
Season 1 · 2026-06-25 → 2026-07-12 · 90 region-days (18 dates × 5 regions) · scored 2026-08-05 → 2026-08-06 · closed pilot, not a live feed
Daily chaos series
The five readings averaged into one line, day by day. Higher means a day that was unusual across several kinds of pressure at once; the dashed line is the average for the whole period, so you can see which stretches ran above their own year.
Why the average is taken over the calibrated scores, not the raw ones
The five raw measurements are not in the same unit. Geopolitical is a severity sum built from conflict events; the other four are per-million article rates. Their mean would be a number in no unit at all, dominated by whichever happens to be numerically largest.
The memberships are all on [0.001, 0.999] by construction, so averaging them is coherent. The line averages 2215 readings.
Dashed cyan line is the period mean. Every day in the range carries all 5 vectors.
Threat vectors · 2026-10-17
most recent date with indicatorsThe five kinds of pressure behind that average, for the most recent day. Each score is a 0-to-1 reading of how unusual today is for this region versus its own recent weeks — not a probability. Every card says it in plain words underneath: a score near 1 just means "about as high as it gets here lately", and by design roughly one region-day in ten lands there. The present / absent tag is only which side of the 0.50 midpoint the day fell on, not an alarm.
above the midpoint for this window
above the midpoint for this window
above the midpoint for this window
above the midpoint for this window
below the midpoint for this window
Anchors · v3
percentile@2025-08-01..2026-07-31The measuring stick. For each vector these are the quiet-tenth, middle and busy-tenth raw values of the calibration window; every 0-to-1 score above is that day's raw value read against these three numbers. They are frozen under a version label, so the same score means the same thing every time you come back.
| Vector | full_out quiet tenth (p10) → scores 0 | crossover midpoint (p50) → scores 0.5 | full_in busy tenth (p90) → scores 1 |
|---|---|---|---|
| climate | 309,516.33 | 374,724.57 | 440,448.6 |
| digital | 178,249.46 | 210,765.07 | 271,358.3 |
| economic | 467,379.59 | 556,012.37 | 662,809.92 |
| energy | 88,833.31 | 132,703.13 | 237,984.78 |
| geopolitical | 56,827.9 | 196,579.9 | 729,241.78 |