Methodology
Every instrument on this site is shown with its source, its timestamp, and a printed caveat. This page explains each one in more depth — including what it cannot tell you.
Reading principles
- No single number is "the risk." Each tile measures one thing; the picture is the ensemble.
- Deviation matters more than level. We compare readings to their own trailing baselines.
- Quiet is not automatically good. Unusual drops are flagged just like unusual spikes — analysts caution against reading intent into short pauses ↗.
Prediction markets
Aggregated from Polymarket, Kalshi and Manifold. Covered in detail in the markets explainer. Thin markets are filtered by volume/participation thresholds; multi-option markets are expanded per outcome.
Global news attention (GDELT)
The share of worldwide news coverage — monitored across dozens of languages by the GDELT Project ↗ — that matches a fixed set of Taiwan-conflict terms. The terms are kept stable over time, so today's reading is always comparable with yesterday's. Limitation: this measures how loudly the world is talking, which is reactive — attention spikes after events and during hype cycles. It is context, not early warning.
Financial stress (TWSE / TAIFEX)
The TAIEX index level, published by the Taiwan Stock Exchange ↗, and the put/call ratio from the Taiwan Futures Exchange ↗ — how Taiwan's own investors are pricing the future, and how much they are paying to protect against a fall. Limitation: markets price everything at once — earnings, global rates, semiconductor cycles. A red day is usually about money, not missiles.
Travel advisories (US, UK, Germany, Canada)
The current official travel advice for Taiwan from four governments: US ↗, UK ↗, Germany ↗ and Canada ↗ — a considered judgment by services with access to information the public doesn't have. Limitation: each government assesses risk for its own citizens on its own scale — a US "Level 1" and a Canadian "0" are not the same thing, and advisories are reviewed on schedules, not continuously. The date on each advisory row is that government's own last update to its advice page — routine edits (a typhoon notice, reworded safety text) move it too, so it is not the date the risk level changed, and it is never the time this site fetched the page. What matters most is change, which is why level changes will get alerting in a future release.
Wikipedia reading
Combined human pageviews (bots excluded) across a small basket of cross-strait Wikipedia articles in English and Chinese, using Wikimedia's public pageview statistics ↗ — what ordinary readers are looking up, as opposed to what newsrooms are publishing. Limitation: an attention metric like news coverage, but for readers rather than publishers — equally reactive, differently biased. Articles with too little traffic to support a truthful comparison are excluded from the combined figure automatically — the tile states how many articles are actually included.
Cyber & connectivity (Cloudflare Radar)
Internet disruptions observed in Taiwan and Taiwan's share of global application-layer DDoS targeting, from Cloudflare Radar ↗. Limitations: Radar sees the internet through Cloudflare's network — disruptions elsewhere can be missed; attack "targets" are attributed by the attacked customer's billing country; attacker locations are obfuscated by botnets and VPNs.
Data practices
We fetch only public data, respect robots.txt in full, identify our requests where feeds allow custom user agents, and poll no source more than its update cadence warrants. Every tile links to its origin. When a source breaks, the tile says so — we never silently show stale data as fresh.