Methodology

OpenWatch ingests from 40+ live sources across news wires, conflict trackers, scientific agencies, regulatory filings, and curated RSS feeds. Collection is staggered by source cadence to balance freshness against rate limits.

• News & events — GDELT 2.0 every 15 minutes; ACLED weekly batch.
• Disasters — USGS earthquake feeds every 5 minutes; WHO and ReliefWeb hourly.
• Regulatory — SEC EDGAR hourly; HKEX, Japan EDINET, and UK Companies House daily.
• Curated RSS — over 1,000 government, NGO, and trusted-publisher feeds in 45+ languages, polled hourly.
• Markets & policy — central bank decisions, sanctions lists, and commodity prices on event or daily cadence.
• Disclosure — STOCK Act congressional trade filings ingested daily.

Every raw item is hashed and deduplicated against historical records before entering the enrichment pipeline.

Each raw signal passes through an LLM-assisted enrichment pipeline that extracts structure from prose:

• Entity extraction— named-entity recognition for companies, people, places, and instruments using a mix of spaCy models and large-language-model extraction for low-resource languages.
• Country geocoding— mentioned places are resolved to ISO-3166-1 alpha-2 country codes via a hybrid gazetteer (GeoNames) plus disambiguation on context.
• Threat-layer classification— each signal is classified into one or more of eight layers: military, cyber, economic, political, disaster, environmental, health, supply_chain.
• Significance scoring— a 0-100 score combining novelty (against the 30-day rolling baseline for the same country/layer), source authority, entity prominence, and language signals indicating escalation or resolution.

The CII is a daily 0-100 score per country summarising how the macro and signal environment is shifting. It is the weighted sum of four components:

• Signal trend (30 pts) — 7-day signal volume and significance versus the trailing 30-day baseline. Positive economic and political signals add points; conflict, disaster, and crisis signals subtract.
• Market gap (25 pts) — the difference between fundamentals (growth, current account, debt service) and current asset prices, sourced from public data and proxies.
• Source quality (25 pts) — how concentrated coverage is, how diverse the sources are, and whether reporting is corroborated across independent feeds.
• Macro tailwind (20 pts) — central-bank stance, commodity exposure, and currency regime relative to the global cycle.

Each country is assigned a posture — opportunity, risk, or neutral — and a recommended action:

• ENTRY — high CII, improving trend; basket-style exposure considered constructive.
• EARLY — CII rising from a low base; thesis forming but unconfirmed.
• WATCH — mixed or transitioning; no decisive bias.
• NEUTRAL — no edge in either direction.
• REDUCE — CII falling, trend deteriorating.
• EXIT — sustained CII decline with confirming signals.
• AVOID — severe risk concentration; not investable on this framework.

Sectors and companies inherit from the surrounding signal environment, then are adjusted for direct mentions:

• Sector score— aggregates threat-layer activity weighted by relevance to that sector (e.g. supply-chain disruptions weigh heavily on industrials, health signals on pharma).
• Company score — 0.8 × sector_score + 0.2 × direct_mention_score. The direct-mention term reflects the volume and significance of signals naming the company (or its known subsidiaries and brands) in the past 30 days.

Scores are intended as environmentalreads — not buy/sell signals. They tell you what the world is saying about a name, not what the price will do.

Source list is updated as feeds are added or deprecated. See /sources for the live catalogue.

• Signal lag. Even at 15-minute polling, primary sources have their own publication delays. OpenWatch is faster than human aggregation, not faster than the underlying event.
• Source bias.News feeds reflect the editorial choices of their publishers. Country coverage is uneven — English-language reporting on small markets is sparse, and we partly compensate with native-language feeds, but gaps remain.
• Model error. Entity extraction and threat classification are probabilistic. Low-resource languages (Urdu, Hindi, Amharic, etc.) have higher error rates; we surface confidence where possible.
• Backtesting caveat. The CII has been backtested on historical signals, but historical performance does not guarantee future returns. Score definitions change as we add data.
• Not financial advice. OpenWatch is an information service. Nothing here is a recommendation to buy, sell, or hold any security. Consult a licensed advisor before making investment decisions.
• Analytical outputs, not recommendations. Signal scores, posture classifications, and investment implications are analytical outputs intended for research and informational entertainment purposes. They represent structured hypothesis generation, not investment recommendations.