How it works
From the first faint signal
to the confirmed picture.
RiskEvents monitors a curated network of sources across every peril class that matters to specialty insurance, then turns the noise into structured event records and tracks each one through a governed lifecycle. Here is what we are tracking right now — and how it works.
Counts cover the full RiskEvents platform, including Conflict & Political Risk. The Core Market feed shown by default is filtered.
Inside the intelligence
How raw noise becomes intelligence.
An event is the visible tip. Behind each one, the platform runs every incoming item through a multi-stage AI pipeline — and because RiskEvents isn’t tied to any single AI provider, the models and prompts behind it are benchmarked, version-controlled, and evaluation-gated before they are ever used.
Raw source items — arriving around the clock
Watch
Continuously scan a curated network of sources across five classes — official advisories, wire services, trade media, mainstream media, and social/community — plus high-frequency hazard feeds.
Triage for relevance
A fast first pass discards the overwhelming majority of items that aren’t relevant or market-material to specialty insurance, so the deeper work is spent only where it matters.
Read & structure
A deeper pass turns prose into a structured record: the perils and sub-perils involved, the lines of business exposed, the countries affected, named entities, and an impact rating with a written rationale.
Match or create
Decide whether an item is a brand-new event or fresh reporting on one already tracked — attaching it to the right event, merging genuine duplicates, and reconciling reporting into source families.
Corroborate
Actively search for additional independent sources to confirm an emerging event, rather than waiting. Independent corroboration is what lets an event advance.
Locate
Resolve the event’s geographic footprint — a point, a radius, a country, or a hazard zone such as a storm track, a shake-intensity area, or a conflict zone — so exposure can be reasoned about spatially.
Translate to the market
Surface the likely affected Lloyd’s catastrophe codes, risk codes, classes of business, and macro-event context — RiskEvents’ own market assessment, not an official Lloyd’s view.
Assess & re-assess
Judge the likely London-market materiality — affected classes of business, geographic or asset footprint, source quality, and possible insured-loss relevance — and revise it as new evidence arrives.
A structured event record → enters the lifecycle below
The lifecycle
One governed path from Signal to closure.
Once a structured event record exists, RiskEvents tracks it through a governed lifecycle — from a private Signal, to corroborated Developing, to confirmed Active, then Monitoring and Closure — advancing only as the evidence does.
- SignalCount only304
Just detected, not yet corroborated. Held privately.
Advances: 2+ sources agree
- Developing26
Gaining independent corroboration.
Advances: 3+ sources or an authority
- Active6
Confirmed and unfolding now.
Advances: 6h without an update
- Monitoring221
Live, but gone quiet.
Advances: 48h without an update
- Closed153
Resolved, archived for reference.
Live figures are read straight from the database. Signal events stay private; this page shows only the aggregate Signal count, and public event content begins once an event clears the lifecycle’s publication rules. Retractions are handled separately and stay clearly labelled in the public archive.
Our method
Why the numbers mean something.
Volume is easy. Signal is hard. A handful of principles keep the feed honest.
Corroborated, not echoed
One story carried by fifty outlets counts once. We reconcile reporting into source families, so a wave of reprints never masquerades as confirmation.
Confirmed before it climbs
An event advances only as real, independent evidence accumulates, or when a trusted authority confirms it. Social-only signals can never auto-promote on their own.
Tested before it’s trusted
No change to how the platform reads or classifies reaches production until it clears a fixed evaluation battery and human sign-off.
Never locked to one engine
RiskEvents isn’t tied to a single AI provider. Models and prompts are benchmarked, version-controlled, and evaluation-gated before they’re used, and components can change as the platform evolves.
Human-verified where it counts
Reviewers can confirm or correct any classification. Every event is labelled accordingly, and each correction feeds continuous quality monitoring.
Audited end to end
Every change of status is logged with when it moved, why, and on what evidence. The full history behind a closed event stays intact and reviewable.
Coverage
The perils we watch.
Explore the current picture.
Live events, market-filtered and tracked as they develop.