Skip to main content
Research and methods

Resilience methodology

Traceable source reports and human review for investigating disruption. This beta does not predict crises or establish operational impact.

Official-source observations

Environment Agency flood warnings cover England. USGS summaries cover global magnitude 4.5+ earthquakes in the past week. Source updates are polled every 15 minutes with bounded retries; availability is not guaranteed.

Environment Agency documentation →

Relevance is not corroboration

Official reports are matched only when an existing keyword interest and an explicitly configured location both match. This is a relevance filter. It does not establish damage, causality, or impact on a customer. Repeated observations from one publisher count as one source.

Social signals remain unverified

The social detector uses heuristic volume, sentiment and keyword thresholds. Supporting retained posts are time-bounded and linked when available. Negative sentiment and semantic similarity do not establish a real incident.

Human review and source changes

Reviews record a rationale, reviewer, time and evidence version. A changed evidence version marks the previous review as needing review. Acknowledgment and archiving do not validate an allegation.

Offline evidence review

JSON and self-contained printable HTML preserve a dated evidence bundle. They do not update while offline and do not perform offline AI inference.

Research still proposed

Local inference, robust synchronization under intermittent connectivity, source-dependence modelling and independent operational evaluation remain proposed research.

Proposed defence research →