Methodology

From source to signal

Every published statement traces back through a fixed pipeline: source fetch → immutable raw snapshot → parsing and normalisation → deduplication → entity and geography resolution → claim extraction → corroboration search → confidence and severity calculation → human review (or narrowly allowed auto-publication) → publication. Each raw document keeps its original URL, fetch time, content hash and parser version so every claim is auditable back to an exact source fragment.

A signal is the smallest publishable unit of change — a new official marine warning, a material change in ice concentration along a route segment, a licence decision, a sanctions designation, an announced exercise — not a copy of an article. Several articles repeating the same agency release count as one source, not several independent confirmations.

Source quality vs claim truth

We score source reliability and claim confidence separately. An official source is the best evidence of what that institution announced — that does not automatically make every interpretation an independently verified fact. State claims are labelled as claims of that actor. We distinguish announced, observed, assessed and verified, and a Disputed flag exists independently of the numeric score. See how confidence works.

Route Access Pressure

Route Access Pressure is computed per route segment from ice concentration and local anomaly, official ice and marine warnings, navigation restrictions, known infrastructure constraints and data freshness. Labels are Favourable conditions, Constrained, Severely constrained and Insufficient data. When key inputs are missing the label falls to Insufficient data — it is never artificially averaged. We never derive route availability from total Arctic sea ice extent alone, and we do not publish a simple Open/Closed status unless a competent authority has done so.

Route Access Pressure is not a safety-of-navigation product. It must not be used for navigation or voyage planning.

Security coverage limits

Security posture coverage uses exclusively public, legal sources: announced exercises, strategies, budgets, publicly described infrastructure and confirmed incidents. We do not show precise positions of active military units, do not aggregate data beyond the level of official publications, apply regional aggregation and time delay, and never publish personal data of personnel. An announced exercise is not treated as evidence of escalation.

The role of AI

Models help classify documents, extract claims, propose summaries and draft briefings. They cannot publish a serious security signal from a single source, change the meaning of a quoted document, invent facts, treat a machine translation as a primary source, assign a precise location without evidence, flip disputed to verified, or delete corrections. Model-generated fields are recorded with the model, prompt version and generation time, and high-risk categories always require human review before publication.

What we are not

• Not a navigational tool — no safe-passage advice.
• Not a military terminal — no real-time unit tracking.
• Not investment advice and no legal guarantee of sanctions status.
• Not an activist or general climate site.