Evidence, judgment & uncertainty

From signal to assessment

Baseline → change → evidence comparison → assessment → maritime implications → observable indicators. A report is retained as an input. Publication as an intelligence product requires a meaningful maritime consequence and an explicit analytical basis. General political or aviation reporting without such a connection stays in the reporting monitor.

Materiality decisions are MATERIAL, MONITOR ONLY, REQUIRES CORROBORATION or NOT MATERIAL. Current material products are individually researched maritime briefs. Automatically extracted incident notices remain reported source records until individual review. The system does not invent a current assessment for every region.

Research time and publication

New editorial investigations have a minimum five-minute window before approval. Refreshes preserve the open window; a changed source or a new investigation after a review decision starts a new one. Drafts can be saved earlier without publication.

The investigation must document chronology, dated primary follow-up, contrary evidence, comparison with the previous judgment and a maritime synthesis. Approval also requires cited evidence beyond the original source and an evidence-linked inference. These are structural checks, not a guarantee of research quality. The clock does not execute AI work: investigation may take longer, and the owner must examine the sources and reasoning before approving in editorial review. Source collection continues independently.

Source status and confidence are different

HIGH, MODERATE and LOW express confidence in a stated judgment. They do not express severity. Current individually researched briefs use MODERATE confidence because contextual evidence does not independently confirm the incident. Source records display their collection basis; analytical confidence is shown only for reviewed products. A severe reported event can carry low confidence.

Collection scope

Automatic readers, report catalogues, dated editorial selections, sensor datasets and reference links have different capabilities. A configured reader can fail. BBC and NYT each supply more than one channel from the same publisher. Reader counts are not a count of witnesses. Maritime reporting centres can circulate the same originating vessel account.

Regional map counts are source records, not unique incidents, risk rankings or estimates of total traffic. An empty region can mean absent reporting or inadequate coverage. ReCAAP and several official sources currently supply catalogues rather than extracted incident tables. Public rescue releases are selected accounts, not worldwide dispatch data.

Indicators and baselines

AIS counts measure received observations using the source counting method. Completed UTC periods exclude partial days. Previous-period comparisons require complete, comparable histories. The pre-crisis baseline from IMF PortWatch has a different counting basis; only the source's labelled calibration is used for an equivalent comparison. No AIS-dark ratio is calculated without an independent physical-traffic denominator.

Monitoring triggers are collection rules, not threat forecasts. Insurance, sanctions, environmental footprint and military posture have no quantitative score until suitable dated evidence is connected. Missing values are shown as unknown, not zero.

Publication security and visual evidence

Public strategic understanding takes precedence over precise live asset disclosure. Vessel trajectories and crossings are delayed by at least six hours; current coordinates, heading, speed and stationarity times are withheld from public vessel responses. Incident-map positions are reported or approximate historical event areas, not live tracks.

Photographs retain their source and reuse links. Selected-vessel photographs require a verified identity match and reusable licensing; an old photograph is dated and does not show the vessel’s current location or condition. A source photograph can be a file image and is not independent verification of a claim. Static basemaps are geographic context, not a new satellite acquisition. No satellite detections or response locations are fabricated.

Versions and corrections

Source-publication versions are retained in the existing archive. Theatre assessment states are persisted separately when their analytical basis changes, with up to 60 recent versions per theatre. A source refresh alone does not constitute a revised assessment. Historical versions retain their dates and contributing record IDs; earlier states are not reconstructed retrospectively.

Corrections should identify the previous claim, revised claim, supporting source, UTC time and effect on confidence. A source update does not automatically update an older analysis. Where no later vessel-status confirmation is collected, unresolved is not the same as active.