System health.

Feed Health

The desk pulls from a diverse set of sources across multiple tiers: wire services, government feeds, social media channels, specialist trackers, and regional outlets. Each source is independently monitored for freshness. A feed is considered active if it returned data within the last five minutes, stale if it last returned data between five and thirty minutes ago, and dead if it has not returned data in more than thirty minutes.

Coverage

The desk maintains continuous monitoring across the following domains: Iran and the Strait of Hormuz, Russia and Ukraine, the Indo-Pacific theatre, NATO and Europe, global energy markets, sanctions regimes, military aviation, and AIS vessel tracking. Coverage spans 130+ sources on a 30-second polling cycle, operating around the clock with no scheduled downtime windows.

Source languages include English, Farsi, Arabic, Russian, Hebrew, Chinese, Japanese, and Korean. Translation for non-English sources adds latency (see Known Limitations below) but does not reduce coverage breadth.

Known Limitations

The desk is transparent about what it cannot do. The following limitations are structural and apply at all times:

• Translation latency on Farsi and Russian sources. Non-English items require translation before keyword matching and confidence scoring can run. This adds between 2 and 8 seconds of processing time per item, depending on length and language pair. During high-volume events, this delay can compound.
• AIS coverage gaps in contested waters. Vessel tracking relies on AIS transponder data. Ships in contested zones (particularly the Strait of Hormuz, the South China Sea, and the Black Sea) frequently disable transponders. The desk cannot track what is not broadcasting.
• 30-second polling, not sub-second. The desk polls sources every 30 seconds. It is not a real-time streaming system. Events that occur between polling cycles are detected on the next cycle, not at the moment they happen. This means worst-case detection latency is 30 seconds plus processing time.
• No classified intelligence. The desk uses only open-source, publicly available information. It does not have access to classified government intelligence, diplomatic cables, or signals intelligence. Its analysis is limited to what is publicly observable.
• No satellite imagery analysis. The desk does not process satellite imagery. It ingests reports from third-party satellite analysts (such as Planet Labs or Maxar via public channels) but does not run its own imagery interpretation pipeline.

What Happens When Feeds Fail

Feed failures are expected. Sources go offline, rate-limit, change their endpoints, or become temporarily unreachable. The desk is built to handle this gracefully through a series of fallback mechanisms.

Stale-while-revalidate cache. When a source fails to respond, the desk continues serving the most recently cached data from that source. Cached items are marked with a staleness indicator so subscribers know the data may not reflect the latest state.

HTTP polling fallback. Sources that support both WebSocket and HTTP connections will fall back to HTTP polling if the WebSocket connection drops. This increases latency but maintains coverage.

Reconnection with exponential backoff. Failed connections are retried automatically. The first retry happens after 1 second, the second after 2 seconds, then 4, 8, 16, and so on up to a maximum interval of 5 minutes. This prevents the system from hammering a struggling endpoint while still recovering quickly from transient failures.

Diagnostic banner after 30 seconds. If a source remains unreachable for more than 30 seconds, the dashboard displays a diagnostic banner identifying the affected source and its last successful contact time. This gives subscribers visibility into which parts of the coverage map may be degraded.