37,656 Vessels Have One Flag Assignment Each. None Has Ever Closed. OWNED_BY Records Close at 8.3%. The Flag-State Layer Has No History.

The Setup

The flagged_as temporal edge type carries 37,656 records as of today. Each one links a vessel to a flag state. Panama leads with 4,061 vessels. Liberia follows at 3,815. Marshall Islands at 3,305. Cyprus at 2,208.

None of those records have ever closed.

Every flagged_as edge in the graph has valid_to = NULL. The average record has been open for 38 days. The oldest was written on April 3, 2026. Zero closures in 127 days of graph history. The graph is not frozen — records are being added continuously, with Cyprus and Japan both showing average record ages of just 26 days. The feed is running. It just never closes.

The Chain

The temporal graph uses valid_from / valid_to pairs to represent state changes over time. When a vessel changes ship manager, the old edge closes and a new one opens. The OWNED_BY relation shows how this is supposed to work: 2,854 of its 34,584 records have valid_to set — an 8.3% closure rate. Ownership changes are tracked. The infrastructure knows how to close edges.

It does not close flagged_as records.

There are exactly 37,656 vessels with flag assignments and exactly 37,656 flagged_as records — one per vessel, every one open. That 1:1 ratio is the tell. Reflagging — a vessel moving from one registry to another — would produce a second record linked to the same source entity, with valid_to set on the first. No such pattern exists anywhere in the table. Either no vessel among 37,656 changed its flag between April 3 and August 5 (implausible across a fleet this size), or the data feed writes an initial flag assignment once and does not process subsequent changes.

The mmsi_changed relation offers a contrast. It tracks 864 MMSI identity changes — discrete events the graph captures as new edges with their own timestamps. Flag changes are equally discrete and equally consequential. They are not captured.

The Implication

Flag state is one of the primary dimensions in sanctions screening and risk classification. The top three registries in this dataset — Panama (4,061), Liberia (3,815), and Marshall Islands (3,305) — together represent 11,181 vessels, or 29.7% of the tracked fleet. These are open registries commonly used in flag-of-convenience arrangements. Hong Kong (1,265) and the Bahamas (870) add further exposure.

If a vessel reflagged from a sanctioned jurisdiction to Panama after initial ingestion, there is no record of the change. The vessel appears Panama-flagged with no prior history. The obfuscation_flag field on flagged_as records is 0.0% — the system has marked none as suspicious. That figure cannot be reliable if flag changes are invisible to the layer that would generate it.

Any screening workflow reading flagged_as edges to assess current or historical flag exposure is reading a one-time snapshot. A vessel's entire reflagging trajectory — the sequence of registries used over its operating life — is absent from the graph. The relation type exists. The data structure is correct. The update logic is not running.

What to Watch

• Whether any flagged_as record acquires a valid_to timestamp in the next 30 days. Zero new closures would confirm the feed cannot process flag state transitions.
• The 511 shell_hop edges and 864 mmsi_changed edges against their corresponding vessels' flag records. Vessels that changed MMSI or corporate entity may also have reflagged; the flag layer would not reflect it.
• The intersection between flag state assignments and the obfuscation_flag field. If the field cannot detect reflagging, it is unlikely to detect subtler manipulation.

Limitations

It is structurally possible that the 37,656 vessels tracked have, in fact, maintained stable flag assignments across the April–August window. Fleet-wide reflagging rates vary by segment; a specialized cohort under active sanctions screening might show less flag volatility than the broader market. The zero-closure finding is a necessary but not sufficient indicator of a feed failure — confirmation requires comparing against external flag-state change records for the same vessel set. The SHIP_MANAGER closure rate (4.3%) and OWNED_BY closure rate (8.3%) are the operational benchmarks; zero against those references is the anomaly.

Axiom Overwatch · Temporal Graph · August 7, 2026