Drag to pan, scroll to zoom. Click the tagged truck to investigate it — most cases are clean, one carrier per case is running a real fraud pattern. Missions / free-roam and day-night-audio polish are not built yet.
A seeded European freight network, running on its own. Trucks move over one real topology, things happen to them, and only what was recorded reaches the panels below.
What this mode is, and what it is not
Live Freight Network
How to read this drawing
Trucks
| ID | Status | Driver | Signals |
|---|
Shift & Oversight Coverage
The port runs a 24h cycle. Traffic volume, which disruptions are plausible, and how much gets recorded all change with the shift. Counts below measure what was observed, not what happened -- a quiet night is a coverage artefact before it is anything else.
Sites & Observation Bias
Where a movement is decides how much of what happens to it gets written down, the same way the shift does. A gatehouse reconciles every movement; a remote depot reconciles in batch the next morning. So the counts below are counts of records, and ranking sites by them puts the best-watched site at the top. Both orderings are shown; neither is a risk ranking, for the reasons in the third card.
Fraud Network View
Entities that have co-occurred in the same case. A connection means "appeared together in a case," not an accusation -- click a node for detail, then "Trace path from here" and a second node to see the shortest chain of shared cases between them. Every entity in a case is linked to every other, so a dense little cluster from one case is guaranteed by the construction and means nothing on its own -- the structure list says which groups actually span several cases. Sites (grey, on the outermost circle) are drawn for orientation and excluded from that analysis: every movement passes through a handful of gates and yards, so a site connects to nearly everything whatever is happening.
Analyst Calibration
A mirror, not a score. Cases you closed are checked against what this simulation actually recorded about them -- not against the real world, and never to tell you that you were right or wrong.
Exposure & Cost Model
Investigative hours are measured by this simulation. The rate that turns them into money, and the value of the cargo, are stated assumptions. Expected loss and loss avoided are refused — the third card explains why, and it is there at the same weight as the other two on purpose.
Portfolio Analytics
Aggregation only — nothing on this panel is measured here, and every figure is reported against the base it was actually computed over. Percentages on a dashboard look alike and are not: some are counts over a sample of closures, and the amber ones are numbers stated in a model file, which cannot move with the data at all. Counts stay visible beside every percentage, and a rate whose base is too small to divide stays a count.
Discovery Lab
A candidate is auto-derived, purely informational: it names a signature moEngine still classifies Potential New MO or Emerging Behavior, sighted often enough to be worth naming. Nothing past that moves on its own -- Start Review, then Validate or Reject, are the only two actions that ever change a record's state, and skipping Start Review is refused.
Candidate JSON downloads are synthetic simulator output, not real incidents or taxonomy entries. Export does not upload or promote anything.
MO Intelligence Center
CATCH THE FRAUD BEFORE THE DEPOT
Shipments roll in from the left. Most are clean. Some are running a real fraud pattern — read the clues on the radio feed, click the truck, then bust it or wave it through before it reaches the gate.