본문으로 건너뛰기
D.Hub
KOEN
Defense
90 min

Turn natural-language commands into safe chart control and audit history

Act as a fictional maritime operator and complete a flow that executes display commands safely without requiring zone IDs or coordinates while recording why requests were allowed or blocked.

0/7 chapters complete

An operator who does not know the coordinates

A request arrives in a fictional maritime operations room: “Show Busan Port Entry/Exit Control Zone.” The operator knows the zone name, but not its internal ID, latitude, longitude, or recommended scale. Sending natural language straight to the chart could also execute a misclassified query or configuration change.

In this Workshop, act as the maritime operator and answer one question.

How can an operator execute a natural-language command safely without knowing a zone ID or coordinates, while preserving the reasons and history for both allowed and blocked requests?

Fictional maritime operations-room illustration where a zone display command reaches map coordinates, a data query stops at a safety gate, and both requests enter an audit history
Fictional scenario illustration — an AI-generated depiction of allowed display control and a blocked query, not an actual product screen.

Decision criteria for this mission

  • Perspectives in play — Operations wants to display a zone quickly from its name. Safety and audit must stop requests outside the allowed scope while preserving the reason for every allowed or blocked path.
  • Decision constraints — The current Agent executes only UI_CONTROL and blocks DATA_QUERY. Eight batch rows and Agent runtime events follow different execution contracts and must not be interpreted as one source.
  • Completion signals — Turn the Busan request into TZ-008 WGS84 coordinate JSON, stop the data query before execution, and find both outcomes again in the audit history for the same user.

Workshop goals

  • Explain how a zone name in natural language becomes an internal ID and WGS84 coordinates.
  • Separate display-control commands from operational-data questions before execution.
  • Turn an allowed command into structured JSON and record a blocked request as an audit event.
  • Distinguish reproducible batch records from Agent runtime events.
  • Explain both the evidence and current linking limits across the dataset, graph, and dashboard.

Seven decisions

Loading the diagram. Mermaid source:

flowchart LR
    accTitle: Seven decisions in natural-language chart control
    accDescr: Frame the operator request, connect a zone to coordinates, run allowed and blocked paths, trace batch and runtime history, check the control boundary, and explain the safe loop.
    request["1. Frame request"] --> zone["2. Link zone and coordinates"]
    zone --> gate["3. Decide allowed scope"]
    gate --> run["4. Run both paths"]
    run --> trace["5. Trace two histories"]
    trace --> check["6. Check the boundary"]
    check --> explain["7. Explain control loop"]

Every chapter follows question → reason → action → observation → interpretation → next decision. The full asset inventory and implementation details stay under Deep dive, so you can follow the core route first.

Prerequisites

  • A D.Hub Portal analyst or engineer account with Editor permissions or higher
  • About 37 KB of download space for one scenario ZIP
  • 90 minutes

You do not need a terminal, Python, or a clone of dhub2-examples. If scenario import is new to you, start with Import and tour a complete hands-on scenario.

Learning journey

  1. Start with the request from an operator who does not know the coordinates15 minImport the Map Control scenario and define the evidence needed to handle one natural-language display request.
  2. Connect a zone name to map coordinates15 minInspect eight tactical zones and run the batch pipeline that produces WGS84 coordinates and scale.
  3. Separate display commands from data questions15 minCompare two real requests and understand the allow-list gate that executes only UI_CONTROL.
  4. Run the allowed command and blocked question in one session20 minStart the event pipeline and compare classification, coordinate JSON, and audit publication across both paths.
  5. Trace batch reference rows and runtime events separately15 minSeparate two execution contracts in event history and inspect the scope and current limits of the ontology graph.
  6. Check the safe-control boundary5 minBefore the final explanation, review zone resolution, the intent gate, audit events, and the graph boundary in four questions.
  7. Explain the safe natural-language control loop and finish the Workshop5 minConfirm allowed and blocked history on the dashboard, then connect the original request to its audit evidence.