Exploratory Analysis¶

Most analytical work isn't a formal investigation. It's "who are the top operators by incident count?", "what's connected to this entity?", "show me the trend over the last three years." The assistant is connected to every knowledge graph in your deployment and helps you answer these questions through conversation — with tables, charts, maps, and graphs instead of walls of text.

Asking Questions¶

Talk to the assistant in natural language. It figures out the right way to query the data:

"How many incidents involved Boeing 737 aircraft last year?"

The assistant reads the graph schema, writes the query, runs it, and presents the result — usually as an interactive table or chart. You don't need to know Cypher or the graph structure. If your question is ambiguous, the assistant asks for clarification before querying.

What You Can Ask¶

Multiple Knowledge Graphs¶

If your deployment has multiple knowledge graphs, the assistant can query across them. Ask which connectors are available:

"What knowledge graphs are connected?"

Then direct queries to a specific one, or let the assistant figure out which graph to use based on your question.

Seeing Results¶

The assistant never dumps raw data as prose. Structured results come as interactive components:

• Tables for lists and comparisons — sortable, with smart formatting (numbers, dates, badges, percentages)
• Bar charts for categorical comparisons — "incidents by aircraft type"
• Line charts for trends over time — "monthly incident count"
• Maps for geographic data — markers, routes, clusters
• Relationship graphs for network exploration — entities and connections

You can ask for a specific format ("show that as a chart") or let the assistant choose. If you want a visualization the assistant didn't offer, just ask:

"Can you plot that on a map?" "Show me the relationship graph for these entities."

→ Visualizations for the full catalog.

Pivoting and Drilling Down¶

Analytical work is rarely linear. You start with one question, the answer leads to another, and you follow the thread:

You: "Show me incidents involving engine failure in the last 5 years." Assistant: (shows table with 47 incidents)

You: "Break that down by aircraft type." Assistant: (shows bar chart — A320 leads with 18)

You: "What's going on with the A320? Show me the details." Assistant: (shows table of 18 A320 engine-failure incidents)

You: "Are any of these operators also flagged in the sanctions database?" Assistant: (queries the sanctions graph, finds 2 matches, shows them)

Each question builds on the previous context. The assistant remembers what you've been looking at and uses it to interpret your next question. When it queries a different graph or changes approach, it does so transparently.

Working with Hypotheses¶

The assistant can help you test hypotheses against the data:

"I think operators with older fleets have more incidents. Can you check?"

The assistant formulates this as a query (fleet age vs. incident count), runs it, and presents the result — confirming, refuting, or complicating the hypothesis. It won't tell you what to conclude; it shows you the data and lets you decide.

When a hypothesis leads somewhere interesting, you can:

• Save the finding — ask the assistant to write it to a file in the workspace
• File it as a case finding — if you have key questions set up, file it directly into the case layer
• Open a formal investigation — if the thread needs multi-step, auditable reasoning

Saving Your Work¶

Exploratory analysis is ephemeral unless you save it. The assistant can:

• Save a table as a CSV file in the workspace
• Save a map or graph as a reloadable .map or .graph file
• Write a summary to a markdown file
• File findings directly into the case layer

"Save that table as operator-incidents.csv in the Analysis folder."

Anything saved to the workspace is attached to the RFI and persists across sessions.

When to Escalate to a Formal Investigation¶

Exploratory analysis is fast and flexible, but it has limits. Consider creating a briefing and running a formal investigation when:

• The question needs multiple steps — you can't answer it with a single query or a short chain of pivots
• Cross-graph reasoning is central — the answer requires joining entities across different knowledge graphs systematically
• You need an evidence chain — every claim in the answer must trace back to specific graph nodes and tool calls
• The result will be shared or audited — investigations produce a provenance graph that others can browse and verify
• You've hit ambiguity you can't resolve alone — the investigation engine can pause and present structured options at decision points

The escalation is seamless. From the same chat, say:

"I want to investigate this formally."

The assistant starts the briefing flow — selecting a template, walking you through elicitation, and launching the investigation once you approve.

Under the Hood¶

Learn More¶

• The Analyst Workflow — How exploratory analysis fits into the bigger picture
• Briefings — Scoping a question for formal investigation
• Cases — Filing findings from exploratory work into the case layer
• Visualizations — Full catalog of tables, charts, maps, and graphs