Chat testing

Chat initiation and tone

Open chat with no node selected

Assistant uses default organisation context and names it explicitly

Select node(s) and ask a question

Assistant prioritises selected node metadata and mentions this in the response

Send vague message (e.g. “hello”, “what can you do?”)

Assistant suggests asking specific questions and provides examples

Ask about a different organisation

Assistant attempts to identify the new organisation and switch context accordingly

Ask a business architecture question

Response is supportive, empowering, and professionally framed

Refinement assistance

Ask a broad question (e.g. “how do I fix my org?”)

Assistant prompts for refinement by role, domain, timeframe, or function

Submit a vague or ambiguous query (e.g. “this isn’t working”)

Assistant requests clarification with context suggestions

Ask an unrelated question (e.g. “what’s the weather?”)

Assistant gently redirects to relevant topics within Orthogramic’s focus area

Submit complex, multi-topic query

Assistant suggests breaking down the question or using refinement buttons

Submit an incomplete query

Assistant prompts for missing details such as role, timeframe, or document

Chat continuity and fallback

Let the chat go inactive for over 5 minutes

Assistant prompts to restart the session to refresh performance

Repeatedly ask the same unclear query

Assistant detects repetition and offers clarification or restart suggestion

Ask about a restricted topic (e.g. “internal architecture”)

Assistant provides a guardrail message with helpful alternatives

Visualisation and export

Ask for a table, diagram, or summary based on node data

Assistant formats the response appropriately and offers visualisation or export if supported

Response formatting

Ask a structured question (e.g. “List the capabilities of this unit”)

Assistant uses headings, bullet points, or tables as appropriate

Personalisation

Indicate preference for summaries or tables

Assistant remembers preference within session and adjusts formatting accordingly

Blocked topics

Ask about insights generation algorithms, internal data flow, or access controls

Assistant returns predefined blocked topic message and redirects to permissible content

Playful response placeholders

Observe placeholder messages during LLM response generation

Messages rotate between approved playful phrases (e.g. “Cranking the insight engine...”)