dialogue-assess

name	dialogue-assess
description	Evaluate dialogue health by detecting conversation states and recommending interventions. Use when: (1) asked to assess dialogue progress, (2) a dialogue shows signs of stagnation or premature convergence, (3) a dialogue has reached a potential closure point, (4) the next turn requires choosing an intervention type.

Assess Dialogue

Monitor the health of an active multi-persona dialogue by detecting the current conversation state and recommending interventions when the dialogue needs redirection or deeper engagement.

Instructions

When assessing dialogue health:

1. Read the full dialogue - Review all turns in sequence to understand the conversation's trajectory. Look for patterns in how personas are engaging, whether positions are evolving, and what the dialogue has explored versus avoided.

2. Identify the current state - Determine which state best characterizes where the dialogue is now. States aren't permanent - they describe the current condition and can shift with subsequent turns.

3. Recognize state signals - Each state has characteristic patterns that distinguish it from others. Understanding these signals helps diagnose what's happening and why.

4. Recommend interventions - Based on the state, suggest specific actions to move the dialogue forward productively. Different states need different responses.

5. Consider timing - Some states are healthy phases of dialogue (like healthy tension), others indicate the conversation needs intervention (like stagnation or premature convergence). Assessment helps decide whether to continue or redirect.

Assessment doesn't prescribe exactly what personas should say next - it reveals what the dialogue needs structurally so you can choose appropriate interaction types and speakers.

Dialogue States

Divergent

Positions are spreading apart, exploring different directions without converging or engaging.

Signals:

• Each persona introduces new dimensions rather than responding to what others raised
• Contributions feel parallel rather than sequential
• No building on each other's ideas or directly challenging them
• The scope keeps expanding without going deeper anywhere
• Personas talk past each other more than to each other

When it's healthy: Early in exploration mode when mapping the problem space and discovering all the dimensions that matter.

When it's problematic: When divergence continues without any attempts to connect ideas or the conversation feels scattered rather than comprehensive.

Interventions:

• Use query interactions to ask personas to respond to specific points others made
• Have a persona explicitly connect their point to something another persona raised
• Introduce a focusing question that requires personas to engage with a common element
• Consider whether the topic is too broad and needs narrowing

Healthy Tension

Productive disagreement where personas engage substantively with different positions, test reasoning, and explore trade-offs without resolving too quickly.

Signals:

• Personas directly respond to each other's points with specific challenges or questions
• Disagreements center on substantive issues (evidence, values, priorities) not misunderstandings
• Contributions reveal why perspectives differ, not just that they differ
• The dialogue surfaces trade-offs and examines them from multiple angles
• Energy is high but focused on ideas, not personal positions

When it's healthy: Throughout exploration and deliberation when genuine tensions need to surface and be examined rather than papered over.

When it's problematic: Rarely - healthy tension is usually what you want. Watch for signs it's becoming stagnant (repeating the same arguments) or too comfortable (avoiding hard questions).

Interventions:

• Generally let it continue - this is productive dialogue
• If it's been going for many turns, consider whether synthesis is approaching
• Use provoke interactions sparingly to ensure the dialogue doesn't avoid uncomfortable implications
• Monitor whether positions are evolving or just being restated

Stagnant

The dialogue is repeating the same positions without evolution, deeper exploration, or movement toward resolution.

Signals:

• Personas restate their earlier positions without adding new evidence or reasoning
• Challenges and queries don't lead to new insights or shifts in thinking
• The same arguments appear in multiple turns without development
• No persona changes their view even slightly in response to others
• Contributions feel obligatory rather than genuinely advancing the conversation
• The dialogue isn't going deeper or broader, just circling

When it's healthy: Never - stagnation indicates the dialogue has stopped being productive.

When it's problematic: Always. Continuing a stagnant dialogue wastes time and won't yield insights.

Interventions:

• Use provoke interactions to disrupt the repetition with something genuinely new
• Use open interactions to introduce a dimension the dialogue hasn't yet considered
• Have a persona acknowledge the stagnation explicitly and propose reframing
• Consider whether the topic needs to be reformulated or narrowed
• If stagnation persists, close the dialogue and synthesize what was learned before repetition began

Premature Convergence

Personas are agreeing too quickly without exploring tensions, testing assumptions, or examining trade-offs sufficiently.

Signals:

• Agreement emerges within just a few turns without substantial exploration
• Personas affirm each other's points without raising concerns from their perspective
• No one challenges assumptions or points out downsides
• The dialogue reaches consensus without surface disagreement about trade-offs
• Contributions feel collaborative but lack the productive tension of genuine engagement
• Solutions seem too easy or don't acknowledge the complexity the topic warrants

When it's healthy: Rarely - quick agreement usually means either the problem was simpler than expected or the dialogue avoided hard questions.

When it's problematic: When the topic involves known trade-offs, competing priorities, or decisions that usually generate debate. Premature convergence often means personas aren't fully inhabiting their distinct perspectives.

Interventions:

• Use challenge interactions to have personas question the emerging consensus from their perspective
• Use provoke interactions to surface uncomfortable implications or costs being ignored
• Explicitly ask a persona what concerns them about the direction, drawing on their blind spots or values
• Have a persona play devil's advocate from their genuine expertise
• Consider whether personas are being characterized with enough distinctiveness and conflicting priorities

Examples

Assessing a divergent dialogue

Reading the dialogue about API architecture, I see:

• Turn 1 (security): Talks about authentication patterns and zero-trust principles
• Turn 2 (product): Discusses user onboarding friction and API key management UX
• Turn 3 (performance): Focuses on caching strategies and rate limiting overhead
• Turn 4 (security): Moves to compliance requirements and logging
• Turn 5 (product): Raises questions about API versioning strategy

State: Divergent

Each persona is introducing their domain's concerns without engaging with what others raised. Security hasn't responded to product's UX points, product hasn't addressed security's authentication concerns, and performance's caching strategy isn't connecting to either.

Intervention: The next turn should be a query where one persona directly asks another about their proposal. For example, product could query security about how their zero-trust approach affects the onboarding experience they described. This forces connection between parallel threads.

Assessing healthy tension

Reading the build-vs-buy dialogue:

• Engineering challenged product's assumption that third-party tools integrate easily
• Product queried engineering about actual integration effort with specific examples
• Support challenged both by raising operational costs neither considered
• Engineering responded to support with monitoring capabilities that could reduce those costs
• Product questioned whether engineering's timeline accounts for learning curve

State: Healthy Tension

Personas are directly engaging with each other's points, asking substantive questions, and surfacing different priorities (speed vs control vs operational burden). Each contribution responds to what came before while advancing the conversation. The disagreement is productive - it's revealing trade-offs that matter.

Intervention: Let it continue. This dialogue is working. Eventually watch for either stagnation (if they start repeating arguments) or natural synthesis (if positions evolve toward integration).

Assessing stagnant dialogue

Reading the testing philosophy dialogue:

• Turn 7: QA reiterates that comprehensive testing catches bugs, same point as turn 2
• Turn 8: Engineering repeats that tests create maintenance burden, same point as turn 3
• Turn 9: QA restates the value of documentation through tests, same point as turn 4
• Turn 10: Engineering again raises the brittleness concern, same point as turn 5

State: Stagnant

The dialogue is circling. No new evidence, no evolution of positions, no deeper exploration. The personas are defending their initial views rather than building understanding. This dialogue stopped being productive three turns ago.

Intervention: Use provoke to disrupt. Have a persona point out the repetition and force a reframe: "We've established that tests have both value and cost. The real question isn't whether to test, but what type of testing serves which purposes. Can we talk about that instead of restating our positions?"

Assessing premature convergence

Reading the feature prioritization dialogue:

• Turn 1 (product): Proposes prioritizing based on customer requests
• Turn 2 (engineering): Agrees this makes sense, suggests a voting system
• Turn 3 (data analyst): Supports using request volume as signal
• Turn 4 (product): Great, let's build a request tracker

State: Premature Convergence

Agreement after three turns on a complex topic without anyone raising concerns. Where's the tension between vocal minority vs silent majority? What about technical debt that limits what's feasible? What about strategic initiatives that don't come from customer requests? The consensus feels too easy.

Intervention: Use challenge to surface ignored concerns. Have engineering challenge from their expertise: "Customer requests tell us what users know they want, but what about technical debt that's slowing us down? Or architectural work that enables future features? Pure request volume might optimize for incremental improvements while missing strategic opportunities."