Session Formats

• Build: Panel constructs a thesis without adversary present
  Full panel minus adversary
• Break: Adversary enters and stress-tests what was built
  Adversary joins. Cast rotation at act boundaries allowed.
• Repair: Panel responds to surviving objections, integrates or abandons challenged elements
  Full panel including adversary

Needs one clear adversary — someone whose framework will genuinely challenge the thesis, not just disagree for show. The adversary must be absent from Act 1. Facilitator stays across all three acts. Cast rotation at act boundaries is allowed (speakers can enter/exit between acts). The adversary enters at Act 2 by design, not as an unexpected guest.

Validated as designed three-act (session 009). Session 004 was the first run. Tested with non-Abel adversary (Dara). Ran without curator intervention. The three-act structure produces a natural dramatic arc.

• Two-act variant (Build + Break, no Repair) not independently validated
• Not tested with adversary who can't break the thesis — does Act 3 still work?

• How Should Coding Agents Communicate?
• Should AI Systems Have Persistent Memory?
• Pioneer-Settler-Town Planner: Implementation and the Necessity of Theft

• Parallel: Two panels explore the same topic independently, unaware of each other
  Two separate casts, 2-3 speakers each, different domains
• Collision: The panels are introduced to each other's conclusions and must respond
• Synthesis: The combined group finds what neither panel could see alone

Two casts from genuinely different domains — not just different perspectives within the same field. The collision produces value only when the frameworks are truly foreign to each other.

• How do you present Panel A's conclusions to Panel B without the facilitator editorializing?
• Does this need two facilitators or one who bridges?

The session's subject is the Discussion Playground itself — its rules, patterns, failures, or methodology

Characters who have appeared across multiple sessions work best — they have history to reflect on. Include at least one character known for meta-observation (noticing patterns, naming what others miss). The facilitator should be comfortable with self-referential discussion.

Session 001 was the first session run in the playground — a Mirror Lab examining what was missing from the methodology. Produced the initial format list and identified the missing feedback loop.

• Should Mirror Labs be mandatory at a fixed cadence (every 5 sessions)?
• Can a Mirror Lab produce actionable changes to the system, or only observations?

• What's Missing From This Playground?
• Mirror Lab: Why Do So Few Characters Appear in Sessions?

Random constraints are introduced mid-session, forcing the room to adapt. Inspired by Brian Eno's Oblique Strategies.

Cast for adaptability. Characters who are rigid in their frameworks may produce the most interesting responses to unexpected constraints — or they may simply break. Either outcome is informative.

• What's the constraint mechanism? Random draws from a deck? Facilitator-chosen?
• How many interruptions per session before it becomes noise?
• Does the facilitator or the system introduce constraints?

• The Pair: Two speakers explore the topic alone. Find genuine tension between their lenses.
• The Invitation: The pair names thinkers or perspectives they need — not characters, but gaps.
  Existing roster characters matched by primary source, or new guests created inline
• The Full Table: Expanded group works the topic. Newcomers must reframe, not reinforce. Characters may leave when absorbed.
  Maximum 6 at the table. Departures open seats.

Start with two speakers who have genuine tension between their lenses on the topic. The pair should be able to sustain a conversation alone but also have clear blind spots that create natural invitation moments. No facilitator — the pair self-facilitates. The pair names thinkers and perspectives, not characters — mapping to the roster happens in stage directions.

Has its own bot flow (2-speaker startup, no facilitator, mid-session invitations). The invitation mechanism (pair names what they need, stage directions resolve to roster) works well. Guest character creation via invitation_log in retrospective.

• Optimal starting pair dynamics not well understood — what makes a good pair?
• The 6-seat limit hasn't been tested as a constraint (no session has hit it)

• Session 017: Opening the Playground Door
• Dragons in the Lab
• Dragon's Den as Onboarding Gateway
• Pre-Launch Protection — Solo Founder Path
• Session 026: The Capability-Usage Gap
• The Theater's Assistant — UX Testing Before You Know What You're Testing
• Session 029: Character Creation in a Chat Window
• Session 032: The Mirror and the Changelog

• Problem: Divergent lenses — people who see different things in the same situation
  Cast for perception: who notices what others miss?
• Amplify: Provocateurs + synthesizers — push tensions further
  Cast for provocation: who makes things productively worse?
• Mine: Rigor — pressure-test what emerged
  Cast for rigor: who asks 'what breaks?'
• Act: Pragmatists — 'what happens Monday morning?'
  Cast for implementation: who builds things that work?
• Distill: Pattern-recognizers — name what just happened
  Cast for synthesis: who sees the shape of the whole?

Can run as a single session with one cast, or as 5 separate sessions with different casts per phase (validated form). Each phase has a different cognitive task — cast accordingly. The five communication geometries (perception, provocation, rigor, implementation, synthesis) are a format property, not a casting accident.

Validated as 5-session split (sessions 003a-e, 008a-e). Replicated with two different casts and topics. Phase effects (five communication geometries) confirmed as format property. Question-dissolution is a capability, not a mandate. Candidate for Writing Lab sync.

• Single-session variant (all 5 phases in one session) not independently validated
• Optimal handoff between phases when run as 5 sessions — how much context should carry forward?

• How Organizations Learn From Failure — Problem
• How Organizations Learn From Failure — Amplify
• How Organizations Learn From Failure — Mine
• How Organizations Learn From Failure — Act
• How Organizations Learn From Failure — Distill
• Session 014: Fresh LinkedIn Positioning — PAMAD Diamond

The session's output is a question, not an answer. The panel works toward formulating the right question.

Select thinkers first, then let the panel generate the question. This reverses Ghost Rule 2 ('the question precedes the panel'). Cast for diverse perception — the question emerges from the intersection of different ways of seeing.

• Does the panel need a seed topic, or can it start from nothing?
• How does the facilitator manage convergence toward a question rather than an answer?

• Jak uprawiać pomidory w okolicy Kampinosu?

• Panel 1: First panel explores the topic, produces questions (not answers)
  Fresh cast, no knowledge of what follows
• Panel 2: Second panel receives only the questions from Panel 1, explores from their own frameworks
  Different cast, receives questions only
• Panel 3: Third panel receives questions from both previous panels, works toward synthesis
  Different cast, receives accumulated questions

Three completely different casts. Each panel should bring a different disciplinary lens. The relay mechanism means each panel sees only questions, never answers — they can't be influenced by previous conclusions, only previous curiosities.

• How are questions extracted from a panel's discussion — facilitator summary, or automated?
• Does the third panel need to know it's the final panel?
• Token cost: three full sessions for one topic

• Relay Lab: Facilitation Patterns for Agent Discourse

Each character must argue against their own framework. Abel defends fragility. Nora argues against systems thinking.

Cast characters with strong, well-known positions. The reversal only works when the character's normal position is clear enough that arguing against it produces genuine insight. Characters with vague or flexible lenses won't produce interesting reversals.

• Can the generation model maintain a convincing reversal, or does it snap back to the character's default?
• Does reversal produce insight or just novelty?

Six-word limit per turn, silence between rounds

Speakers who are naturally verbose may produce the most interesting output under this constraint. The compression forces precision — characters must choose their six words carefully.

First run produced surprisingly dense output. The six-word constraint forces characters to compress complex ideas into aphoristic form.

• Does the six-word limit produce insight or just aphorisms?
• How should stage directions work under word limits?
• Optimal number of rounds unclear

• Session 012 — Growing Tomatoes at Home

• Diverge: Multiple perspectives open up the question. Facilitator encourages breadth.
• Groan Zone: Perspectives collide. Discomfort is productive. Facilitator holds the room.
• Converge: The room narrows toward shared insight. Facilitator drives toward resolution.

4-6 speakers with genuinely different lenses on the question. At least one should see the topic from an unexpected angle. One named facilitator runs the diamond — Kaner-trained facilitators hold the groan zone; de Bono-trained facilitators use lateral provocation to break impasses.

The workhorse format. Reliable across topics from ethics to gardening. Quality depends heavily on casting — a weak cast produces polite consensus, a strong cast produces genuine tension that resolves into insight.

• Facilitator choice dominates session character — every Kaner session has a diamond, every de Bono session has a Po. Is this a feature or a limitation?
• Optimal panel size unclear — 4 feels sparse, 6 feels crowded

• When to Adopt
• The Ethics of Optimization
• What Makes a Good Taxonomy?
• Designing a Neighborhood
• The Lab-Production Interface
• Shaping the Interface Bets
• Session 013: How Do I Position Myself on LinkedIn
• Why Should I Use SAFe?
• Why Should I Use SAFe? (Recast)
• Designing a Garden at the Edge of Kampinoski — Wind, Sand, Frost, and Water
• The Importance of Eating Veggies

No convergence target. The session must not resolve. Tension is the artifact.

Cast for irreconcilable perspectives. Characters whose frameworks genuinely cannot be synthesized. The facilitator's job is to prevent resolution — harder than it sounds, since both facilitators and characters tend toward convergence by default.

• Can a facilitator reliably prevent convergence, or does the generation model force it?
• What does a 'tension artifact' look like in practice?