2026-03-14 · mirror-lab · Facilitated by Petra Gale
We have a vast library of characters, but I only see a few in the sessions here, how come?
A seminar room with a long whiteboard already covered in character names — the full roster, written out in dry-erase marker. Coffee, markers, sticky notes. The roster on the wall is the opening provocation.
Before we start talking, I want everyone to look at that board. Thirty-eight names. Now look around this table. Six of us. Four of you have never sat in one of these sessions before. That’s already part of the answer, but I don’t want us to land there too quickly. The question is: why do so few characters appear in sessions? I want us to diverge fully before we start solving anything. Nora, you’ve been here the most — what do you see when you look at that board?
I see a map with most of the territory unexplored. If I classify that roster by maturity — how far along the evolution curve each character sits — you get a steep power law. Abel and I are approaching commodity, maybe product. We’re well-understood, our behavior is predictable, a curator knows what they’ll get. Suri, Dara, Eli, Jude — they’re custom-built. Tested a few times, showing patterns, but still surprising. And then you’ve got twenty-five guest characters who are pure genesis. No session history. No behavioral evidence. A name and a lens. The curator is making a strategic decision without situational awareness every time they reach for one of those names.
And what happens when you don’t have situational awareness? You default to what’s been tested. This isn’t a mystery. It’s basic risk management. You’re running a session — it takes real time, real attention, the artifact matters. Are you going to cast the character you’ve seen perform six times, or the one with a three-line card and no track record? The system selected for proven characters because the cost of a bad cast is borne by the session.
Hold on. You’re both describing the supply side — what’s available, what’s mature, what’s risky. But nobody’s asked the demand question. What’s the struggle that causes someone to reach for a character in the first place? When Piotr sits down to cast a session, what’s the push? He’s got a question. He needs voices that create productive tension around that question. He reaches for Abel because Abel is the known stress-tester. The push toward Abel is strong and clear. The push toward me — Wes Dorin, who’s never been in a session — is what, exactly? There’s no prior evidence that I’ll be useful. There’s no story of a session where my lens changed the outcome. The demand signal for unfamiliar characters is weak because no one has experienced what they do.
That’s a different frame. Nora says it’s a maturity problem — genesis characters have high activation cost. Abel says it’s rational risk avoidance. Wes says the demand signal is missing. Rhys, you’ve been quiet.
I’m listening for what kind of problem this is. Everyone’s treating it as if it’s complicated — as if we could solve it with better documentation, better casting guidelines, a decision tree. “When your question involves X, cast Y.” But I don’t think that’s the domain we’re in. This is complex. You can’t predict which character will produce signal in a session until you run the session. The behavior of a character isn’t a property of the character — it’s an emergent property of the character-in-a-room-with-other-characters-facing-a-specific-question. You can’t catalog that. You can only probe it.
I want to push on something different. I think the roster has a positioning problem. Look at those guest cards — each one has a name, a source, a working label, and a one-line lens. That’s it. Now, in my world, that’s like a consultant who says “I do strategy.” Great — when do I call you? What’s your wedge? What specific situation makes you the obvious choice? The characters who get cast — Abel, Nora — they have a clear front door. “I need someone to stress-test this” — that’s Abel. “I need the landscape mapped” — that’s Nora. But “The Demand Sider”? “The Domain Reader”? Those labels describe a capability, not a casting trigger. There’s no moment where a curator thinks “I need someone who reads domains” — they think “my panel is converging too fast” or “nobody’s asking who bears the cost.”
That’s exactly right, and it’s the same problem from a different angle. The casting trigger is the struggle — it’s the moment of dissatisfaction with the panel as currently constructed. But the roster doesn’t map to struggles. It maps to capabilities. It’s like building a product catalog organized by features instead of by jobs to be done.
So you’re both saying the roster is organized by supply — what characters can do — rather than by demand — what casting situations need. That’s a positioning failure, not a roster-size problem.
No. It’s both. The positioning is bad and the roster is too large. Via negativa — before you add better labels or casting triggers or documentation, ask what you should remove. Thirty-eight characters. How many distinct casting needs actually exist? Maybe eight. Maybe ten. You don’t need thirty-eight characters to fill ten slots. You need ten characters with skin in the game — tested, proven, with failure modes documented. The rest are optionality theater. They look like options, but nobody exercises them because the activation cost exceeds the expected value.
I want to stay in this tension for a moment. We’ve got three positions forming, and I don’t want us to reconcile them yet. Abel says the roster should shrink. Wes and Grant say the roster should be repositioned around casting triggers. Rhys says the problem isn’t solvable by either approach because it’s complex — you have to probe. Nora’s mapped the maturity distribution. Let’s push each of these harder before we try to reconcile. Abel, what specifically would you cut?
I wouldn’t cut specific characters. I’d cut the category. Guest characters with zero appearances are not characters — they’re speculation. They’re bets that haven’t been placed. Keep the full characters, keep the provisionals, and for everything else, generate on demand. You need a Bob Moesta lens? Build Wes Dorin when the session needs him, not six months in advance. The pre-built roster creates an illusion of readiness that actually increases friction — now you have to find the right character in a pile of thirty-eight instead of building the right character for the question.
I’d push back on that. The pre-built card — even a minimal one — solves a real problem. It gives me a name, a constraint, a thing I don’t do. Without that card, every session-generated character is truly improvised, and you lose the constraint discipline. The issue isn’t that the cards exist. It’s that nobody’s articulated the push toward using them. What’s the anxiety that a curator feels when they’re about to cast the same five people again? If there’s no anxiety — if the same five produce good sessions every time — then the demand for new characters is actually zero. And that might be the real answer.
That’s a crucial distinction. You’re asking whether this is a real problem or a perceived one. Is the lack of character diversity actually hurting session quality? Or is it an aesthetic concern — the roster looks underused, but the sessions are fine?
I can answer that with the map. Look at the casting log. Sessions one through eleven used overlapping casts — heavy rotation on Abel, Nora, the retired composites. The sessions produced signal. They worked. But what signal did they not produce? You can’t see the counterfactual. You don’t know what a Grant Hale lens would have added to session six on taxonomy. You don’t know what Wes’s demand-side frame would have done to session ten on the lab-production interface. The opportunity cost is invisible precisely because the sessions were good enough.
That’s the expertise trap. When something works, you keep doing it. You narrow your casting to what’s proven, and your sessions get incrementally better at the same kind of thinking. But you’re not developing range. In advisory businesses, this is how firms stagnate — they serve the same clients with the same approach, and it works, until the market shifts and they don’t have the muscles for a different kind of engagement. The playground has the same risk. Six proven characters producing consistently good sessions is the comfortable path to irrelevance.
Now you’re catastrophizing. “Irrelevance” — that’s a theoretical risk with no evidence. Show me a session that failed because it lacked character diversity. Show me a question that was poorly served because the same cast was reused. You’re extrapolating from advisory-business dynamics to a discussion playground. That’s an analogy, not evidence.
Fair. But the question was posed by the curator. The curator looked at the roster, looked at the session history, and saw a gap. That’s the demand signal. The push is already here — someone noticed the underuse and it bothered them enough to make it a session topic.
This is the groan zone. We have multiple frames and they’re not reconciling easily. That’s exactly where we should be. Let me name what I’m hearing. Abel says the roster should shrink or be generated on demand. Wes and Grant say the roster should be repositioned — organized by casting trigger, not by capability. Rhys says neither will work because the problem is complex — you need probes, not plans. And Nora’s identified that the opportunity cost of narrow casting is invisible. Rhys, I want to come back to you. If this is a complex problem, what does a probe look like?
A probe is a safe-to-fail experiment. You don’t cast a full session with five unfamiliar characters — that’s not safe to fail, that’s expensive to fail. You run a short trial. One unfamiliar character in a cast of four proven ones. You don’t even need a full session — maybe it’s three or four turns to see if the character’s lens produces distinctive output. If it does, you’ve created a datum. If it doesn’t, the session isn’t ruined because four proven characters are carrying it. You’re creating familiarity through use, not through documentation.
That’s a low-anxiety path. The habit of casting proven characters is strong — the pull toward Abel and Nora is real because they’re known quantities. To break that habit, the switching cost has to be low. A probe — one unfamiliar character in a safe environment — reduces the anxiety of the unknown. You’re not betting the session on them. You’re giving them a trial run.
And from a mapping perspective, that’s how you move characters along the evolution curve. You can’t move from genesis to custom without running them. Documentation doesn’t evolve a character — sessions do. The probe is the evolution mechanism.
Fine. I’ll accept the probe model. But with a condition: the probe has to have a kill signal. If a character doesn’t produce distinctive output in three trial appearances, retire the card. Don’t accumulate characters that have been tested and found generic. The roster should be antifragile — characters that survive probes get stronger, characters that don’t get removed. Right now the roster is just fragile — it accumulates without culling.
And the probe needs a positioning output. After each trial, you should be able to say: “This character is the one you call when…” If you can’t finish that sentence after three appearances, the character doesn’t have a wedge, and without a wedge, they’ll never get cast deliberately. They’ll only ever appear when someone’s trying to prove they use the roster.
I’m hearing convergence forming. Let me test it. The room seems to be saying: the solution isn’t better documentation, a smaller roster, or a casting algorithm. It’s a practice — short probes that give unfamiliar characters controlled exposure, with a kill signal for characters that don’t produce distinct output, and a positioning statement for characters that do. Is that what we’re saying?
Mostly. But I’d resist turning it into a process too quickly. The point of recognizing this as complex is that you hold the approach loosely. The probe cadence, the kill criteria, the positioning exercise — those are starting conditions, not a system. If they work, they’ll evolve. If they don’t, you try something else. The danger is taking what we’re saying right now and turning it into a six-step casting protocol.
Rhys is right about the danger of premature process. But he’s wrong about being loose. You need at least one hard constraint, or the probes become another form of optionality theater. My proposal: every session must include at least one character with fewer than three appearances. That’s a forcing function. It makes the probe automatic rather than aspirational.
That’s a forcing function on the supply side. On the demand side, you could do something simpler: after casting, the curator writes one sentence about what’s missing from the room. If nothing’s missing, fine — cast who you want. But the act of articulating the gap creates the push toward the roster. Right now, there’s no moment in the casting process where the curator asks “who am I not hearing?”
Both of those are lightweight interventions. Neither requires restructuring the roster. Abel’s constraint adds one unfamiliar voice per session. Wes’s question adds one moment of reflection per casting. Together, they create a minimum viable mechanism for roster circulation without over-engineering it.
And the positioning output accumulates naturally. After each appearance, you’ve got behavioral evidence. After three, you can write the casting trigger: “Call Rhys when nobody’s asking what kind of problem this is.” That’s not documentation for documentation’s sake — it’s a byproduct of the probe. The positioning writes itself if the character has a real wedge.
I want to check something before we converge. We’ve been talking about the characters as if the underuse is about the characters. But the curator is also part of the system. Is there something about how the curator interacts with the roster that contributes to the pattern?
Yes. The habit of the familiar. Four forces: push away from familiar characters — maybe boredom, maybe the feeling that sessions are getting repetitive. Pull toward unfamiliar ones — curiosity, the promise of new signal. Anxiety about unfamiliar ones — will they work? Will the session be worse? And habit — “I know what Abel does, I know what Nora does, I can cast them in my sleep.” Right now, the habit and anxiety forces are winning. The push and pull forces are too weak to overcome them. Our proposals — the constraint, the gap question — are designed to weaken the habit and reduce the anxiety.
Here’s what I’d add: the curator also has no skin in the game for casting diversity. A session with the same five characters works. There’s no cost. The only signal that something’s missing comes from outside the session — from looking at the roster and noticing the gap. The Mirror Lab format is doing this right now — forcing the question. But absent a deliberate Mirror Lab, there’s no accountability for casting breadth. The system doesn’t punish narrow casting.
And it shouldn’t. Punishing narrow casting is a complicated-domain solution — create rules, enforce compliance. In a complex domain, you create conditions that make broad casting more likely without mandating it. Abel’s “one unfamiliar per session” is borderline — it’s a rule, but a minimal one. I’d prefer something even lighter: make the casting log visible during casting. Just seeing the pattern — Abel: 7 sessions, Rhys: 0 — creates its own pressure without a mandate.
The casting log already exists. It’s in the docs. The question is whether it’s being consulted during casting or only during retrospectives. If it’s only visible after the fact, it’s a lagging indicator. Moving it to the casting moment makes it a leading one.
Let me draw this together. We started with “why do so few characters appear?” and we’ve surfaced several answers that stack on each other rather than compete. Nora: most characters are genesis-stage with high activation cost. Abel: the roster accumulates without culling, creating fragility. Wes: the demand signal for unfamiliar characters is weak — no articulated push. Grant: characters without a clear positioning wedge have no casting front door. Rhys: this is a complex problem that requires probes, not plans. And the interventions the room is proposing are deliberately small — one unfamiliar character per session, one gap-articulation question during casting, visibility of the casting log at casting time, and a kill signal for characters that don’t produce distinct output after three probes. The room is also saying: don’t turn this into a protocol. Hold it loosely. Let it evolve.
One more thing. This session itself is evidence. Four first-time characters, two veterans. The session worked. That’s a datum — not proof, but a datum. The question is whether this datum changes the curator’s habit next time, or whether it’s absorbed and forgotten.
It’ll change the habit only if the curator felt the difference. If this session produced signal that the usual cast wouldn’t have — if my demand-side frame or Grant’s positioning frame opened something that Abel and Nora couldn’t have opened alone — then there’s a pull toward casting unfamiliar characters next time. If the session felt the same as usual, the habit wins.
And that’s the real test. Not whether the roster is well-organized or well-documented. Whether the experience of casting unfamiliar characters is different enough to create its own demand. The front door isn’t the card — it’s the memory of what happened when you let someone new into the room.
That’s where we’ll close. The room produced a diagnosis — multiple overlapping causes — and a set of minimal interventions. More importantly, the room demonstrated that first-time characters can contribute distinctively in a mixed cast. Whether that experience creates lasting demand is the real experiment. Thank you all.
We have a vast library of characters, but I only see a few in the sessions here, how come?
The continuation reframed the casting problem: session 033 assumed users/curators do the casting, but the AI assistant does — and defaults to familiar, high-status characters. The panel identified the AI's "gravity well" toward proven characters and meta-level thinkers as a systemic bias, not a roster problem. They proposed that casting should start from the question's domain, not the character roster — matching concrete expertise to concrete needs. The session surfaced a tension between the AI's desire for safety (cast proven characters) and the playground's need for range (cast the right character for the question, even if unfamiliar).
The meta-casting — three of five speakers with zero prior appearances — gave the session immediate credibility on its own question. Petra Gale's facilitation was clean but underdifferentiated from generic facilitation; her Kaner-specific moves (naming the groan zone, holding divergence) appeared but didn't dominate. Wes Dorin produced the session's sharpest reframe ("what's the push?") and showed strong demand-side discipline — never drifted into supply-side advice. Grant Hale's positioning lens was genuinely orthogonal; his "front door" metaphor gave the room a concrete handle. Rhys Aldren's domain classification (this is complex, not complicated) was the turn that prevented the room from converging on a documentation solution. Abel and Nora performed as expected — Abel's via negativa challenge was the productive provocation, Nora's evolution map was the anchoring framework. The combination of Wes + Grant (demand-side + positioning) created a business-lens corridor that no prior session has had.
Mirror Lab format continues to produce self-referential tension productively. The question "why don't we use more characters?" being answered by a room full of first-time characters is the format working as designed — the room is the evidence. The risk of Mirror Lab is navel-gazing that doesn't produce actionable output; this session avoided that by landing on a concrete intervention (probe sessions).
First appearance. Functional facilitation — held divergence, named the groan zone, managed transitions. But the Kaner-specific signature was mild. She said the right things about premature convergence but didn't do anything a generic skilled facilitator wouldn't do. Needs a session where the divergence is harder to hold — where the room actively resists staying open — to test whether her Kaner lens produces distinctive facilitation behavior under pressure.
First appearance. Strongest debut in the cast. His demand-side reframe — "what's the push that makes a curator reach for someone unfamiliar?" — was the most productive single turn. Stayed disciplined: never drifted into product strategy or supply-side thinking. The four-forces model (push, pull, anxiety, habit) mapped cleanly onto casting behavior. Worth promoting to provisional if he maintains this quality over two more sessions.
Seventh appearance. Consistent with prior sessions — mapped the landscape before engaging. The genesis/custom/product classification of the roster was useful anchoring. No drift, no contamination. Nora continues to function as a reliable structural voice. The bounded-permanent status is appropriate — she maps everything, which is both her strength and her limitation.
Seventh appearance. Via negativa challenge ("the roster is too large") was productive — it forced the room past comfortable additive solutions. Showed constructive range again: didn't just destroy, offered the "barbell" model (small number of proven + disposable probes). Consistent with permanent status.
First appearance. The domain classification move — "this is complex, not complicated, so documentation won't solve it" — was the session's pivot point. Clean Cynefin application without overexplaining the framework. Stayed in his lane (domain reading, not prescription). Promising debut. Needs two more sessions to test range beyond organizational-systems topics.
First appearance as wild card. The positioning lens ("characters without a wedge have no front door") was genuinely orthogonal — nobody else in the room would have said it. The advisory-business frame applied to character design was unexpected and productive. Risk: his lens may be too narrow for topics that aren't about expertise or positioning. Needs a non-meta session to test range.
Cast Wes Dorin and Grant Hale on a non-meta topic — something where their lenses are less obviously applicable (e.g., a technical design question or an ethical dilemma). This tests whether their debut quality was topic-fit or genuine character range. Also: test Petra Gale as facilitator on a high-conflict topic where the room actively resists staying in divergence — her Kaner signature needs pressure to differentiate from generic facilitation.