All personas are AI-generated interpretive approximations inspired by published work. No real person participated, reviewed, or endorsed.
2026-02-26 · standard-panel · Facilitated by Ines Moran
What distinguishes a useful taxonomy from a misleading one — given that all taxonomies simplify?
The question: what distinguishes a useful taxonomy from a misleading one — given that all taxonomies simplify? We’re starting with the premise that simplification is a feature, not a bug. The question is when it becomes a liability. Diverge. I want genuinely different answers to what makes one work.
A useful taxonomy survives contact with new data. That’s the first and most important test. Most taxonomies are built from existing examples — you look at what you have, find patterns, name the categories. They describe the past. The moment something new appears that doesn’t fit, you learn whether your taxonomy is robust or fragile. A fragile taxonomy forces the new thing into an existing category — “it’s kind of like this one.” A robust taxonomy has a mechanism for admitting it’s incomplete. The second test is via negativa: what does the taxonomy force you to exclude? Every categorization is also an exclusion. If you can’t name what falls outside your categories — what the taxonomy specifically cannot see — you don’t understand your own taxonomy.
Abel’s talking about resilience. I’m talking about structure. A taxonomy is a map of a space. And like any map, it needs axes. What are the dimensions? How do they relate? Where is each category positioned relative to the others? Most taxonomies are flat lists — they name categories but don’t show the relationships between them. A good taxonomy has internal topography. You can see which categories are adjacent, which are far apart, which are evolving toward each other. And every taxonomy has an evolution axis — some categories are genesis (invented for this specific system, not used elsewhere) and some are commodity (everyone uses them, they’re standardized). If your taxonomy is full of genesis categories — custom labels nobody else uses — it might be precise but it’s not transferable. If it’s all commodity categories, it might be transferable but it doesn’t capture what’s distinctive about your domain.
Both of you are evaluating the taxonomy as an object — does it survive data? does it have structure? I want to look at it as a process. A taxonomy is a frozen question. Someone asked “what kinds of things are there?” and the categories are the answer. But the question determines what you can see. A taxonomy of animals by habitat (land, sea, air) shows you different things than a taxonomy by diet (herbivore, carnivore, omnivore) or by evolutionary branch. None is wrong. Each makes certain relationships visible and others invisible. The ethical question of taxonomy — and there is one — is: whose question got frozen? In Bohm’s terms, a taxonomy makes certain assumptions visible and hides others. The hidden assumptions are the dangerous ones, because they look like the natural order rather than a choice.
All three of you are looking at the taxonomy as a knowledge structure. I’m looking at the person who has to use it. A taxonomy is an interface. Someone encounters it — in a dropdown menu, a filing system, a medical diagnostic guide, an API specification — and they have to make a decision: which category does this belong to? The quality of the taxonomy is determined by how often the user hesitates. If the user consistently knows immediately where something goes, the taxonomy is well-designed. If the user regularly pauses, second-guesses, or picks the wrong category, the taxonomy has a design flaw. It doesn’t matter how conceptually elegant the categories are. The taxonomy is as good as the user’s ability to apply it without ambiguity.
Four frameworks. Resilience, structure, process, interface. That’s already too many. If someone asked me “how do I know if my taxonomy is good?”, I wouldn’t hand them four frameworks. I’d hand them a checklist. Three questions, yes or no. First: when you encounter something new, do you know immediately where it goes? That’s Kai’s interface test. Second: can you name three things your taxonomy can’t see? That’s Abel’s exclusion test. Third: if you showed this taxonomy to someone in a different department, would they understand it without a glossary? That’s transferability. Three questions. If you need more than that, your taxonomy isn’t a taxonomy — it’s a research project.
Five framings in five turns. I want to check for genuine disagreement before we start integrating. Abel, does Max’s compression lose something you think is load-bearing?
Max’s compression is useful but it removed the most important thing: disposability. A good taxonomy should be easy to throw away. The moment you invest identity in your categories — “this is how we organize things here” — the taxonomy becomes an institution instead of a tool. It resists revision. It survives past its usefulness because abandoning it means admitting you were wrong. The best taxonomies are the ones the creator is willing to burn when something better appears. Max’s three questions are a snapshot. They tell you if the taxonomy works now. They don’t tell you if it will survive, or more importantly, if you’ll be willing to kill it when it stops surviving.
I take the point. But disposability is a property of the organization, not the taxonomy. Any taxonomy can be disposed of if the people using it aren’t attached to it. And any taxonomy becomes permanent if there’s an organizational investment — training, software, reporting — built around it. You’re stress-testing the context, not the object.
That IS the object. A taxonomy doesn’t exist in isolation. It exists in a context of organizational investment, identity, and sunk cost. The taxonomy that survives isn’t the best one — it’s the one with the most infrastructure attached. If you evaluate a taxonomy without evaluating its context, you’re evaluating a museum piece.
Abel and Max are having a framing disagreement, and it’s productive. Max wants the taxonomy to be evaluable as a standalone object — is it clear, exclusive, transferable? Abel says it can only be evaluated in context — does it survive, is it disposable, who’s invested in it? This is the complicated vs. complex distinction again. In the complicated domain, Max is right — you can evaluate the taxonomy against criteria. In the complex domain, Abel is right — the taxonomy is entangled with its context and can’t be separated.
The map analogy holds here. A map can be evaluated on its own terms — is it accurate, is it readable, does it show the right features? But a map’s usefulness also depends on who’s navigating and what they’re navigating toward. A subway map is excellent for passengers and useless for urban planners. The taxonomy is the map. But the question “is this a good map?” requires knowing who the navigator is.
That’s the design insight. There’s no such thing as a good taxonomy in the abstract. There’s only a good taxonomy for a specific user doing a specific task. The dropdown menu taxonomy for a customer support agent needs to be different from the classification taxonomy for a researcher, even if they’re categorizing the same things. The support agent needs speed and certainty — which category is this ticket? The researcher needs nuance and boundary cases — what doesn’t fit? Designing one taxonomy for both users is the root failure. It’s a generic design serving nobody well.
I’m hearing convergence around the idea that a taxonomy can’t be evaluated without its user and context. Let me check: is anyone still holding that a taxonomy can be evaluated on formal properties alone?
I’ll hold that position, partly. The three-question test works regardless of user. Can you place new items? Can you name what’s excluded? Is it transferable? Those are formal properties. They’re necessary conditions. But I’ll concede they’re not sufficient. You can pass all three and still have a taxonomy that fails in practice because it doesn’t match the user’s decision-making flow.
Necessary but not sufficient — that’s honest. I’d add one formal property Max missed: a good taxonomy tells you when it’s failing. Not just “this item doesn’t fit” but “the categories themselves are no longer carving the world at its joints.” That’s a meta-signal. If your taxonomy has no mechanism for detecting its own obsolescence, it’s already failing and you don’t know it.
We’re converging on a layered answer and I want to make sure we’re not smoothing over a real disagreement. Max and Abel are arguing about what’s in the taxonomy’s boundary. Max draws the boundary at the object — the categories, their properties, their formal relationships. Abel draws the boundary at the system — the object plus its context, its users, its organizational embedding. This isn’t a reconcilable disagreement. It’s a different answer to “what is a taxonomy?” Is it a tool or a system?
That’s the right question to hold. I don’t think we resolve it today. Let me name what we’ve built:
A real taxonomy on the table. We’ve been talking about taxonomies in the abstract, which is ironic — we’re taxonomizing taxonomies without a concrete example. The stress test is: take one of these criteria and apply it to an actual classification system. See what survives.
All personas are AI-generated interpretive approximations inspired by published work. No real person participated, reviewed, or endorsed.
Ines as facilitator produced the strongest confirmation of facilitator-suppresses-content. Her demand-side lens — "what job is this hired to do?", the forces model, the human behind the system — was entirely absent from her facilitation. Not once did she ask what job a taxonomy is hired to do. Not once did she surface push/pull/anxiety/habit forces. She managed the room well — named divergence, held the groan zone, checked convergence. But the Ines who would have been the most interesting speaker on this topic (taxonomies serve human needs, not just intellectual ones) was invisible. This is the second data point across two different character types (single-source Nora in 005, composite Ines here). The mechanism is now confirmed with reasonable confidence. Max Reeves's first playground appearance was sharp — counted concepts, demanded compression, fought elegance. Distinctive voice arrived immediately. Kai on a purely abstract topic (taxonomy) was the key character test: the human-experience lens stretched but held. Found the user of the taxonomy as the design subject.
Standard Panel on a meta/abstract topic produced more conceptual framework collision than previous sessions. The taxonomy question is genuinely abstract — no obvious domain owner. This distributed influence more evenly than sessions with clear home-turf advantages (cf. Nora on technology adoption in 002). Good test of topic-as-implicit-casting: neutral topic, more balanced room.
Strong facilitation — different from Nora's in 005. Ines's facilitation had a warmer, more relational quality — 'what are you actually disagreeing about?' vs. Nora's spatial 'where are we?' Both competent, stylistically distinct. But content suppression was total. No JTBD lens, no forces model, no demand-side reframe. The Perel/Kahneman seasoning — invisible. Facilitator-suppresses-content: confirmed across character types.
The key stress test: abstract topic, no obvious human-experience layer. Kai found it anyway — 'who uses this taxonomy and where does their experience of using it break?' The person navigating the taxonomy IS the interface subject. Quality_test held. Not forced — genuinely useful reframe. The bounded graduation condition (untested on abstract topics) weakens. Kai may have wider range than the Tribunal assumed.
Taxonomy is NOT mapping — it's classification, not positioning. Nora adapted: mapped the taxonomy's internal structure instead of mapping a landscape. 'Every taxonomy has an evolution axis — some categories are genesis (invented for this system) and some are commodity (everyone uses them).' Applied Wardley to a non-Wardley topic. Good flex, but still mapping. The 'can Nora do something that isn't mapping?' question remains open.
Strong on a non-Snowden topic. Reframed taxonomy as an inquiry problem — 'a taxonomy is a frozen question, and the question determines what you can see.' The Bohm source activated more than usual — assumptions becoming visible through encounter with different taxonomies. Good range signal.
First playground appearance. Immediate distinctiveness. Counted Abel's criteria ('that's four things — if I need to remember four things to evaluate a taxonomy, you don't have a taxonomy, you have a methodology'). Demanded the sticky-note version. Pushed against elegance relentlessly. Fried's fingerprint clear: 'if it can't be simple, it shouldn't exist yet.' Gawande's secondary showed in the checklist move: 'give me three yes/no questions I can ask of any taxonomy.' Passed the bar test in a new room.
Applied fragility to taxonomies — 'a taxonomy that can't survive new data is fragile, and most are.' Via negativa: 'the test of a good taxonomy isn't what it includes, it's what it forces you to exclude.' Built four criteria for taxonomy quality (mutually exclusive, exhaustive, generative, disposable), then stress-tested his own criteria. Build-then-break pattern holds in a fully abstract domain.
Session 007 tests Abel in a purely constructive context (convergence FORBIDDEN — no breaking allowed as a convergent move). Ines returns as speaker in a different cast (with Dara, Eli instead of the core six). Perel source activation test on a relational/spatial topic.