Sorting Things Out: every taxonomy describes and decides
The book Sorting Things Out examines how categories and standards organize the world while distributing visibility, work, and consequences. Reading it through the lens of a skills architecture extends that inquiry to the decisions a taxonomy makes possible.
Geoffrey C. Bowker and Susan Leigh Star study very different classification systems, from medical categories to standards and infrastructure. The book does not offer a method for building corporate catalogs. Its contribution lies in teaching us to see what a classification makes normal, what it leaves out, and the work required to sustain its boundaries. MIT Press, Sorting Things Out (opens in a new tab).
A skills taxonomy looks like an ordered list. Once it is connected to assessment, mobility, or investment, it starts distributing possibilities.
Naming makes things visible—and also draws boundaries
A named capability can be found, observed, and developed. Its definition creates a shared point for comparing experiences that previously used different languages.
That benefit requires a boundary. The category decides which manifestations count, which belong to another entity, and which remain noise. No definition captures all the work. The problem begins when a necessary simplification is presented as a complete and neutral description.
Hybrid roles reveal the cost. If a contribution crosses two domains, forcing it to choose one may hide its integrative function. Creating a skill for every combination preserves detail and destroys reuse. The decision depends on which boundaries and relationships the architecture must preserve and which uses the classification activates.
Residue contains information about the system
Bowker and Star pay attention to cases that do not fit neatly. These are not minor failures that should always be corrected in the person or the data. They may indicate emerging work, a poorly drawn boundary, or context that was missing during design.
One exception does not require changing the taxonomy. An accumulation of similar mismatches deserves monitoring. It is useful to record where it occurs, which decision is affected, and how it is resolved while the model learns.
Residue also generates invisible work. Someone translates, justifies the exception, preserves the evidence, and prevents the person from disappearing between boxes. If no one owns that task, the cost often falls on those whose work is represented least well.
The same category changes risk with its use
An approximate classification may be useful for exploring interests. It may be insufficient to exclude someone from an opportunity. The necessary quality depends on the consequence.
This requires connecting taxonomy, version, and purpose. A term that is useful for guiding learning does not automatically become valid for selection. An update that improves the current catalog may also alter the interpretation of historical results.
Governance must answer which uses each version supports, how a classification can be challenged, and which decisions require additional evidence. The question is not only whether the label looks right. It is what happens to it afterward.
Standards hide maintenance decisions
A stable category produces comparability. Keeping it too long can preserve a boundary the work has already crossed. Changing it with every new development makes the history impossible to interpret.
The book helps us recognize this tension without resolving it through a universal cadence. The taxonomy needs review signals: repeated exceptions, incompatible interpretations, new decisions, or groups that are systematically misrepresented. It also needs authority to decide when the mismatch justifies splitting, merging, redefining, or retiring a category.
In Memory Practices in the Sciences, Bowker extends his interest in the relationship among infrastructure, memory, and what a practice preserves. This perspective is relevant to versioning: changing a category also changes what can be remembered and compared. MIT Press, Memory Practices in the Sciences (opens in a new tab).
Design mechanisms for challenging the category
A responsible taxonomy does not depend solely on the central team detecting errors. The people who use and inhabit the categories need a way to point out that a definition does not represent their work or that an assignment produces an improper consequence.
An objection should preserve context: the category challenged, the evidence, the affected use, and the proposed alternative. Not every disagreement requires changing the catalog. It may reveal incorrect application, a local need, or a definition that requires clearer examples. Recording the resolution prevents the same tension from being negotiated from scratch in every area.
Observe the pattern of objections. If certain areas, career paths, or forms of contribution need to appeal more often, do not treat each case as an independent anomaly. The distribution of disagreement is information about the architecture.
A responsible architecture observes its own consequences
Classification will remain necessary. Without common categories, the organization loses language, traceability, and the ability to connect work with decisions. The responsible alternative is to treat every classification as a revisable intervention.
That means observing who appears clearly, what is scattered across labels, who absorbs the exceptions, and which decisions rely on the taxonomy. It also means admitting that a boundary useful today may become unfair or irrelevant tomorrow.
Sorting Things Out leaves an editorial and operational discipline: look at the boxes, but also at their edges. That is often where the earliest evidence appears that reality changed before the system responsible for describing it. By design, the possibility remains open that a category useful today will need a different boundary tomorrow.
To extend this reading, see Minimum viable governance for a skills architecture, which develops a complementary dimension of the problem.
Keep exploring
How to maintain an architecture without turning it into a permanent project
Maintaining an architecture starts with distinguishing a signal worth recording, a small change with a clear owner, and a structural decision that deserves deliberate review.
Minimum viable governance for a skills architecture
Before scaling an architecture, define which decisions need an owner, a threshold, traceability, and the ability to pause.
Who can see what: access, privacy, and purpose in talent data
Who needs to see a piece of talent data, in how much detail, and for how long? The answer begins with the authorized task, not hierarchical rank.
How to tell whether a skills-first initiative creates value after the pilot
To decide what comes after the pilot, examine which decisions changed, which outcomes followed, and what it costs to sustain the mechanism. Participation and coverage do not yet demonstrate value.
How long does a skills assessment remain current
An older assessment may still guide development yet be insufficient for assigning a critical responsibility. Its date alone does not determine whether it remains useful.
What AI can infer about a skill—and what it should not infer
What can AI claim about a skill from digital traces? Moving from data to signal, inference, and decision involves leaps that require control.