Understanding anything requires leaving most of it out. You cannot hold a situation in full complexity, so you select, abstract, and represent. Mathematics does this precisely. Language does it differently but with the same necessity: the sentence that captures something real does so by discarding what would obscure it. Both are simplification tools, and more useful for being so.
Organisations do this too, though rarely with the same awareness. A strategy is a simplification. A set of KPIs is a simplification. A consulting framework, a budget model, a job architecture: each selects what to count and what to leave out. That selection determines what can be seen, and therefore what can be done.
When the model becomes the thing
The problem is not simplification. The problem is losing track of the fact that you are simplifying.
When a model is held as a model, provisional and open to revision, it remains a tool for seeing. When it is held as reality, it becomes a constraint on what can be noticed. The statistician George Box put this precisely: all models are wrong, but some are useful. That sentence is often read as licence to be approximate. It is better read as a standing instruction: keep track of the wrongness. When you can no longer say what your model cannot show you, you have stopped holding it as a model.
In organisations, this transition tends to be quiet. There is rarely a decision to treat the framework as reality. It happens through repetition: the same categories used in enough meetings, the same metrics cited in enough reviews, until asking whether they fit the situation stops feeling necessary. proxy-capture describes one version of this precisely: when a metric becomes the target, the underlying thing it was meant to measure quietly degrades. The metric has forgotten it was an approximation. repetition-and-revelation makes the same observation at the level of the frame itself: the assumptions you work within most consistently are the ones you stop being able to see.
The language chosen is not neutral
The language used to do the simplifying matters more than is usually acknowledged. A financial model and an ecological model of the same situation are not two descriptions of the same underlying reality. They make different things visible. They define what counts as a problem and what counts as a solution. what-a-frame-is describes this: the frame does not just shape how you look at a situation, it determines what the situation can be. Choosing a language is choosing the range of available responses before the analysis begins.
This means the order matters. The sequence that works: understand the situation, then choose the simplification that fits it. What tends to happen instead is the reverse. The available language arrives first, the consultant’s framework, the sector’s standard metrics, the methodology the team already knows, and the situation is then read through it. The fit is usually close enough to be plausible. your-data-architecture-isnt-technical shows how this works with data systems: the choice of architecture embeds assumptions about the nature of knowledge, and those assumptions shape what the organisation can learn from its own data. The philosophical commitment is made early, often without noticing, and everything downstream follows from it.
Why the model cannot correct itself
the-frame-cannot-see-itself describes the structural consequence: the instruments of perception are built from the same assumptions as the frame. The model that has forgotten it is a model has no internal mechanism for registering when it stops fitting. Feedback arrives late, and by then tends to look like an execution problem.
separated-knowledge adds a further difficulty. The people most likely to notice where the model does not fit tend to be separated from those with authority to act on it. The mismatch is visible at the edge of the organisation. At the centre, where the model feels most like reality, it is invisible.
the-dimensions-of-not-knowing offers a taxonomy of what cannot be seen from inside a frame. The relevant category here is structural ignorance: not knowing what the current frame of inquiry cannot reach. That is precisely what a model forgets when it forgets it is a model.
The question to keep asking
The right simplification is never universal. Which dimensions to flatten, which complexity to preserve, which language to apply: this depends on what is actually generating the difficulty in a given situation. It is discovered, not imported, and it requires holding available models loosely enough to let the situation push back against them.
The question worth asking, regularly and before things become urgent: is this model working in service of understanding the situation, or is the situation being made to fit the model? The two can look identical from inside the model. The difference tends to become visible when something important starts falling outside the frame and no one can find a way to say so.
Further reading:
↳ two-kinds-of-elegance — on the difference between smoothness within a frame and the harder question of whether the frame is the right one (draft)
Garden notes
- what-a-frame-is — a frame is the enacted simplification; this piece asks what happens when that stops being acknowledged
- the-frame-cannot-see-itself — the structural consequence of a model that has forgotten it is one
- proxy-capture — Goodhart’s Law as a concrete case: the metric forgets it was an approximation of the thing, not the thing itself
- repetition-and-revelation — how the frame calcifies through repetition until it stops feeling like an assumption
- your-data-architecture-isnt-technical — how the choice of language for simplifying embeds philosophical commitments, made early and mostly unnoticed
- separated-knowledge — those who can see where the model does not fit are separated from those who can act on it
- the-dimensions-of-not-knowing — structural ignorance is what a model cannot see about its own limits
- contextual-excess — why long proximity to a situation makes the simplification harder, not easier, to question
- observation-and-reasoning — the two modes of inquiry this piece assumes must be held together