---
name: Bounded Rationality
status: draft
territory: interest-and-power-analysis
host_mode: boundary-critique
also_loadable_in: []
msi_wired: false
sources:
  - title: Simon, Herbert A. (1955), "A Behavioral Model of Rational Choice," Quarterly Journal of Economics 69(1):99-118
    url: https://doi.org/10.2307/1884852
  - title: "Simon, Herbert A. (1957), Models of Man: Social and Rational, Wiley"
    url: https://openlibrary.org/works/OL1205030W
---

# Bounded Rationality

## Why it matters

Real decision-makers don't pick the best option — they can't. With limited time, limited information, and limited mental horsepower, they take the first option that clears a "good enough" bar and stop looking. The shape of their limits shapes the choice as much as the facts do.

For example: a family hunting for an apartment cannot tour every unit in the city, price each one against every other, and forecast which block will gentrify — the search would never end, and the perfect apartment would be rented out before they found it. So they set a floor — under this rent, this many bedrooms, this far from work — walk through a handful, and sign the first one that clears the bar. They never see the slightly-better place two listings down, and they were never going to. A company setting a price does the same: it doesn't compute the revenue-maximizing number across every possible demand curve, it lands on a figure that beats last year and covers costs, and moves on. In both cases the choice that gets made is governed less by what's out there than by where the searcher's limits made them stop. Herbert Simon called this *satisficing* — searching until something is good enough, not until it is best — and won a Nobel Prize for showing it is how real choice actually works.

- **What it reveals.** That a decision was shaped by the limits of the chooser — what they had time to see, could afford to compute, were able to know — not just by the facts, so a "best" outcome was never on the table and the realistic ceiling was "good enough."
- **How it changes the read.** You stop asking whether someone made the optimal choice and start asking what they could actually have known and computed — and whether the option set they searched, and the bar they stopped at, were the real constraint.
- **When to foreground it.** When an actor is being judged for missing the best option, or when a plan, model, or expert is being trusted as if it could survey everything — any time a decision is treated as if its maker were omniscient.
- **What you'd miss without it.** That the environment — how options are ordered, what default is pre-set, what information is shoved to the front — quietly steers the choice, so a bad outcome no individual could fix is the design's fault, not the chooser's.
- **Where it misleads.** It explains why optimization fails; it does not excuse a careless decision. Stretched into "they were doing their best under constraints," it becomes a blanket alibi for any outcome — when the real question is whether the bar they set was defensible for the stakes.

## How to invoke it in Ora

You have a plan, policy, study, or design in front of you that looks rigorous, and you suspect its framing has quietly settled questions that should be open — who it's really for, what it counts as success, who never got asked. Bounded rationality rides along inside that critique: it's one of the thinking tools the boundary work always has on hand to test what the plan's makers could actually have known.

Describe the artifact and what feels off about its scope, and ask:

> "Do a boundary critique of this redevelopment plan — surface the boundary judgments it's treating as given, and tell me who's affected by it but has no voice in it."

Ora names the artifact under critique, then walks the twelve boundary categories in four groups — motivation, control, expertise, legitimacy — asking each twice: what the plan assumes now, and what it would assume if the affected counted. When the critique reaches the expertise group — who is treated as the expert, what knowledge counts, what *guarantees* the plan will work — bounded rationality is what lets Ora ask the harder question underneath: could anyone actually know and compute what this plan assumes its decision-maker knows? A plan that leans on data, models, or expert authority as its guarantee of success is exactly where an assumption of unbounded rationality likes to hide.

One thing to know: the words *boundary critique*, *who is excluded*, *whose voice is missing* are what route you here, to the host analysis — not the phrase "bounded rationality" on its own. You don't summon this model directly; you ask for the boundary critique, and the model is always available inside it. A request for a neutral who-benefits read goes elsewhere (to a cui-bono analysis) — boundary critique takes a deliberately *critical* stance, surfacing contestation rather than smoothing it.

Name the system and what feels excluded, and Ora supplies the twelve-category structure itself — you don't need to know the framework, and you don't need to name bounded rationality for it to do its work. If you only have a vague unease ("this plan trusts its own forecasts too much"), that's enough; Ora will ask which assumption you suspect is being treated as settled.

One thing Ora won't do: hand you a "balanced" boundary that resolves the disagreement. It reports the gaps as live political contestation owned by the affected parties, not as objective findings the analyst gets to settle — because making boundary judgments visible and arguable is the whole job; eliminating them is not on offer.

## How it works

In the middle of the twentieth century, economics ran on a particular fiction. The textbook decision-maker — call him *Econ* — was perfectly rational: faced with a choice, he surveyed every available option, scored each one against every relevant criterion, and selected the precise optimum. He had all the information, infinite patience, and a mind that never strained. He was also a creature who has never existed.

Herbert Simon, an economist and psychologist who refused to look away from how people actually decide, broke with that picture. Watch a real person, he pointed out, and you see nothing like Econ. Watch a family looking for somewhere to live. They do not — cannot — examine every apartment in the city, because there are thousands, the listings change daily, and a lifetime wouldn't be enough to tour them all and compare. So they do something else entirely. They decide in advance what would be acceptable — a rent ceiling, a number of bedrooms, a tolerable commute — they look at a handful of places, and they take the first one that clears the bar. The same thing happens when a company sets a price: it does not solve for the single revenue-maximizing figure across every conceivable curve of customer demand, because it cannot see that curve. It picks a number that covers its costs and beats last year, and it gets on with business.

Simon gave the move a name: *satisficing* — a blend of *satisfy* and *suffice*. You search until you find something good enough, and then you stop. You do not search until you find the best, because finding the best would cost more than the difference is worth, and usually it cannot be done at all. His sharpest image for it was the needle in the haystack. The old picture imagined you sifting the entire haystack to find the single sharpest needle. But that is not what anyone does, or should. You rummage until you find a needle sharp enough to sew with, and then you sew. The point of the search was never the sharpest needle. It was a finished seam.

This was not a claim that people are stupid or lazy. It was a claim that the mind is *finite* — that, as Simon put it, the capacity of the human mind for solving complex problems is very small next to the size of the problems the world hands us. Given that gap, satisficing is not a failure of rationality; it is what rationality looks like once you account for the cost of thinking. And it carries a consequence that turns out to matter enormously: if people search until "good enough" and then stop, then the *shape of the environment* — which options they encounter first, what gets set as the default, which facts are pushed to the front and which buried — quietly decides what they end up choosing. Move the good option to the top of the list, or make it the default, and more people land on it, with no one's intelligence changed at all. The lever on better decisions is often not a better decider but a better-arranged world.

Simon won the Nobel Prize in Economics in 1978 for this — for showing that real choice is shaped by the boundaries of attention and computation rather than by flawless optimization. The reframe is durable because the limits are permanent: there will always be more options than anyone can weigh, and the realistic question is never "did they find the best?" but "did they know where to stop, and was the world arranged so that stopping there was wise?"

## Framework & implementation

*This section uses Ora's own terms for the parts of an analysis, so that if you open the actual mode and lens files they line up. Each is glossed in plain language on first use.*

### Pipeline execution

Bounded rationality is one of the **always-loaded mental models** of the Boundary Critique analysis — a thinking tool the mode keeps on hand for every run, listed in the mode's **`ANALYTICAL PERSPECTIVES`** block under "always loaded." Unlike Ulrich's twelve boundary categories, which *are* the mode's output skeleton, a mental model supplies no sections of its own. It informs the critique from inside: here, it sharpens the **expertise cluster** (the categories *expert*, *expertise*, and *guarantor*) and the **decision-environment** category under control. The premise it forces is that what a plan's decision-maker can actually know and compute is itself a boundary — so treating the planner as omniscient quietly hides the line their cognitive and informational limits draw around the whole analysis. The mode runs at **Gear 4**, Ora's most thorough setting — a **Depth analyst** and a **Breadth analyst** read the artifact independently, each critiques the other's reading, both revise under that critique, a flagged factual claim is web-checked, and a consolidator and formatter assemble the result. The model threads through those stages like this.

**Detection.** The model becomes relevant on the cases in its **Detection Signals** — an actor's decision is being judged against a fully-rational-actor standard that doesn't match its actual capacity; an outcome is being blamed on the chooser when the environment that constrained them is the real cause; a system produces poor results that no individual user can correct without re-architecting it. Inside a boundary critique these surface as a particular kind of buried boundary: the artifact appeals to expertise, data, or models as the guarantee that it will work, without surfacing what that expertise could not have known.

**The Depth and Breadth analysts.** Two models read the artifact in parallel; neither sees the other's work. As each walks the **expertise cluster**, bounded rationality runs the lens's **Application Steps** against the *guarantor* category — the artifact's answer to "what guarantees this plan succeeds?" It asks: was the binding constraint on the decision-maker identified (time, information, attention, computation)? Were the aspiration levels — the "good enough" bar — set explicitly, or is the plan satisficing blindly while presenting itself as optimal? Is the decision environment designed to steer choices toward good outcomes or against them? The **Depth analyst** presses one such gap hard; the **Breadth analyst** checks whether the same unbounded-rationality assumption recurs across the *expert* and *expertise* categories and into the *decision-environment* category, where a plan often treats as fixed an information environment it could in fact redesign. The model never overrides the twelve-category structure — it supplies the harder question to ask *within* the categories where claims of knowledge live.

**Cross-adversarial evaluation.** Each analyst's reading is handed to the *other* to critique, and the model's signature failures, keyed to its **Common Failure Modes**, are caught here. The chief one is **chooser-blame**: a draft that pins a bad outcome on the decision-maker when an identical chooser in a better-arranged environment would have done better — the tell is that no environmental redesign is even considered. The opposite error is **satisficing-as-license**: treating "they were constrained" as a complete defense of the outcome without ever asking whether the aspiration level was defensible for the stakes. The evaluator also catches **bias-conflation** — the draft using "bounded rationality" as a loose synonym for "made a mistake" rather than naming the structural condition precisely.

**Revision and claim-check.** The reviser restores the environment-versus-chooser distinction wherever the draft collapsed it, and makes the guarantor critique explicit: where a plan rests its case on data, a model, or expert authority, the revision names what that guarantor could not have known and computed. Crucially, it **resists revising toward neutrality** — the mode's character is critical, so a passing artifact keeps the edge and does not soften into "everyone did their reasonable best." Where a reading rests on a checkable factual claim about the artifact or its decision-makers, that claim is flagged and sent to a web-search tool before the revised draft proceeds.

**Consolidation and output.** The model contributes no section of its own; its work shows up *inside* the mode's set sections, organized by the consolidator and placed by the formatter. The artifact is named in **System under critique**; the load-bearing assumptions, including any reliance on an all-knowing decision-maker, in **Boundary judgments currently embedded**. The bounded-rationality findings land in the **Per-category audit** — specifically in the **Expertise** cluster (*expert* / *expertise* / *guarantor*) and the **decision-environment** category under **Control**, each rendered as the three explicit lines `is` / `ought` / `gap`. Where the planner's limits turn out to draw the deepest line, that surfaces in the **Worldview (category 12) — extended** block. The constituencies a bounded decision-maker could not see or weigh appear in **Affected-but-not-involved parties**. The model's most direct contribution is to **Implications for action**: naming where a plan assumes an omniscient decision-maker, and what changes when you don't — typically, that the fix is to redesign the choice environment or lower the plan's reliance on a guarantee no one could honor, rather than to demand a smarter planner. The closing **Boundary judgments as contestation** block and the per-gap **Confidence per gap** record complete the output.

**What the analysis will not assert.** It never proposes a "balanced" boundary that resolves the disagreement, and it never presents the gaps as objective findings — they are political judgments owned by the affected, not by the analyst. And it does not let bounded rationality become an alibi: the model is used to expose where a plan over-trusts what its makers could know, not to excuse a careless decision as the inevitable product of constraints.

### Origin and evidence

The model is Herbert A. Simon's. He set it out in "A Behavioral Model of Rational Choice" (1955), in the *Quarterly Journal of Economics*, which replaces the optimizing agent of classical economics with a decision-maker who has limited information and limited computing power and therefore searches sequentially against an aspiration level rather than evaluating all options at once. He developed the account in *Models of Man: Social and Rational* (1957), the synthesis of his early decision-theoretic work, where the term *satisficing* and the analysis of aspiration-level dynamics are laid out. Simon's central claim is the **finitude of the chooser**: optimization across all options on all criteria is exponential in the number of dimensions and infeasible at any realistic scale, so a tractable substitute — set a "good enough" bar, search until the first option clears every bar, then stop — is what real rationality consists of. He restated the position for an economics audience in his 1978 Nobel Memorial Prize lecture, "Rational Decision Making in Business Organizations." The line of work continues in two directions that the lens names as adjacent: Kahneman and Tversky's heuristics-and-biases program, which catalogs the particular patterns of error that occur within bounded rationality, and Gerd Gigerenzer's ecological-rationality tradition, which argues that simple heuristics are often well-adapted to the specific environments in which they are used. Inside a boundary critique, Simon's finitude claim does a precise job: it converts "the planner knows" from an assumption into a question.

### Applications and common uses

Bounded rationality is a working model anywhere a decision is being judged, or a system designed, for real and finite people rather than idealized optimizers — and, inside a boundary critique, anywhere a plan's credibility rests on a claim of knowledge.

- **Interrogating a plan's guarantor.** Its native job in this host: when a policy or design points to its data, its forecasting model, or its experts as the reason it will work, bounded rationality asks what that guarantor could not have known or computed — turning an appeal to authority into a checkable boundary on knowledge rather than a settled fact.
- **Decision-quality review.** Judging a past decision by what its maker could realistically have known and searched, not by hindsight's full information — and separating a defensible "good enough" bar from a negligently low one.
- **System and choice-environment design.** Because the lever on outcomes is often the environment, not the chooser, the model justifies investing in better defaults, clearer information ordering, and simpler option sets — the recognition that a better-arranged world beats a demand for better deciders.
- **Organizational analysis.** Modeling how firms and bureaucracies actually decide — through satisficing routines, standard operating procedures, and aspiration levels that adjust to feedback — rather than through a fully-rational-actor fiction that does not match their capacity.
- **Program and policy evaluation.** Surfacing whether the people who designed a program could have known what it assumes they knew, and whether its measure of success was set at a level appropriate to the stakes.

In every case the contribution is the same: a claim that someone made the right call, or that a model guarantees an outcome, is re-described as a claim about what a finite mind could know and compute — and is opened to argument on exactly that ground.

### Failure modes and when not to use it

The model's characteristic ways of going wrong are catalogued in its **Common Failure Modes**:

- **Satisficing-as-license.** Bounded rationality is invoked to defend any decision regardless of whether the aspiration level was appropriate. The tell is that the analyst defends the outcome by citing the chooser's constraints without examining whether those constraints justified the *specific* decision. The correction: the model explains why optimization fails, not why poor satisficing succeeds — the "good enough" bar must itself be defensible for the stakes.
- **Chooser-blame.** Poor outcomes are pinned on the decision-maker when the environment was the cause. The tell is that an identical chooser in a better-designed environment would have done better, and no one is asking about the design. The correction: redesign the choice environment before blaming the chooser.
- **Aspiration-stagnation.** The "good enough" bar stays fixed despite feedback that it no longer fits the problem. The tell is the same decision rules producing repeatedly poor outcomes. The correction: review and update aspiration levels on a cycle proportionate to the decision's stakes.
- **Bias-conflation.** "Bounded rationality" is used as a synonym for "made a mistake," collapsing the structural condition into the catalog of specific biases. The tell is the term standing in for any error at all. The correction: bounded rationality is the structural condition; particular biases are particular patterns of error within it — name the specific bias if one applies.

**When not to reach for it.** When the decision-maker genuinely *could* survey the full option space — a small, closed, well-specified choice where exhaustive evaluation is cheap — the satisficing frame adds nothing, because optimization was actually feasible. When the real question is *which* error a chooser made, the precise diagnosis is a specific bias, not the structural condition that contains it. And as a guard against its own misuse: when an actor is being held to account for a careless decision, bounded rationality is a tool for examining whether the aspiration level was defensible, not a ready-made excuse for the outcome — reaching for it to end the inquiry rather than to sharpen it inverts the model.

## Related

- **Boundary Critique** — the analysis this model is loaded inside; it walks Ulrich's twelve boundary categories, and bounded rationality sharpens the expertise cluster by asking what the planner could actually know.
- **Ulrich CSH Boundary Categories** — the required, foundational lens that *is* the boundary critique's twelve-category skeleton; bounded rationality informs that skeleton rather than supplying it.
- **Tragedy of the Commons** — another model often live inside a boundary critique: where bounded rationality bounds what the planner can *know*, the commons structure names what the planner's frame can deplete when no one speaks for a shared resource.
- **Free-Rider Problem** — a commons-adjacent model about who benefits from a public good without bearing its cost — a question, like bounded rationality's, about a limit the planner's frame treats as settled.

## Sources

- [Simon, Herbert A. (1955), "A Behavioral Model of Rational Choice," Quarterly Journal of Economics 69(1):99-118](https://doi.org/10.2307/1884852)
- [Simon, Herbert A. (1957), Models of Man: Social and Rational, Wiley](https://openlibrary.org/works/OL1205030W)