---
name: Occam's Razor
status: draft
territory: artifact-evaluation-by-stance
host_mode: balanced-critique
also_loadable_in: []
msi_wired: false
sources:
  - title: "Sober, Elliott (2015), Ockham's Razors: A User's Manual, Cambridge University Press"
    url: https://openlibrary.org/works/OL21108524W
  - title: Solomonoff, Ray (1964), A Formal Theory of Inductive Inference, Information and Control 7(1):1–22
    url: https://doi.org/10.1016/S0019-9958(64)90223-2
---

# Occam's Razor

## Why it matters

Between two explanations that fit the same facts, the one carrying the fewest assumptions is usually the right one — not because the universe loves simplicity, but because every extra assumption is one more thing that has to be true, and one more chance to be wrong. The more moving parts an explanation needs to work, the less likely it is that all of them happen to hold.

For example: your laptop won't connect to the internet, and the explanation that springs to mind is elaborate — your provider must be having a regional outage, or a recent update corrupted the network driver, or the router's firmware silently failed overnight. Each is possible. But the spare explanation, the one assuming the least, is that the Wi-Fi got switched off — a stray keystroke, a toggle bumped in a settings menu. Check that first. It costs ten seconds, and most of the time the unglamorous cause is the real one; the elaborate stories only earn their keep after the simple check comes back clean.

- **What it reveals.** How many separate assumptions an explanation has to stack up to work — and therefore how much has to go right, all at once, for it to be true.
- **How it changes the read.** You stop asking *"could this explanation be true?"* and start asking *"how many independent things must be true for it to hold, and is there an account that needs fewer?"*
- **When to foreground it.** Two or more explanations fit the same evidence; a proposed cause needs a long chain of coincidences; a debugging session is wandering into exotic territory; a mundane account is competing with a baroque one.
- **What you'd miss without it.** That an explanation can sound thorough precisely *because* it is overbuilt — each added mechanism feels like rigor while quietly lowering the odds the whole thing is correct.
- **Where it misleads.** Pushed too far it cuts past the evidence — a simpler story that fails to cover the facts is just wrong, not parsimonious; and "simpler" can hide assumptions, so a one-sentence explanation may rest on more tacit conditions than the longer one it beat.

## How to invoke it in Ora

You have an artifact to evaluate — a proposal, a diagnosis, a post-mortem, a plan — and part of its case rests on a chain of assumptions about why something happened or will happen. You want a fair read that notices when a simpler account of the same facts would do.

Share the artifact and ask:

> "Give me a balanced critique of this incident report — strengths and weaknesses, and flag where a claim is leaning on a stack of assumptions when a simpler explanation would fit."

This rides inside the Balanced Critique analysis, which surfaces an artifact's strengths and weaknesses at equal depth. The Occam's-razor lens is one of the always-present points of view that rides along: when a claimed strength or weakness depends on a long chain of assumptions, it weighs whether a leaner account of the same behavior is more credible, and it guards against critiques that pile up assumptions to manufacture a flaw.

One thing to know: phrases like *balanced critique*, *fair evaluation*, *strengths and weaknesses*, *what holds up and what doesn't*, or *neutral read* are what route you here. Naming the lens alone — "apply Occam's razor" — does not route; describe the artifact and ask for the balanced read.

Give it the actual artifact, not just its conclusion; the lens works by counting the assumptions an explanation actually requires, and it can only count what's on the page.

One thing Ora won't do: treat "simpler" as automatically true. It checks that the leaner explanation actually covers all the evidence before preferring it, and it treats a too-simple account that leaves data unexplained as wrong, not as a winner — parsimony never licenses ignoring the facts.

## How it works

You text a friend and hear nothing back for hours, and the mind gets to work. They're angry about something you said. They're reconsidering the whole friendship. They've decided, quietly, to let you drift. By dinnertime you've assembled a small tragedy — when the explanation that needs almost nothing is that their phone is face-down on silent in another room. Both stories fit the same single fact, total silence. One of them asks you to believe in a hidden grievance, a private deliberation, and a decision; the other asks you to believe a phone is on silent. We reach for the elaborate version constantly, and rarely notice we've done it.

Here's the engine underneath. Every independent assumption an explanation rests on is a separate bet, and each bet can lose. When the assumptions are independent, their odds multiply — and multiplication is merciless. Suppose an explanation needs five separate things to be true, and each one, generously, is 80% likely. The whole explanation is only as good as all five holding at once: 0.8 multiplied by itself five times is about 0.33. A story that sounded careful, each piece individually plausible, turns out to be roughly a one-in-three shot. Cut it down to two assumptions and you're back above 60%. That is *why* the spare explanation tends to win — not magic, not aesthetics, just arithmetic. Fewer load-bearing guesses means fewer ways for the whole structure to fall down.

So the working rule is: when several explanations fit the same evidence, start with the one that assumes the least. A web app starts throwing 500 errors right after a deploy. The baroque hypothesis is a race condition in the new async code causing a deadlock under production-only load patterns that only show up because of network-latency differences. The plain one is that the deploy introduced a typo in an environment variable, and the database connection string is now wrong. Check the plain one first — open the config. Nine times out of ten the mundane cause is the real one, and you've spent a minute instead of an afternoon chasing a ghost. Medicine carries the same proverb for the same reason: *when you hear hoofbeats, think horses, not zebras.* Hoofbeats fit both animals; horses are simply far more common, so you bet on horses first.

The nuance is where this earns its keep, and it cuts the other way too. The razor is *not* a law that the simpler explanation is true, and it is *not* a verdict that ends the inquiry. It is a rule for ordering what you test first. A simpler explanation that fails to cover the evidence isn't elegant — it's wrong, and you keep the messier account that actually fits the data. The phone-on-silent story collapses the moment your friend turns out to have read the message and said nothing; the typo theory dies the moment the config checks out clean. Parsimony tells you where to look first; it never licenses looking away from facts that don't fit. The principle is old enough to have a name. It's called Occam's razor, after a 14th-century English friar, William of Ockham, who put it as a blade for cutting needless complexity out of an argument: *entities must not be multiplied beyond necessity* — do not invent more causes than the evidence forces you to.

## 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

Occam's razor is one of the **always-loaded mental models** in the Balanced Critique analysis — it sits in the mode's **`ANALYTICAL PERSPECTIVES`** block under "always loaded," alongside narrative instinct, confirmation bias, Bayesian reasoning, devil's advocacy, and Walton's argument schemes. It is *not* the mode's method (Balanced Critique has no single required lens; its method is the symmetric strengths-and-weaknesses discipline itself); Occam's razor **informs** the read. The mode runs at **Gear 4**, Ora's most thorough setting — a **Depth analyst** and a **Breadth analyst** evaluate the artifact in parallel, critique each other (**cross-adversarial evaluation**), and revise. Balanced Critique surfaces an artifact's strengths and weaknesses at *parallel depth* and refuses to collapse them into a tidy verdict — and Occam's razor is the perspective that fires when a strength or weakness rests on a stack of assumptions a leaner account would not need.

**Honest host-fit note.** Occam's razor's lens file scopes it to hypothesis-evaluation, debugging, and abductive reasoning — choosing among competing *explanations* of the same evidence. Balanced Critique is its **public host**, and a reader meets it here as a general thinking aid for fair evaluation, while its native use is breaking ties between rival explanations of one set of facts. Be fair about the fit: it's a broadly-useful reasoning discipline, loaded into this mode to keep critique honest, not a tool purpose-built for artifact evaluation — it earns its place by weighing how many assumptions a critique demands, not by being made for the job.

**Where the lens engages.** It activates on its **Detection Signals** — two or more hypotheses explaining the same observations equally well; a proposed explanation requiring a long chain of unlikely coincidences; a debugging session going down increasingly exotic rabbit holes; a conspiracy account competing with a mundane one; a design accumulating complexity without clear benefit. Its **Application Steps** list all plausible explanations for the evidence, count the independent assumptions each one requires, check that the simpler ones truly account for *all* the evidence (not just most), prefer the fewest-assumption explanation as the working hypothesis, and test the simplest first — escalating to complex accounts only when the simple ones fail.

**What it contributes to the analysis.** It sharpens the mode's **Weaknesses** section and its **Assumptions and uncertainties** section: when a claimed flaw depends on an unlikely stack of premises, the razor weighs it down to its real probability rather than letting the assumption-count pass as rigor. And it backs the **opinion-as-evaluation** guard (the mode's `opinion-as-evaluation`, **CQ4**) — a critique resting on a long chain of independent guesses is weak evidence for a flaw, no matter how detailed it sounds, because the chain's odds multiply down. It also informs the **Net assessment with residual tensions**, keeping an over-engineered objection from inflating the case against an artifact.

**Cross-adversarial evaluation.** At Gear 4 each analyst's reading is critiqued by the other, which catches the lens's signature failures, keyed to its **Critical Questions** and **Common Failure Modes**: **premature simplification** — preferring a leaner explanation that fails to cover the evidence; **assumption miscount** — a "simple" account that, once spelled out, smuggles in many tacit conditions and is not actually simpler; and **razor-as-verdict** — treating parsimony as settling the question rather than ordering what to test. The evaluator presses the core check: *does the simpler account actually explain all the evidence, and were the assumptions counted consistently across the competing explanations?*

**What the analysis will not do.** It will not treat "fewer words" as "fewer assumptions"; will not prefer a simpler explanation that leaves evidence unexplained; and will not let parsimony close the question — simpler-first means *test first*, not *decided first*.

### Origin and evidence

The principle is old and the formal case for it is recent. William of Ockham (c. 1323, *Summa Logicae*) gave it its enduring statement — entities must not be multiplied beyond necessity — though as a methodological maxim, not a proof. The modern philosophical reckoning is Elliott Sober's *Ockham's Razors: A User's Manual* (2015), which is careful about exactly *when* a preference for parsimony is justified and when it is not — there is no single razor but several, each warranted by different reasoning, and none of them a guarantee that the simpler hypothesis is true. The probabilistic backbone comes from Bayesian analysis: Jefferys and Berger's "Ockham's razor and Bayesian analysis" (1992) shows why simpler models that fit the data are favored — a model with fewer free parameters spreads its probability over fewer possibilities, so when it fits, it fits more sharply and earns more credit than a flexible model that could have fit almost anything. The deepest formalization is Ray Solomonoff's algorithmic theory of inductive inference (1964, frontmatter), which makes "simplicity" precise as *description length*: the best explanation is the shortest program that reproduces the observations, turning Ockham's maxim into mathematics.

### Applications and common uses

Occam's razor is a working tool wherever competing explanations contend for the same evidence and one of them is overbuilt.

- **Debugging and incident response.** Its native ground: checking the boring cause — a typo, a stale config, an off switch — before the exotic one, and ordering the investigation from fewest assumptions to most.
- **Diagnosis and differential reasoning.** Medicine's "horses, not zebras" — favoring the common, few-assumption condition that fits the symptoms before reaching for the rare syndrome, while still testing rather than declaring.
- **Evaluating arguments and theories.** Comparing rival accounts of the same data and preferring the one that needs the fewest independent things to be true — the everyday use against conspiracy-shaped explanations whose plausibility leaks away as the coincidences multiply.
- **System and product design.** Resisting architecture that accumulates mechanisms without clear benefit — every added component is another assumption about what the system needs and another way for it to break.
- **Reviewing critiques and post-mortems.** Catching an objection that sounds rigorous because it is elaborate, and asking whether a simpler account of the same behavior is more likely than the multi-step one being offered.

In every case the payoff is the same: the assumptions an explanation actually requires get counted, the leaner account is preferred *only* when it still covers the evidence, and parsimony orders the testing rather than ending it.

### Failure modes and when not to use it

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

- **Premature simplification.** Accepting a simple explanation that does not cover all the evidence. The tell is residual unexplained data sitting there after the "simple" verdict. Keep the explanation that actually fits the facts, even when it is the more complex one.
- **Assumption miscount.** Implicit assumptions in the "simple" hypothesis make it actually more complex than its rival. The tell is that when you force the simple account to be spelled out, it requires many tacit conditions. Enumerate the assumptions of *both* explanations explicitly before comparing counts.
- **Razor-as-verdict.** Using parsimony to declare the question settled rather than to order what gets tested. The tell is no empirical follow-up planned — "it's simpler, therefore it's true." Simpler-first means test first, not decided first; the razor ranks hypotheses, it does not confirm them.

**When not to reach for it.** When only one explanation actually fits the evidence, there is no tie to break and the razor has nothing to do — it adjudicates *between* competing accounts, not within a single one. When the difference in assumption count is trivial, preferring "the simpler" is splitting hairs and adds noise rather than signal. And the razor ranks explanations by plausibility; it does not, by itself, establish which is true — that still takes the empirical test the principle is telling you to run first, which is a separate task from noticing that one account assumes less.

## Related

- **Balanced Critique** — the analysis this lens rides in; surfaces an artifact's strengths and weaknesses at equal depth, where Occam's razor weighs whether a claim's assumptions outrun what a simpler account would need.
- **Narrative Instinct** — the sibling always-loaded mental model: where the razor distrusts the explanation carrying the most *assumptions*, narrative instinct distrusts the one carrying the most satisfying *shape* — two different guards against a too-comfortable account.
- **Falsifiability** — the operational partner: the parsimonious explanation must still be testable, because a too-simple account that can't be checked isn't a win — it's just an untested guess wearing the costume of elegance.
- **Confirmation Bias** — the distortion underneath the miscount: we under-count the assumptions of explanations we favor and over-count those of explanations we dislike, which is exactly why the assumption-counting has to be done consistently across both sides.

## Sources

- [Sober, Elliott (2015), Ockham's Razors: A User's Manual, Cambridge University Press](https://openlibrary.org/works/OL21108524W)
- [Solomonoff, Ray (1964), A Formal Theory of Inductive Inference, Information and Control 7(1):1–22](https://doi.org/10.1016/S0019-9958(64)90223-2)