A razor is a tool for cutting — for shaving away the explanations you shouldn’t bet on first. But here’s the thing about sharp tools: held the wrong way, they cut you. Occam’s Razor and Hanlon’s Razor are brilliant at one job — telling you where to place your first bet. They are catastrophic at a different job — telling you what is true. And the most common way smart people hurt themselves with these heuristics is by quietly promoting them from “best opening guess” to “final ruling,” then defending the ruling like it’s their honor on the line.
This is the lesson that makes the other lessons safe to use. We’re going to break both razors on purpose — show you exactly where Occam over-trims and where Hanlon lets the guilty walk free — so you never get blindsided by your own favorite shortcut.
Before you read — take a guess
You've learned that Occam's Razor favors the explanation with the fewest assumptions, and Hanlon's Razor says don't assume malice when incompetence explains it. What is the single most dangerous way to use them?
A razor is a prior, not a proof
Let’s name the through-line of this whole course in one sentence: the razors tell you where to bet first, not what is true.
Think of a weather forecast. “70% chance of sun” is an excellent reason to leave the umbrella home — it’s a smart default. But it is not a guarantee, and only a fool stands in the rain insisting the forecast said sun. A razor works the same way. It’s a calibrated guess about which explanation is most likely before you’ve looked hard. The moment new facts arrive, the honest move is to update — not to argue with the sky.
In the language of an earlier lesson: a razor sets a prior — your starting probability — that you then revise as evidence comes in. That’s the Bayesian spirit. A prior is supposed to move. A prior that never moves isn’t a prior; it’s a prejudice.
The one-line trap
A heuristic becomes a hazard the instant it stops your inquiry instead of starting it. “Simplest, so done.” “Probably incompetence, so done.” That little word done is where people get hurt.
When you treat a razor as a verdict, two old enemies show up. First, confirmation bias: the lean answer becomes a story you’re now emotionally invested in, so you notice every fact that fits it and wave away every fact that doesn’t. Second, you skip first-principles thinking — rebuilding the answer from the raw facts — precisely when the stakes are high enough to deserve it. The razor was meant to save you the expensive work when it doesn’t matter. It was never meant to forbid the work when it does.
Where Occam over-trims: reality isn’t always simple
Occam’s Razor says: prefer the explanation that needs the fewest assumptions. Great rule. But it comes with fine print that people love to ignore. The full quote (often attributed to Einstein) is the whole point:
Make things as simple as possible — but no simpler.
That second clause is the entire safety mechanism. Simplicity is only a virtue when the simple explanation actually fits the facts. If the lean story leaves evidence unexplained, it isn’t elegant — it’s just wrong, wearing a nicer suit. A simple explanation that doesn’t fit beats a complex one that does zero of the time.
And some corners of reality are genuinely, irreducibly complex. The simplest explanation is often plain incorrect in:
- Biology and the human body — riddled with feedback loops, redundancy, and exceptions.
- Economies and markets — millions of agents, no single lever.
- Climate systems — coupled, nonlinear, full of lag.
- Large software systems — where “the obvious one-line cause” is famously a liar.
The assumption-counting con
Occam only works when the explanations genuinely fit equally well and you count assumptions honestly. But “an assumption” is slippery. You can hide a mountain of complexity inside one confident word (“it’s just fraud”) and call it simple, or split one clean idea into five steps and call it complex. People game the count — usually without noticing — to make the answer they already liked look “simpler.”
There’s a third way Occam over-trims: it tempts you to force one cause when the truth is a conjunction. Multiple things can be true at once. The real story behind an outage is often “a bug and a bad config and a retry storm that turned a small failure into an avalanche.” Insisting on a single root cause there doesn’t make you rigorous — it makes you incomplete.
Complete the safety clause on Occam's Razor.
Pick the right option for each blank, then check.
Make things as simple as possible, but . The simple explanation only wins if it still .
Occam’s counterweight: Hickam’s dictum
Doctors learned this lesson the hard way, so they minted a counter-razor. Occam in medicine says: find the single diagnosis that explains all the symptoms — elegant, satisfying, often right in a young, otherwise-healthy patient. But against it stands Hickam’s dictum:
A patient can have as many diseases as they damn well please.
Reality is not obligated to be tidy for your convenience.
Worked example. An 82-year-old arrives with shortness of breath, swollen ankles, a mild fever, and confusion. The Occam-elegant move is to hunt for one disease that ties it all together. But in an older patient, multiple chronic conditions are the norm, not the exception — heart failure (the swelling and breathlessness), a separate brewing infection (the fever), and dehydration or a medication interaction (the confusion) can all be true at once. Chase the single unifying diagnosis and you can treat one problem while two others quietly kill the patient.
The lesson isn’t “Occam is bad.” It’s that Occam’s prior — bet on one cause first — must yield the instant the facts demand more than one. Base rates decide which razor leads: in a healthy 20-year-old, lean Occam; in a frail 82-year-old, Hickam is waiting in the wings, because the prior probability of multiple simultaneous conditions is just higher.
A senior engineer insists the production outage has 'one root cause' and dismisses teammates who point at a second contributing factor. Which mistake is this?
Where Hanlon excuses real bad actors
Hanlon’s Razor — never attribute to malice what is adequately explained by incompetence — is a gift for your blood pressure and your relationships. Most of the slights you take personally really were just someone being busy, tired, or clumsy. But used as a verdict, Hanlon becomes the perfect camouflage for people who are counting on you to give them the benefit of the doubt.
Here is the signature that should make Hanlon let go: a pattern of “mistakes” that all break in the same direction — toward the same person’s benefit. Random incompetence scatters. It makes errors that hurt the person making them, that point every which way, that get fixed once noticed. Intent has a direction. When every “oops” mysteriously profits the same party and somehow never gets fixed, that’s not a streak of bad luck. That’s design.
Once is incompetence; a pattern pointing one way is a strategy
Repeated, self-serving, asymmetric “errors” are the fingerprint of intent — fraud, manipulation, and abuse all wear the “oops, my bad” mask because it’s the cheapest disguise available.
Worked example. A subscription company makes signing up a one-click delight, but cancelling requires phoning a hotline that’s only open Tuesdays. It “accidentally” pre-checks a box that opts you into a paid add-on. Its “billing glitches” always charge you more, never less, and refunds take three emails. Apply Hanlon to any single event and sure — software is hard, mistakes happen. But step back: every error flows one way, toward their revenue, and none of the profitable “bugs” ever get fixed. That asymmetry is the evidence. This even has a name in the design world — dark patterns — precisely because it’s deliberate.
Two more nails in the coffin of naïve Hanlon:
- Power and stakes raise the prior on malice. When someone has clear motive, the means, and a track record, defaulting to “they probably just messed up” isn’t generous — it’s gullible. The more someone stands to gain, the more you should at least check for intent.
- Incompetence and malice are not mutually exclusive. It’s not a coin flip between the two. A malicious actor can also be sloppy; a scammer can have a buggy checkout. Hanlon tempts you into a false either/or. The real question isn’t “incompetence or malice” — it’s “how much of each, and which one should I act on?”
When it carries the three marks at once: direction (the error reliably benefits the same party), persistence (it survives being pointed out — they had every chance to fix it and didn’t), and stakes/motive (the benefiting party had something real to gain). One isolated, self-correcting, harmless slip stays in Hanlon’s column. A directed, sticky, profitable one has left it. You don’t need certainty of intent — you need enough signal to stop assuming its absence and start investigating.
Weight your reliance by the stakes
Both razors share one master dial: the cost of being wrong. The higher that cost, the less you should lean on the heuristic and the more you should actually investigate. A razor is a shortcut, and you take shortcuts on errands, not on the edge of a cliff.
This matters most for Hanlon, because the error is asymmetric. Assume incompetence when it’s really malice in a low-stakes setting — a coworker’s rude email — and the worst case is you were a little too kind. Make the same assumption in a high-stakes setting — security, scams, abuse, child safety — and you’ve handed an attacker exactly the cover they were betting on. There, you deliberately lower your reliance on Hanlon and raise your default suspicion, because the price of a false “it’s probably nothing” is unacceptable.
Same logic for Occam: a simple guess about why your toast burned costs nothing if it’s wrong. A simple guess about why a patient is crashing, or why a bridge is swaying, can cost everything. High stakes are the signal to drop the heuristic and rebuild from first principles.
For each situation, decide: does the razor still apply as a reasonable default, or does it break down (high stakes / a directed pattern / facts that no longer fit)?
Place each item in the right group.
- A stranger emails an "urgent" request for your password — "must be an honest IT mix-up"
- Your simplest theory of the outage leaves three log entries unexplained
- A vendor's "billing errors" always overcharge, never undercharge, and never get fixed
- A friend forgot to reply to one text last week
- An 82-year-old has four symptoms and you want one tidy diagnosis
- A new app crashes once on an unusual phone model
How to hold a razor correctly
So how do you keep the blade pointed away from yourself? Hold every razor as a starting bet you actively try to disconfirm. Not a conclusion to protect — a hypothesis to attack.
A practical loop:
- State the lean answer out loud. “Simplest: one root cause.” / “Probably just incompetence.” Naming it makes it a candidate, not a reflex.
- Try to break it. What evidence would prove this wrong? Go looking for that evidence on purpose — this is the direct antidote to confirmation bias, where you’d otherwise only collect what flatters the guess.
- Check fit. Does the lean explanation account for all the facts, or is it ignoring some? Unexplained facts mean it doesn’t fit — and fit beats simplicity.
- Weight by stakes. The more it costs to be wrong, the less you trust the shortcut and the more you rebuild from first principles.
- Drop it without ego. When the lean answer stops fitting, let it go. You were never defending it; you were testing it. Updating a prior on new evidence isn’t a defeat — it’s the entire skill.
The razor's real job
A good razor doesn’t end the conversation. It starts it — by giving you the smartest first thing to check. Used that way, it makes you faster and harder to fool.
Which behaviors mean you're holding a razor correctly (as a prior to test), not incorrectly (as a verdict to defend)? Select all that apply.
Match each failure mode to what it actually is.
Pick a term, then click its definition.
Recap
Big picture
Razors are priors, not proofs
- Hold a razor correctly
- Prior, not proof
- Sets your first bet
- Update on evidence (Bayesian)
- As verdict = blindfold
- Occam over-trims
- Complex systems: bodies, markets, climate, software
- "As simple as possible, but no simpler"
- Hickam: many causes at once
- Hanlon excuses bad actors
- One-way patterns = intent
- Motive + means + track record
- Malice and incompetence both possible
- Weight by stakes
- Higher cost = less razor, more inquiry
- Asymmetric error (security/scams)
- Connects to
- First principles when it matters
- Confirmation bias — don't defend it
- Prior, not proof
Prove you can break a razor on purpose
What is the core meta-point that makes both razors safe to use?
Check your answer to continue.
You now know the tools and their failure modes — where Occam over-trims, where Hanlon lets the guilty hide, and how to hold both as bets you keep testing rather than verdicts you defend. That’s the whole skill. Next up is the Final Exam: one shot, no retries — bring everything you’ve learned and prove you can wield the razors without cutting yourself.