The 20 Wrong Answers People Are Most Confident About

Ask a room what weapon defined the early samurai. Most will say the sword. In our production data, 56.2% of players said exactly that, and only 21.9% picked the answer the historians give.

A misconception index ranks wrong answers by how often people choose them, not by how often a question gets missed. This one was built from a July 2026 read-only export of LearnClash duel data: 24,065 questions carrying real first-encounter answers, 267 of them publishable after the volume and quality gates. Twenty of those wrong answers are below.

Seven of the eight detailed rows were picked more often than the truth. A curated cut of the row-level data is public on Kaggle under CC BY 4.0 if you would rather run your own analysis of it.

What a Misconception Index Measures

A hard question and a misconception look identical on a scoreboard. Both show a low percentage correct. They are not the same thing, and the difference is where the wrong answers land.

On a genuinely hard question, misses scatter. Players who do not know the answer spread across every option roughly evenly, because none of the wrong answers is more attractive than the others. On a misconception, the misses pile onto one specific option. Something in the culture, in a movie, in a half-remembered classroom, points everybody at the same wrong door.

So the metric here is not the miss rate. It is the share of all first-encounter answers that went to the single most-chosen wrong option. Our 12 most-missed questions piece ranked questions by wrong-answer rate. This one ranks the wrong answers themselves, which turns out to be a different list.

The Eight Headline Misconceptions

These eight came through a fact-check pass and a statistical pass. Six clear the strict Bonferroni signal bar the published dataset uses; the stars row and the clams row reach only nominal significance, so read those two with an extra grain of salt. Sample sizes are small by survey standards, so every row below carries its n. Read them as strong signals, not as national polling.

The samurai carried a bow

Picked: curved longsword, 56.2%. Correct: bow and arrow, 21.9%. n=32.

This is the widest margin in the whole published cut. Under the guessing model, one specific wrong option on a question like this should attract about 15% of answers. Eighteen of thirty-two players landing on the katana has a probability of roughly one in twelve million by chance.

The myth has a clear source: four centuries of art, film, and tourism built around a sword. But the sword became the samurai’s signature weapon late. From the tenth through the fourteenth centuries the bow reigned on Japanese battlefields, and the warrior class described its own calling as kyuba no michi, the way of the horse and bow.

The early samurai was a mounted archer first and a swordsman second. The katana became the emblem after the fighting style that defined the class had already changed.

The strongest insect is a dung beetle

Picked: leafcutter ant, 42.9%. Correct: dung beetle, 20.0%. n=35.

Ants own this category in the popular imagination, and leafcutters specifically, because everyone has seen the footage of a column hauling leaf fragments many times their own size. The measurement says otherwise. Rob Knell and Leigh Simmons put horned dung beetles on a pulling rig and published the result in Proceedings of the Royal Society B: the strongest individual pulled 1,141 times its own body weight.

Relative to body weight, the horned dung beetle (Onthophagus taurus) is the strongest insect ever measured. The leafcutter footage is real. It is just not the record.

The shortest war lasted under an hour

Picked: the Six-Day War, 42.4%. Correct: the Anglo-Zanzibar War, 18.2%. n=33.

Six days sounds impressively short, and the name does the persuading. The actual record is not close. On 27 August 1896 the British bombarded the palace at Zanzibar after a succession dispute, and the fighting was over in between 38 and 45 minutes, with 38 the figure most often cited. Around 500 Zanzibari defenders were killed or wounded in that window against one wounded British sailor, which is the part the fun-fact version usually drops.

The Anglo-Zanzibar War of 1896 is the shortest recorded war in history, measured in minutes rather than days.

Dogs see blue and yellow

Picked: red and green, 41.5%. Correct: yellow and blue, 26.8%. n=41.

This one is a straight inversion. Dogs are close to red-green color blind, which is exactly the pair most players chose as the colors dogs can see. The 1989 work by Neitz, Geist and Jacobs found two cone pigments in the dog retina, peaking around 429 and 555 nanometers, and behavioral color-matching tests that fit dichromatic vision.

Dogs are dichromats: they discriminate blues and yellows well and struggle with reds and greens. The red ball on green grass that your dog cannot find is a genuine visual problem, not a training failure.

Alcohol sits between carbohydrate and fat

Picked: dietary fiber, 34.1%. Correct: pure ethanol, 12.2%. n=41.

Only 12% knew pure ethanol carries more energy per gram than carbohydrate and less than fat. The FAO conversion factors are 4 kcal per gram for carbohydrate, 7 for alcohol, and 9 for fat. Fiber goes the other way, contributing roughly 2.

Why fiber wins the vote is easy to guess: it is the answer that sounds like a nutrition question should sound. Ethanol reads as a chemistry answer, so people rule it out before they do the arithmetic.

Pure ethanol delivers about 7 kcal per gram, closer to fat than to sugar.

Stars twinkle because of our air, not theirs

Picked: “the stars’ own atmospheres bend their light”, 29.0%. Correct: air turbulence bends the light, 9.7%. n=31.

This row is the strangest in the set. Only 9.7% of players got it right, which is below the 25% a pure guesser would average on a four-option question. When the correct answer scores below chance, players are not merely failing to know it. They are being actively pulled somewhere else.

The pull is a plausible-sounding phrase. “The star’s atmosphere” is a real thing, it sounds astronomical, and it puts the cause out in space where the star is. The actual mechanism is much closer to home: astronomical scintillation is starlight refracting through moving layers of Earth’s atmosphere. Planets sit close enough to show a small disk, so their light averages across many turbulent paths and holds steady.

Stars twinkle because of turbulence in Earth’s atmosphere. That is also why planets, which show a disk rather than a point, mostly do not.

The iron answer is clams, with an asterisk

Picked: oysters, 27.0%. Correct: clams, 10.8%. n=37.

Oysters are the shellfish everyone associates with mineral density, so they collect the vote. Clams are the answer, and this is the row where I want to show my work, because checking it turned up a mess worth knowing about.

The number that circulates in iron-rich-food lists is 23.8 mg per 3 ounce serving of canned drained clams, against roughly 5 mg for a same-size serving of beef liver. It comes from reference tables, not from a measured product, and measured products scatter: a survey of canned clams found 0.86 to 59.46 mg per 100 g, with whole baby clams averaging 30.5 mg per 100 g and chopped or minced ones averaging 1.4 mg per 100 g. The paper is titled Inconsistent values for iron content complicate recommending clams as an iron source, which tells you how settled this is. I pulled the current USDA FoodData Central entry for mixed-species canned drained clams while writing this section and it reads 2.68 mg per 100 g, below raw beef liver at 4.9 mg per 100 g.

One more caveat. Clam iron is mostly non-heme (measured at roughly 0.5 mg heme against 4.5 mg non-heme per 100 g in Manila clams), and non-heme iron is absorbed far less efficiently than the heme iron in liver.

Whole clams can carry several times the iron of beef liver per serving, but the figure swings by an order of magnitude with species and processing, and less of it is absorbed. Oysters are not a bad guess either: on the same USDA basis, eastern canned oysters read 6.7 mg per 100 g, above beef liver. Whole clams are just capable of far more. Players who feel the correct answer deserves an asterisk are right.

Cleopatra grew up speaking Greek

Picked: Egyptian, 25.2%. Correct: Greek, 34.0%. n=103.

This is the largest sample in the published cut, and the only one of the eight where the correct answer still won. A quarter of players said Egyptian, which is a lot of people confidently placing the wrong mother tongue on the most famous Egyptian in history.

Cleopatra belonged to the Ptolemaic dynasty, Macedonian Greek rulers who held Egypt for nearly three centuries while conducting their court in Greek. Plutarch’s Life of Antony reports that she was the first of her line to learn Egyptian, which is presented as remarkable precisely because her predecessors had not bothered.

Cleopatra VII’s native language was Koine Greek. Egyptian was a language she learned, and by her dynasty’s standards that was unusual.

Twelve More the Index Flagged

These twelve cleared the same volume gate and the same fact-check, but their margins are narrower, so they belong in a list rather than in a headline. Question wording is verbatim from the published dataset. The share column is the portion of all first-encounter answers that went to that one wrong option.

Question	Most-picked wrong answer	The answer	Share (n)
Why do sloths descend from trees once a week?	To drink fresh water	To defecate on the ground	39.4% (33)
What bias makes past events seem more predictable than they actually were?	Confirmation bias	Hindsight bias	34.4% (32)
What surprising color is the Knight Bus in Harry Potter?	Red	Purple	33.3% (30)
What childhood bone disease is caused by vitamin D deficiency?	Osteoporosis	Rickets	32.4% (37)
On which planet in our solar system is a day longer than a year?	Jupiter	Venus	32.4% (37)
Which mammal has the longest gestation period of any species?	Blue whale	African elephant	31.7% (41)
Who is the queen and wife of Jupiter in Roman mythology?	Venus	Juno	30.0% (30)
Which common food has more potassium than a banana?	Broccoli	Potato	29.9% (77)
What color is the skin of a polar bear?	White	Black	28.9% (45)
In nutrition, which everyday fruit is botanically classified as a berry?	Cherry	Banana	28.6% (49)
Oxford University is older than which famous Mesoamerican empire?	Mayan Empire	Aztec Empire	28.2% (39)
What famous line is never actually spoken in the classic film Casablanca?	”Here’s looking at you, kid"	"Play it again, Sam”	28.1% (32)

A note on the banana: that question exists twice in our catalog, in two different topics, and both copies show the same knowledge gap, though players split onto different wrong answers (cherry in one, raspberry in the other). It appears once here.

How the Index Was Built, and the Number I Got Wrong First

The export script runs read-only against production and counts one thing: a player’s first encounter with a question inside a 1v1 duel. Timeouts and joker-assisted answers are excluded. Repeat encounters are excluded, which matters, because a question you have already seen once tells you about our spaced repetition scheduling rather than about what you believed before we told you.

The funnel, run on 27 July 2026:

Stage	Count
Question documents scanned	39,113
Questions with first-encounter answers	24,065
Topics in the corpus at export time	691
Rows with at least 10 answers	2,902
Rows with at least 30 answers	268
Publishable after quality filters	267
Rows where one wrong answer out-polled the correct one	21
Rows in the statistically defensible signal cut	28

Here is the part I got wrong on the first pass, and it changed the whole analysis.

I wrote the first version of this against a 25% guessing floor. Four options on screen, so a random click lands on any one of them a quarter of the time. Obvious. It is also wrong for wrong answers, and our internal methodology review caught it before anything was published.

LearnClash rotates distractors. Each exposure shows the correct answer plus three distractors sampled from a master pool, and for every row in this article that pool holds five. So a specific wrong option is only on screen about three times in five, and its expected share under pure guessing is 0.75 divided by 5, which is 15%. The correct answer is different: it is displayed every single time, so its guessing baseline really is 25%.

That asymmetry is why the star-twinkle row is worth more than it first appears. A correct answer at 9.7% is below its own 25% chance line, and a wrong answer at 29.0% is nearly double its 15% line. Both numbers move in the same direction, and neither would have been readable against a single flat 25% baseline.

Each published row therefore carries a one-sided exact binomial p-value against that rotation-aware null, with a Bonferroni correction for pool size before a row counts as signal.

What we left out. Eight rows that cleared the volume gate are sitting in a hold queue because our fact-check found the question wording, not the players, to be the problem. One row was worse: the question itself was defective, we rewrote it in production the same week, and the version bump reset its answer counters, so it dropped out of the publishable cut entirely. It has to earn its way back with fresh answers. That is the intended behavior of the version gate, and it is the main reason the index is smaller than the raw export.

Three caveats travel with every number here. Duel answers only, so Solo mode play does not feed this. Counts aggregate players across all app locales while the question text shown above is English. And at n between 30 and 50, margins under roughly 15 percentage points are inside sampling error, which is exactly why the twelve-row table is a list and not a ranking of eight.

Why a Confident Wrong Answer Beats a Blank

For anyone writing questions, a misconception is more useful than a hard fact. A question nobody knows produces a shrug. A question where the audience commits to a specific wrong answer produces an argument, and arguments are what people remember afterwards.

For anyone studying, the same asymmetry works in your favor. Getting a question wrong when you were sure creates a much stronger memory hook than drawing a blank, which is one reason our 3-stage SRS re-serves missed questions after 7 and 90 days instead of quietly retiring them. The surprise is doing work, and the retention curve those checkpoints produce has its own writeup.

And for anyone teaching, this list is a decent diagnostic. If a class confidently answers “leafcutter ant” or “the Six-Day War”, they are not short on knowledge. They are carrying a specific and correctable belief, which is a much easier thing to fix. Our wider catalog and rating numbers sit in the LearnClash statistics report, and the question sets themselves are indexed in the trivia questions hub.

Three Questions I Could Not Answer With This Export

The dataset is public, and I would genuinely like someone to take it further than we did. Three things I wanted and could not get:

1. Does a misconception survive the correction? Our counters stop at the first encounter by design, so I cannot yet see whether a player who picked the katana picks it again six weeks later. That join exists in principle. It did not exist in this export.
2. Are these misconceptions cultural or universal? Counts aggregate every locale, so a belief that is strong in one language community and absent in another looks like a moderate global effect. Splitting by locale would be the single most interesting cut available.
3. How much of the pull comes from wording? Eight rows are held back because our own phrasing was doing some of the misleading. A systematic way to separate “the audience believes something false” from “the question is badly built” would improve this index more than another thousand answers would.

The starter notebook loads the published cut and reproduces the top-20 chart, the strict-signal table, and the per-topic pull. One thing to know before you run it: the public file is the curated cut described above, not the full export, so its row and signal counts come out smaller than the funnel table’s. If you publish anything from it, I would like to read it.