Man vs. Bear Seed 42
A Monte Carlo experiment10,000 encounters11 variables24 possible worlds

ManVS.Bear

Everyone has an opinion about who wins. We ran the numbers instead. Ten bear species, twelve kinds of human, thirteen things you might be holding, five tactics, seven landscapes, four group sizes and four distances from help, all rolled at random and played out second by second. Is it a bluff? Does the spray come out in time? Who gives up first?

See the results

The headline numbers

Each figure comes from 10,000 simulated fights; the small print shows the range across 24 possible worlds where the model’s coefficients shift. Fix any variable in your fight and everything on the page recalculates.

Every square is one fight

Sorted by species, each horizontal band is one bear, polar bear at the top, sun bear at the bottom. Hover a square for details; click to replay that fight second by second.

MAN/BEAR Run your own
01To scale

The contenders

Adult males standing on their hind legs, average weights, drawn to the same scale as a 1.75 m human. Under each bear is how often its fights end in a death in this run. Tap a bear to rerun all 10,000 fights against it.

02The experiment

Set up your fight

By default, every variable is rolled at random for each of the 10,000 fights. Fix any of them and every fight happens under that condition, while the rest stay random. Pick a preset, or build your own nightmare.

Presets
How should we count?

In real life most people meet a black bear or a brown bear empty-handed, not a panda while holding a spear at 75. “Real-world” weights each fight by how often that situation actually happens. “Equal” counts every option the same, which is the fairer lab view for comparing species or weapons head to head.

03Summary

What 10,000 fights say

04The internet’s favorite question

How many people does it take to beat a grizzly?

A crowd of average adults against one bear. Only eight can reach it at once; the rest wait their turn. The bear picks a target, switches every few seconds and swats at the crowd. Two versions: a realistic one, where the bear can quit and people can break and run, and a fight to the death.

—

Simulating crowds…

Opponent
Armed with

UnhurtDown, survivesKilled
05Figures 1–3

How a fight unfolds

Each ball is one fight. First the bear’s charge splits into bluffs and real attacks. Real attacks then end in an escape, an early stop by spray, gunfire or noise, or physical contact, before landing in an outcome bin. Pipe thickness matches the number of fights passing through.

Fig. 1 · Bins left to right: walked away, badly hurt, killed, beat the bear
Fig. 2

What decides the outcome?

For each variable, the gap in death rate between its deadliest and safest option, in percentage points. The longer the bar, the more that choice changes your fate.

Fig. 3

The risk mountain

Bear weight and human age against the death rate, as terrain. The cliff rises where heavy bears meet older humans. Drag to rotate, hold Ctrl (⌘ on Mac) and scroll to zoom, and pick a weapon class to watch the mountain sink.

Front edge: bear weight (log scale)Side edge: human ageHeight and color: death rate
06Figures 4–6

Wounds, blood and time

Every blow lands on a body region that depends on how you’re positioned at that moment. When the fight ends, a second race begins: bleeding, first aid, and how long help takes to arrive. A tourniquet or firm pressure stops most limb bleeding.

Where the blows land

Fig. 4

If help takes this long

Fig. 5

What killed them

Fig. 6

07Calibration

Reality check

A simulation is only worth something if it reproduces what really happens. The model was tuned against seven numbers from published field studies of real bear encounters. The white ring is the published figure, the orange dot is the model. The gaps that remain are shown, not hidden.

Published studyThis model

Where the numbers come from

    Definitions don’t line up perfectly. Field studies count every close encounter in which someone carried a deterrent; the model only counts fights in which the bear charged. Treat gaps of a few points as noise.

    08Figure 7

    How sure are we?

    The model’s coefficients are estimates, not measurements. So the simulation replays the same fights in 24 “possible worlds”, each with strike power, bleeding rate, spray effectiveness and six other coefficients nudged within plausible ranges. The dot is the main estimate, the thick box covers the middle half of the worlds and the thin line covers 90%. Where two lines overlap, the difference between them isn’t trustworthy.

    The coefficients that move

    90% range

    Each world draws one multiplier per coefficient from a log-normal distribution. All worlds replay the same fights with the same random numbers, so the differences come from the coefficients, not from luck. These ranges show the model’s own uncertainty, not sampling error.

    09Figures 8–13

    Every variable, broken down

    Each row shows how fights under that option ended. The number on the right is the death rate. Unordered variables are sorted from deadliest to safest.

    Fig. 9

    Which bear, which weapon?

    Fig. 10

    Age and weapons

    Death rate by age, split by what the human is holding.

    Fig. 11

    When does the estimate settle?

    The survival estimate stabilizing as more fights are run. The shaded band is the 95% sampling interval.

    Fig. 12

    How long the struggle lasts

    Only fights that reached physical contact. Most short ones end when the bear decides the threat is neutralized.

    Fig. 13

    How each fight ended

    The turning point of each encounter. Bluffs and early interventions end most fights before they start.

    10The tape

    Watch one fight, second by second

    Every fight has its own seed and can be replayed exactly. Click a square in the map at the top, or pull one below.

      11Under the hood

      How the model works

      Sampling

      • Human type, age, species, terrain, time of day, weapon, tactic and time to rescue are drawn from a 50/50 mix of a uniform distribution and rough real-world frequencies, so rare combinations still get enough samples.
      • Results are recounted with importance weights for the view you pick: each fight’s weight is its probability under the target distribution divided by its probability under the sampling mix. The effective sample size is shown in the setup panel.
      • Bear age is weighted realistically (adult 45%, young 28%, old 15%, cub 12%). The bear’s state of mind depends on species: polar bears are often predatory, giant pandas almost never.
      • Every fight has its own sub-seed. The same seed always produces the same 10,000 fights.

      The fight

      • Starting distance comes from the visibility of the terrain and time of day. Under 15 m counts as a surprise encounter.
      • The bear first decides between a bluff and a real charge. Running makes a real charge more likely; standing your ground makes it less likely.
      • If the human reacts in time before contact, they can deploy spray, fire, sound a horn or brace a spear.
      • After contact, each second the bear strikes and the human responds. The bear has a resolve score and retreats when it hits zero. A defensive bear quickly loses interest in someone playing dead; a predatory one doesn’t.
      • Each human has stamina. When it runs low, blows weaken and they’re more likely to curl up. Obesity, heart disease and limited mobility make the same blow do more damage, increasingly so after 40.

      Wounds and rescue

      • Each blow lands on one of five regions. Odds depend on posture: arms and face when fighting upright, back and legs when face down, back and backs of legs when caught running.
      • Regions bleed at different rates, the neck most of all. The initial bleeding rate is a share of blood volume per hour and slows exponentially as clotting kicks in.
      • A conscious human applies a tourniquet or pressure with a probability that depends on training and composure. That cuts limb bleeding by 80% and other bleeding by 30%.
      • Losing about 40% of blood volume is fatal, less for vulnerable profiles. Snow, night and wet ground add hypothermia; waits past 12 hours add infection and shock. Deep head or neck wounds can kill on the spot. Heart attack mortality rises as help is delayed.

      Uncertainty

      • Nine coefficients (strike power, bleeding rate, spray effectiveness and more) shift by a log-normal multiplier in each possible world.
      • Each of the 24 worlds replays the same 4,000 fights with the same random numbers (common random numbers), so differences between worlds come from the coefficients rather than chance.
      • Ranges are the 5th and 95th percentiles across worlds. The headline estimate comes from the 10,000 fights run with nominal coefficients.

      Groups, crowds and real cases

      • Group size changes how early the bear notices you, how likely a real charge is, who it chases when people run, whether companions pile in, and how fast first aid and rescue arrive. It was tuned so that 63% of brown bear attack victims are alone, as in the worldwide data.
      • The crowd model runs a separate simulation: only eight people reach the bear at once, bare hands do little against fur, fat and bone, and in the realistic version the crowd breaks once a third of it is down.
      • Each replayed fight is matched against 13 documented real encounters by species, mood, tactic, weapon, group and outcome. Matches are approximate; the linked reports tell the real story.

      If you actually meet a bear

      • Carry bear spray where you can reach it in under two seconds. Field studies found it prevented injury in roughly nine out of ten close encounters, and the model agrees.
      • Never run. Bears can top 50 km/h.
      • Hike in groups of three or more and make noise. Parties larger than two are much less likely to be attacked.
      • Against a defensive brown bear, lie face down and protect your neck. Against a black bear, or any bear that is stalking you, fight back.
      • Carry a tourniquet and tell someone your route. In the model, they are among the cheapest lifesavers there are.

      This is a toy model, not survival advice. The coefficients and “real-world” frequencies are rough estimates inspired by the literature, not measured values. Follow the guidance of your local wildlife or park authority.