A survival game · the odds come from real vampire bat studies

You are a vampire bat.

You are a young vampire bat. Each night, bite a sleeping cow or go hungry. Two empty nights in a row and you have no blood left, so you have to beg or starve, and only the friends you groomed and fed will feed you. Choose where you sleep, who you groom, and who you save. Survive ten nights.

Learn the real science →

Show tips for the first two nights

Every run is recorded anonymously: your choices, never your name. The logs describe how people play and will one day let the other bats hunt with real players' strategies.

Top seasons

Show

#	Player	Species	Score	Survived	When
Loading scores…

Achievements

Earned once per browser, kept between seasons. Three are secret.

How the odds work

Where the numbers come from. Every number in this game is read from one data file, bat-data.js, and every entry there carries either a citation or a model flag with its design reasoning written beside it. Where a field study gives a number, the game uses it. Where no field analog exists, the constant is flagged and its reasoning is stated; no constant ships with neither. That is a deliberate policy choice, and it is disclosed here rather than hidden. These notes read the same file, so what you read here cannot drift from what you play against. Full references are on the Learn more page.

Sourced numbers. Adult bats fail to feed on about 7% of nights; bats under two years old fail on roughly a third (Wilkinson 1984). You start young: your hidden hunting skill is drawn around a 28% fail rate and improves by ×0.93 with every fed night, down to a floor of 8%, the direction real juveniles learn in. Roughly 70 fasting hours is fatal (Wilkinson 1984; McNab 1973). Blood reads out of 100, a full crop: you begin with 67, two nights of fuel, and any unfed night costs 33, or 50 in rain, in the storms, or after a flight to the far valley, whether or not a bite was ever rolled. When your blood hits zero, each roost-mate who fed that night rescues you with P = 0.02 + 0.60 × bond^k (k is the season's exponent, below), capped at 0.9: real donations track prior grooming and sharing, not kinship alone (Wilkinson 1984; Carter & Wilkinson 2013). On your first two nights the colony carries a youngster: if you beg and any roost-mate the beg can ask is hanging beside you, she feeds you, no roll (most of Wilkinson's observed sharing went to young bats; the rule is model). From night three the curve above decides. You do not have to hit zero to be fed: on any unfed night that still leaves you some blood, a present partner who ate and trusts you at least 0.70 brings up part of her meal with P 0.35 (17 blood, 10 points), which is what a couple of close friends are for; the values are model. Donations are graded, a sip to a full crop, and build bonds in proportion; grooming escalates into sharing (Carter et al. 2020). Regurgitation is one bout of one meal, so you can bring up one donation a dawn, a second only while your crop is still nearly full. You can give on any dawn, fed or not, as long as you keep 33 of your 100 blood, a night of fuel (what you bring up is the last meal you took, not a reserve; model), and blood is never free to give: whatever is left in your crop at the season's end scores 45 points for every 33, and flying out with 33 or less adds 12 points to your fail chance (a bat on empty hunts worse; model). Grooming runs well beyond hygienic need and tracks partnerships (Wilkinson 1986; Carter & Leffer 2015), and it keeps the mites off: once your mite risk passes 50% they can move in and drain 17 blood a night (Lewis 1995). Bright moonlight suppresses vampire bat foraging (Crespo, Burns, Mitchell & Linhart 1972, J Mammal 53:366-368; Saldaña-Vázquez & Munguía-Rosas 2013): a bright night adds 10 points to every bat's fail chance and a dim one 5; the magnitudes are model. Bonded bats forage together and find each other by individually distinctive contact calls (Ripperger & Carter 2021; Carter et al. 2012; Carter & Wilkinson 2016), which is why the bat already at a wound is one of your roost-mates 50% of the time.

The pasture. Each night the paddock deals one to five animals (most often three; how many of a herd sleep within reach is never the same twice) from three tiers, and size is the legend:

Calf: meal 25 blood, thin skin (bite −4%), occupied 10% of the time. Real vampires hit calves and the most accessible animals first (Turner 1975); the values are model.
Steer: meal 33 blood, the standard dinner, occupied 25% of the time.
Prime cow: meal 50 blood, and someone usually got there first: occupied 55% of the time on night one (every animal is half as likely to be taken on the first two nights), rising 3 points a night to 75%, because favored hosts are re-attacked night after night and old wounds reopened (Greenhall 1972; Turner 1975); the occupancy curve is model.

A full meal is 40 to 60% of a bat's body mass, 15 to 25 mL (Greenhall 1972; McFarland & Wimsatt 1969); the in-game scale is relative, model. Reopening a wound another bat opened is near-instant (Greenhall 1972), so any bite at a contested or waited wound is 5 points easier. Wilkinson's failure rates are measured per night and a disturbed herd is abandoned (Greenhall, Schmidt & Lopez-Forment 1971), so you get exactly one bite roll a night: a miss wakes the herd and the hunt is over. The far valley, open on nights 3 to 8 (shut in a squall or a drought), is the gamble: it deals its own one to five animals with prime cows weighted ×2, cattle nobody bleeds are fatter (meals ×1.3), few bats fly that far (occupancy ×0.5), and a meal taken there pays 20 extra points. Against that, your bite is +10% more likely to fail and a hungry night costs the 50 blood flight. In the P37 sim the far valley, flown from the hollow, still costs 4 to 6 points on average but pays off on a nearly empty crop or a rich night, and raises the ceiling and the death rate (flown from the outpost it is the better hunt, see the roosts below): several-kilometer excursions are routine (Wilkinson 1985a); every value is model. Sleeping at the valley outpost scouts the far herd (bite −3% there).

Dark hours. Each night gives you three dark hours, two in the storms: a model framing of usable darkness, since feeding bouts run 9 to 40 minutes and several bouts a night are observed (Greenhall 1972; Wilkinson 1985a). Approaching an animal costs one hour, each wait costs one, and at zero you fly home unfed. Squalls (7% of nights 3 to 8, bite +18%) and the storm finale (bite +15%) are model, with the energetic direction from McNab 1973. A rich night (8% of dark, calm nights) is bite −5% and meal +8 blood, also model. The picked animal's bite lands bar is your chance of landing a bite tonight with every term folded in (your skill, the moon, the weather, the far valley, an open wound, fever, hunger, the luck damper); tap it for the itemized list. Hunger does not count against you on the first two nights. Dark hours themselves are not shown as a bar: the pasture line tells you how many are left.

The encounter. If another bat is at the wound you see who she is and your real chance with every move. A stranger's type is read from her tell from the first meeting (a young bat, an adult, a big dominant female). Share the wound is offered only with a roost-mate, and is tolerated with P = 0.05 + 0.50 × bond; a stranger never shares. Simultaneous feeding is rare in the field, 5 cases in 36.6 hours across 65 bats, mostly mother and young (Wilkinson 1985a), and bond predicts who forages together (Ripperger & Carter 2021); the slope is model. A refusal is a feeding-place confrontation call, no injury, and that animal is done with you tonight (Sailler & Schmidt 1978, a captive study, cited as such). Tolerated, you co-feed and the bond warms +0.05 on both sides (Ripperger & Carter 2021; Carter et al. 2020); a shared wound never pays the fed-alone bonus. Shove wins 50% of the time against a young bat, 30% against an adult and 15% against a dominant one; a roost-mate rolls as an adult (30%), the attempt costs −0.10 bond win or lose, and she remembers. Larger bats displace smaller ones at a wound, an adult male over a yearling (Wilkinson 1985a); nobody is ever hurt, the loser lifts away; the odds are model. A lost shove takes the dark hour (a fight takes time) and that animal is done with you. Since no female roost rank exists (Crisp, Brent & Carter 2021), rank here is feeder-scoped only. A shove is a short wing-off: a marker sweeps a bar with a green zone in the middle, wider the better your odds. A tap in the green adds up to 15% to the shove odds, a tap outside takes up to 10%, and no tap rolls exactly the odds above, which stay the centre. It is there for feel, not biology (model); our simulation puts a good tapper at about 41% of shoves won instead of 33%, worth about 8 points a season. Wait her out: bouts end, so she leaves with P 0.5 each dark hour, and the drained wound then yields 17 blood less (never below 17); sequential feeding at one wound is observed (Wilkinson 1985a), the drain is model. Pick another animal costs nothing beyond the hour already spent.

The roosts. Vampire bats use several roost trees and move between them (Wilkinson 1985a). Each game draws 3 or 4 roosts, the hollow you were born in and the rest, and every one is an independent bundle: 4 to 16 residents (around 8), a hidden generosity, a hidden disease risk and a hidden aggression risk, all model. No roost is the good one by design; you learn a roost by sleeping in it, and after one dawn its card shows who lives there and what you saw. From night 2 you choose where to sleep on the flight home, after the hunt, so the bats who can feed you if you came home empty are the ones at the roost you pick; night 1 and the storm nights lock you to the hollow. Every night every bat sleeps at home unless the season's switching roll (drawn between 2% and 12% a night, far-rangers ×3) sends her to another roost, so a friend is sometimes not where you expect her and a stranger sometimes is; in the storms nobody switches. Your first dawn somewhere new, the residents groom you back at half rate and no one shares with a beggar they trust below 0.30; anywhere, only bats who trust you at least 0.10 will feed you when you beg, because sharing happens among familiar associates (Wilkinson 1984). At dawn you may fall sick, P = 0.02 + 0.05 × crowding + 0.12 × the roost's disease score, ×1.5 on a first visit and ×0.7 at home: two dawns of 67 less stamina (out of 100), groom-backs at 50% of their odds, a bite 15 points likelier to fail, and no bringing up blood for anyone. Sick bats groom less and are groomed less (Stockmaier, Bolnick, Page & Carter 2018; Ripperger, Stockmaier & Carter 2020); the rates are model. Or a resident may bite you, P = 0.01 + 0.10 × the aggression score, ×1.15 while you are the stranger and 0 when a bat you trust at least 0.50 hangs beside you: 11 blood at once and two dawns of a stiff wing (residents guard a roost against strangers, Wilkinson 1985b; rates model). Mites are a separate roll with a separate cure.

The other roosts. Each roost you can sleep at besides the hollow offers one trade, an upside paid for with a downside, and its card says so in these numbers. The barn eaves hang over the herd: the next night a near bite gets −8% on its fail chance and a calm near miss costs 17 blood instead of 33, but the rafters are fought over (aggression at least 0.60). The mango hollow is big and warm: never fewer than 10 bats, bonds grow ×1.5 as fast and newcomers are groomed back in full, but it is damp: the dawn fever chance is ×2. The culvert is cool and quiet: a fever or a wound clears at the dawn there, the fever chance is ×0.3, catching it from a bat you touch ×0.25, and nobody fights, but only 5 bats at most sleep there. The valley outpost overlooks the far valley: the next night a far bite gets −15% on its fail chance and a far miss costs 25 blood instead of 50, but the ridge is cold: 3 blood gone by dawn. These are model trade-offs, plausible (damp, crowded roosts spread disease; a small bat loses heat on an exposed perch; roosting by the cattle shortens the commute) but not sourced; every value is model. Priced in the sim (20,000 seasons a cell): sleeping away for the right reason now costs about nothing (a card-reading bat ends within 2 points of one who never leaves), each roost pays off in its own situation and loses outside it, and from the outpost the far valley is the better hunt every time, which is what the cold night there buys.

The year. Two things have already happened to the colony when a season opens, and they hold all ten nights: one that favours a couple of close friends (a rough year of scarred males, a stingy year where only close friends feed a beggar, a sickly year, a lean year of thin herds) and one that favours a wide circle (a generous year, a restless year of roost-hopping, a crowded year, a quiet year of half-empty roosts). You read them once, in the first dusk's briefing, and their icons stay in the status bar. Each is a trade: the sim holds every year's median score within 5% of the others, so there is no good year to refresh for, only a different one. Two more surprises land on random nights between 4 and 8, drawn from a deck of twelve (hunters, a big storm, cattle disease, plague, dominant bats, injured ears, a cattle drive, drought, newcomers, a mite bloom, bright nights, a cold snap); each pops up with its own sound the night it lands and sits in the status bar while it runs. The culls, storms and roost invasions have field analogs (Wilkinson 1985a, 1985b); the amounts are model.

The season. No two seasons pay off the same way, so a single habit cannot win every game. Four more things are drawn once per season and deliberately not shown: you read them from what happens. Contagion: every bat you groom or share with is a contact, with 25% per contact at full contagion and ×5 if she is coughing; coughing bats are marked on the network, groom back at half rate, and appear more often in roosts with a high disease score. Sick bats groom less and are groomed less (Stockmaier, Bolnick, Page & Carter 2018; Ripperger, Stockmaier & Carter 2020); no study measures infection per grooming bout, so the per-contact rate is model, a cost of contact for the game. The rescue curve: the chance a fed bat feeds you when you beg is P = 0.02 + 0.60 × bond^k, with k drawn between 0.6 and 2: a tight season (k high) rewards a couple of very close partners, an open season (k low) rewards a wide circle. The partner panel shows each bat's chance under this season's curve, so the shape is there to be read. Donation tracks relationship strength (Carter & Wilkinson 2013); the exponent is model. Attention: you can keep 3 to 7 real bonds warm; beyond that every bond fades an extra 0.035 a night per extra bond, and the dawn panel says so. Grooming budgets are finite (Wilkinson 1986); the numbers are model. Switching: the colony's nightly roost-switching rate is drawn between 2% and 12%, so in some seasons a wide circle hedges your friends' absences and in others it is wasted. A meal for a hungry (not starving) bat pays 11 points and builds bond with diminishing returns, while a save pays in full; with one donation a dawn, each has to be worth making.

The luck damper, and hunger. One constant is pure engineering, and it is labeled that way here. After two straight unfed nights your next bite is 10 points easier, and the night says so: you fly careful tonight. It has no field analog and the game does not pretend it has one. Its reason is variance capping: in a ten-night run, two bad rolls must not become a death spiral you never chose. Separately, when your blood is at 33 or below, the high-variance options (an occupied prime cow, the far valley) are tagged risky, the energy-budget rule, risk-prone under deficit (Stephens 1981). It is not framing only: flying out that hungry adds the 12-point hungry term to your bite (not on the first two nights), itemized when you tap the animal's bite lands bar.

Plausible, for play (invented for the game, flagged so you know): the learning rate, bond decay (0.02 a night), your two old friends starting at a bond of 0.55 or more and the rest of the hollow lower, every roost's hidden bundle, roost-mates' hidden personalities (big-hearted, tight-furred, far-ranger, thin), close friends bringing you part of a meal on a hungry night, and roost-mates rescuing each other through their own friendships. Roost-mates can genuinely starve: a begging bat that no one feeds, including you, dies, and its friendships die with it.

Score and balance. Score is a fitness ledger: nights survived, feeds (a prime cow fed alone and a far-valley meal pay extra), co-feeds, displacements, donations scaled by size and need, lives saved, and at the season's end a survival bonus, your bonds valued the way the season values them (bond^k, summed), your single deepest bond (×45), every partner who ever fed you (12 each), the blood left in your crop (×45), and a full-roost bonus if nobody in the hollow died. Death keeps the ledger you earned. The 2026-09-04 relaunch rebuilt this economy because players had found that handing blood to every bat every night was always best: the leaderboard restarted that day, and the earlier board is kept under Before the relaunch. sim/batsim4.py Monte-Carlos 20,000 seasons per policy from the same data file (the October 2026 baseline, 20,000 paired seasons a policy). Medians for a bat who never leaves the hollow: keeping a few close friends 738 (it survives 75% of seasons), sharing widely 737 (69%), and a first-time player who grooms a little and rarely gives 516 (58%). A wide circle wins open and low-contagion seasons, a couple of close friends wins tight and high-contagion ones, and the years swing that gap either way. Every year's median sits within 5% of the others, and a roost wound decides under 3% of deaths. The far valley, always taken, scores the same median as never going (639 against 646) with a higher ceiling and a survival rate of 57% against 64%: a real gamble, not a trap. Only flagged constants were tuned to get there; no cited number moved.

What gets recorded. Every run writes an anonymous log: each choice you make, the state of your bat when you made it, and how the night turned out. It carries a random id for this browser so your separate runs can be told apart from other people's, and nothing else: no name, no exact time, no device details. The leaderboard name you type is never joined to it. The logs are used in aggregate to describe how people play this game and which strategies do well, and later to let the other bats hunt with strategies drawn from real players. Only summaries will ever be published.