Real facts, real animals

We traced 57 sources to the code, and say what each one really did

Before we build an animal or a kitchen substance, we read. A paper on fruit flies says they jump away from whatever looms at them. A paper on garden ants says their trail scent fades. Findings like these decide the mechanism: what the animal senses, and which way it moves. The exact numbers are then set by play, on a sheet of paper about 31 by 23 cm, where a cat is the size of a thumb.

In November 2026 we checked every source behind the animals and the kitchen science, one by one. Each reference was looked up again, by its DOI record, abstract, page or book record, and then traced to the line of code it shaped, if there was one. The chart shows what we found. A handful of sources were applied directly. More of them shaped a mechanism that shipped. Many described behaviour we wrote up and then never built, like the pigeon's hold-and-thrust head bob. Others we found only after building, and we list those as consistent with the game, or not, never as its basis.

We think that's the right order for a game for young children. Real findings decide which way things go, and the toy-sized page decides how fast. Where we checked and the game disagrees with the world, as it does for butter, baking soda and the dog's shake, this page says so.

Where 57 sources stand against the codeThe last bar, found later, is 19 more sources the check itself turned up, each marked consistent or not.

5 sources were applied directly, 7 informed a mechanism that shipped, 7 have no traceable use, 13 shaped a plan never built, 6 were checked only afterwards, and 19 more were found later.

Data table

Where 57 sources stand against the code
	sources
Applied	5
Informed	7
No use	7
Plan only	13
Checked	6
Found later	19

Source: Our source check, November 2026. The first five bars are the 38 sources cited in our own design notes: applied directly in the code; informed a mechanism that shipped; listed with no traceable use; shaped a plan that was never built; and checked against the code only after it was written.

Our ants steer by the same kind of rule measured in real ants

Tap the page with the ants and a scout walks to that spot, laying a scent nobody can see. The ants behind it feel for the scent with their antennae and turn toward the side with more.

How much they turn is the clever part. The game divides the difference between left and right by the total the ant can smell, so a faint trail and a strong one steer an ant equally well. Researchers who studied Argentine ants found turns of the same form: the difference between the two sides divided by their sum ^{Perna 2012}. That paper is in our design notes, but the code doesn't name it, the note asked only for a "probe difference", and the species differ. Dividing by the total is also a common way to steady a steering rule. So we call it a match, not an implementation. Like the ants in the paper, ours keep the same speed on the trail.

The scent fades. In black garden ants, trail scent lasts on average about 47 minutes ^{Beckers 1993}. Ours fades about 63 times faster, so a trail nobody walks is half gone in about half a minute. A child can see it go within one sitting. The same paper found ants lay 43 % more trail from a rich sugar source than a weak one, which in the authors' model is enough for a colony to settle on the richer food. Our ants don't: an ant lays the same whatever it found.

Where they lay it differs too. Real black garden ants lay about equally on the way to food and on the way back, and only once food has been found ^{Beckers 1992}. Ours lay several times more coming home with a crumb, and the scout lays on its way out. On screen the homeward trail stands out, which reads well. It isn't how real garden ants do it, and we don't claim it is.

The most playful part has no paper behind it. Anything lying on a cell wipes the scent there, whoever put it there: water, soap, acid, bleach, flour or baking soda. No source we found says baking soda keeps real ants away. In the game, fine powder wipes the trail, and that's all we claim.

Our ants steer by the same kind of rule measured in real ants

Under the hood

Our ants steer by the same kind of rule measured in real ants

The scent is one shared layer on the paper, coarser than the paper itself, with a ceiling on how strong a spot can get. It fades exponentially, the same shape as the paper's mean lifetime, only far quicker.

Three feelers reach ahead of each ant, one to each side and one straight on:

turn ∝ (s_L − s_R) / (s_L + s_C + s_R)

The centre feeler only adds to the total, which the paper's two-sided rule doesn't have. Below a faint threshold an ant counts as off the trail. A little random wander is added, damped while the ant is on the trail, and turning has a speed limit. Ants keep one walking speed.

Who lays	game	Beckers 1992
scout, going out	lays	none before food
going out	lays a little	about equal both ways
coming home laden	lays most	about equal both ways

Laying depends only on whether the ant carries something, is a scout or is lost, never on the food. A comment in the code says an ant "keeps count of its own steps", as desert ants do ^{Wittlinger 2006}; the code doesn't count, and homing is a pull toward the nest. Pinned by tests. Live.

Ants following a loop of scent, steered by the difference between their left and right feelers — Try it: faint trail, strong trail The game's own rule
Fourteen ants walk a loop of scent (the green haze, which the game never draws). Drag on the page to lay new scent, or press **Lay a line of scent**. Then turn the scent each ant lays right down. **What to look for:** with the game's rule, difference ÷ total, the ants hold the loop on a faint trail as well as a strong one. Switch to *difference only* and a strong trail still works, but on a faint one (about ×0.1 and below) the ants no longer turn enough at the bends and wander off. Switch the fade to real ants' 47 minutes and the trail hardly fades on this timescale. These are the game's feelers, turning, wander, speed, laying and fade on a 160 × 120 cell page, with every ant laying at the game's going-out rate; the game's pull toward a goal and its pause to feel about for a lost trail are left out, so a loop needs no goal.

Scent left on a trail nobody walks

The game's scent is down to 26 % after one minute and under 1 % after six; a real trail with a 47-minute mean lifetime still holds 81 % after ten minutes.

game
real black garden ants: mean lifetime 47 min

Data table

Scent left on a trail nobody walks
minutes	game	real black garden ants: mean lifetime 47 min
0	100	100
0.25	71.65	99.47
0.5	51.34	98.94
0.75	36.79	98.42
1	26.36	97.89
1.25	18.89	97.38
1.5	13.53	96.86
1.75	9.7	96.35
2	6.95	95.83
2.25	4.98	95.33
2.5	3.57	94.82
2.75	2.56	94.32
3	1.83	93.82
3.25	1.31	93.32
3.5	0.94	92.82
3.75	0.67	92.33
4	0.48	91.84
4.25	0.35	91.35
4.5	0.25	90.87
4.75	0.18	90.39
5	0.13	89.91
5.25	0.09	89.43
5.5	0.07	88.96
5.75	0.05	88.48
6	0.03	88.02
6.25	0.02	87.55
6.5	0.02	87.08
6.75	0.01	86.62
7	0.01	86.16
7.25	0.01	85.71
7.5	0	85.25
7.75	0	84.8
8	0	84.35
8.25	0	83.9
8.5	0	83.46
8.75	0	83.01
9	0	82.57
9.25	0	82.13
9.5	0	81.7
9.75	0	81.27
10	0	80.83

Source: Game: the ant scent as the released game fades it. Real: the mean lifetime estimated by ^{Beckers 1993}, drawn the same way.

Startled animals flee straight away, as filmed flies do

Bring a hand down near a fly and it darts off, directly away. Flies filmed at take-off do the same: they jump away from a looming threat, setting the direction of the jump in the fifth of a second before they leave ^{Card 2008}. Our fly matches that paper on direction and nothing more. It has no posture before the jump, no delay, and no sideways bias. Since October 2026 "flee straight away" is one rule shared by every animal in the game, so the fly has nothing of its own here.

A hand that creeps up slowly gets close. That's our own design: slow movement counts for far less than fast movement. Our design note put "slow approaches don't trigger it" right after citing the fly paper, but the paper's experiments didn't test slow approaches, so we don't claim it came from there.

One pigeon taking off sends up the pigeon beside it. Its wings write noise and movement into the layers every animal reads, and the other pigeon hears and sees them with its own senses, and an automated test checks it. The idea came from crested pigeons, whose wings whistle differently in alarm; flock mates took off when they heard the alarm whistle and not the ordinary one ^{Hingee 2009}. The paper's point is that the pattern of the sound carries the alarm. Ours has no calm take-off and no alarmed one. It's just something loud and sudden nearby. The authors only suggest that take-off noise might warn other flocking birds too.

A frightened pigeon drops a few feathers, more after a bigger fright. That touch came from a page on fright moult, which names diamond doves as prone to it ^{Birds Online}. It's a hobby site, about doves rather than pigeons, and we treat the feathers as the small touch they are.

And nothing ever lands on a fly. That isn't physics: it's our rule that no animal is ever hurt, and a fly under a swat is moved out from under it.

The cat and dog borrow a robot's footwork and a vet's food list

Watch the cat run and each paw reaches forward before it lands. The rule comes from legged robots: put the foot down at the "neutral point", and a running machine keeps its speed steady ^{Raibert 1986}. Our design note writes that point as half the distance the body travels while the foot is on the ground. Our pets don't balance. Their bodies move by a steering rule, and the feet follow, so the robot's footwork is used only for how the walk looks.

A wet dog shakes itself dry. Ours rocks 7.6 times a second for 1.2 seconds, about 13° each way. A study that filmed wet mammals found dogs shaking at 4.5 to 8 times a second, from Labradors at 4.3 to 4.6 up to a 2.5 kg Chihuahua at 6.8 ^{Dickerson 2012}. A Labrador swings about 90°. So our shake sits inside the range for dogs, faster than any dog the study filmed, and much smaller. Each drop it throws off carries a share of whatever is on the dog's coat; the study says nothing about which way drops fly, so that part is ours.

Shine the laser near the cat and it chases the dot. Switch the laser off under its paw and the cat has caught it, and is very pleased. The puppy sees the dot and doesn't chase it. A survey of 618 cat owners, which we found after building this, linked frequent laser play with cats that keep chasing lights and shadows, and mentions the suggestion that never catching the prey may frustrate them ^{Kogan 2021}. Letting the cat catch it is consistent with that. The survey is about cats, and it shows a link, not a cause.

Offer the puppy chocolate and it sniffs, sits drooling, looks at you and leaves it, and the game says once why. Chocolate, caffeine, grapes and xylitol are tagged as bad for dogs. A veterinary manual, read after the game was built, agrees on all four ^{Merck: chocolate, Merck: grapes, Merck: xylitol}. Passion fruit is tagged too, and for that we found only pet-care blogs, so we make no claim for it.

The dog's tail wag leans a little to one side, the same whatever it sees. Real dogs wag more to one side or the other depending on what they see ^{Quaranta 2007}, so ours has the lean and not the feeling, and we claim nothing for it.

The cat and dog borrow a robot's footwork and a vet's food list

Under the hood

The cat and dog borrow a robot's footwork and a vet's food list

Foot target, the "neutral point" of legged robotics as Raibert describes it:

x_foot = x_home + v · T_stance / 2

No test checks the rule itself; a test runs the cat and looks for skids.

Shake: a sideways rock that eases out at the end, as measured on the released game:

	shake	swing
game dog	7.6 Hz	±13°
Labrador, 26–41 kg	4.3–4.6 Hz	about 90°
Chihuahua, 2.5 kg	6.8 Hz

Food: chocolate, caffeine, grape, xylitol and passion fruit carry a "bad for dogs" tag on the material itself, so anything that carries the tag is left alone. Pinned by a test. Wag: a small fixed lean. All live; the dog is in the full toolbox.

Seen from above: the puppy sitting with its nose a short way from a bar of milk chocolate on the paper, not eating it — **The puppy leaves the chocolate alone.** The released game: it walked over to the bar, sniffed and sat down a nose-length away. Chocolate carries a tag, and the dog's food rule reads tags, not a list of foods.

One real number, paper's 230 °C, anchors every temperature

The game keeps one temperature number for every spot on the paper, where 0 is room temperature. To turn that into degrees, it needs a fixed point, and the fixed point is paper. The scale is set so that the point where the game's paper catches is 230 °C. Published figures for paper catching by itself run from 218 to 246 °C ^{Autoignition, Wikipedia}, which we found after the fact. Every other threshold sits on the same scale.

Then the kitchen. Chocolate softens at 32 °C. Cocoa butter can crystallise in six ways, and ordinary chocolate is normally in the fifth ^{Wille 1966}; the game has one softening point. Baker's yeast in the game works best at 35 °C, suffers above 50 and dies at 55. A published measurement puts its best growth at 32.3 °C and its limit at 45.4 ^{Salvadó 2011}. Ours is a little warmer, and it models how fast yeast makes gas, not how fast it grows. Egg white sets at 63 °C and yolk at 70, each at a rate. Real yolk can start to set from 56 °C, and how firm it gets depends on time and temperature together ^{Vega 2011}, so a single setting point is our simplification.

One real number, paper's 230 °C, anchors every temperature

Under the hood

One real number, paper's 230 °C, anchors every temperature

One scale: every spot on the paper keeps a single unitless heat number, mapped linearly to degrees, with room temperature at zero and paper's catching point pinned to 230 °C.

material, °C	flame	by itself	published
cooking oil	340	420	406–435 by itself
butter	330	400	never by itself
candle wax	250	300	flash point 199

Between the two points a pool needs a flame on it or beside it. Pinned by a test.

Flash boil: when water lands under something very hot, the share of it that turns to steam is worked out from water's real latent heat, 2,260 J/g, and the spot cools by exactly the heat the steam took. How hard the oil is thrown, the temperature where flashing starts and how much heat a spot of liquid holds are tuned for play.

Ink make-up, next update: each ink's browning strength comes from its sugar, protein and acid per 100 g. The sugar and acid figures come from nutrition tables; milk's fat figure has no source in our notes. Pinned by a test.

Yeast works best at 35 °C, suffers above 50 and dies at 55; egg white turns at 63 °C and yolk at 70; chocolate softens at 32 °C.

A temperature ladder: ten kitchen things with the game’s thresholds and the published figures, and what each does at the chosen heat — Try it: turn up the heat The game's own rule
Slide the heat and watch what changes. Red ticks are the game's thresholds; yellow marks are the published figures we checked them against. **What to look for:** paper catches inside the published band; oil catches by itself inside its band, but between 340 and 420 °C only with a flame beside it; butter catches at 400 °C where the one measurement says it never did; secret ink moves from 80 to 105 °C in the next update. These are the game's thresholds on its own temperature scale; the heating here eases toward the slider in about a second. The open ring on the butter row: one measurement, never caught by itself.

Fats are where the checking bites. Cooking oil in the game lights from a flame at 340 °C and by itself at 420 °C. One study measured cooking oils catching by themselves between 406 and 435 °C ^{Buda-Ortins 2011}, and our 420 sits inside that. The same study found that butter didn't catch by itself at all, while ours does at 400 °C. It's a student study, a useful check rather than an authority, but it's the only measurement we found, and it disagrees with us. Candle wax lights from a flame at 250 °C, above the 199 °C flash point on a safety card ^{ICSC 1457}, which is the right way round. Its 300 °C for catching by itself we couldn't check.

Some real constants go straight in. Pour water onto burning oil and it flashes to steam under the oil and throws it up. How much water flashes is worked out from water's real latent heat, typed in as 2,260 joules per gram. A reference table we checked it against later gives 2,256 ^{Engineering Toolbox}. In the next update next update wet paper also boils dry by latent heat, held near 100 °C until the water is gone.

Secret ink is the clearest case of a source changing the code. Paint lemon juice on paper, heat it, and brown writing appears. In the released game that starts at 80 °C, a number with no source. In the next update next update it starts at 105 °C, the temperature at which fructose, the earliest of the common sugars, begins to caramelise; glucose needs 150 and table sugar 170 ^{Caramelization, Wikipedia}. Those figures come from a standard food-science book ^{McGee 2004}, which we know only through that page. Nobody has measured the point for lemon juice on paper, and we say so. The next update also gives each ink a make-up per 100 g taken mostly from nutrition tables, so milk, with more sugar and protein, browns darker than lemon next update.

Red cabbage follows the printed charts, and baking soda was too strong

Paint red cabbage juice on the paper and drip vinegar or baking soda on it. The juice goes pink where it's acid and blue to green where it's alkaline. The game estimates how acid each spot is from what's dissolved there, then looks the colour up on a table of eight colours from pH 1 to 13. It isn't a calculation of molecules, and the pH is a game estimate.

Two published colour charts for home and school experiments, found after the table was made, agree with it on the order: red in strong acid, then pink, purple, blue, green, and yellow at the far alkaline end ^{RHS chart, Illinois chart}. The shades are our own picks. Our pH 5 is pinker than either chart's colour there, and the two charts don't agree with each other either: one calls pH 4 violet and pH 6 purple, the other the other way round.

Baking soda is where the check found a real error. Baking soda in water sits at about pH 8.3 ^{Science Insights}. In the released game, enough of it reads pH 10, which turns cabbage teal-green. In the next update next update it stops at 8.3, and cabbage on baking soda goes blue, as it does in a kitchen.

Alkali also makes food brown faster. That's why pretzels are dipped in lye ^{Maillard reaction, Wikipedia}. The game speeds browning up by a fixed factor for every pH unit the paper under the food is more alkaline than the food itself, a factor we estimated; the page gave us the direction, not the number.

Red cabbage follows the printed charts, and baking soda was too strong

Under the hood

Red cabbage follows the printed charts, and baking soda was too strong

Eight hand-picked colours across the pH range. The colours are used as absorbance, the way a dye darkens paper, so the more juice lies on a spot, the deeper its colour.

Baking soda's share of the pH rise has a ceiling:

	reads up to	real baking soda
released	pH 10	about pH 8.3
next update	pH 8.3	about pH 8.3

Browning by pH grows exponentially with how much more alkaline the paper is than the food, within limits, and counts only beyond the food's own pH, so ketchup's own acid doesn't slow ketchup. The factor per pH unit is our estimate. Live.

The game’s red cabbage colours from pH 0 to 14, with the colour words of two printed charts underneath — Try it: dip a strip in cabbage juice The game's own rule
Drag along the strip or use the slider to pick a pH. The big swatch is the game's colour; underneath are the colour words from the two printed charts. **What to look for:** the order matches everywhere; the game's pH 5 is a clear pink; and baking soda lands on teal-green in the released game, on blue in the next update, where real baking soda sits. This is the game's colour table, drawn at full strength on white.

A sheet dipped in red cabbage juice, purple all over: a pink stripe where a line of vinegar drops was, and a blue band round a white line of baking soda wetted with water — **Vinegar pink, baking soda blue.** next update A line of vinegar drops (left) and a line of sifted baking soda (right), the sheet dipped in cabbage juice, then water dropped on the soda to dissolve it. The blue is the next update's baking soda; in the released game the same line turns teal-green. Nothing in the vinegar or the soda knows about cabbage.

Ant trails fade and snails crawl faster than in real life, so a child can watch them

A child plays for minutes, not hours, on a page smaller than a placemat. So some things are sped up on purpose. The ants' trail fades about 63 times faster than a real one. A real snail crawls about a millimetre a second; ours was set to three times that, so you can see it move, and comes out at about 3.6 mm a second.

Others ended up faster without anyone aiming. Fire creeps across the released game's paper at about 1.1 mm a second. One published measurement of plain paper burning downward is 0.81 mm a second ^{Ramli 2018}: the same order of size, though we never tuned toward it, and a flat sheet isn't a downward burn. The dog's shake is about 1.7 times a Labrador's.

How many times faster than life four things runOnly the first two were sped up on purpose.

Ant scent fades 62.7 times faster than in life, the snail crawls 3.6 times as fast, the dog shakes 1.71 times as fast and paper burns 1.36 times as fast.

Data table

How many times faster than life four things run
	game ÷ real
ant scent	62.7×
snail	3.6×
dog shake	1.71×
paper fire	1.36×

Source: Ant scent: the game’s fade against a 47-minute mean lifetime ^{Beckers 1993}. Snail: about 3.6 mm/s against about 1 mm/s (a popular page ^{Scienceinsights snail}). Dog: 7.6 Hz against a Labrador’s 4.3–4.6 Hz ^{Dickerson 2012}. Fire: about 1.1 mm/s, measured on the released game for our fact sheet, against 0.81 mm/s for paper burning downward ^{Ramli 2018}.

57

Sources traced to the code they shaped, or to the code they didn't: 38 from our design notes, 19 found in a later check.

5

Sources applied directly in the code: the robot's foot rule, the caramel point and the book behind it, and two nutrition tables. All but the first reach players in the next update.

63×

How much faster our ant trail fades than a real one, which lasts about 47 minutes.

230 °C

Paper's catching point, inside the published 218–246 °C. It fixes the game's whole temperature scale.

3 of 107

Facts in our own fact sheet rated "measured", its strongest grade. All three are electricity. The other 104 are rated modelled, simplified or not yet known.

2,260 J/g

Water's real latent heat, typed into the code. It decides how much water flashes to steam under burning oil.

Some research we read, and chose not to build

Thirteen of the sources in our notes shaped plans that never reached the game. Some were overtaken by a later rule, such as "no animal is ever hurt" or "animals sense the world through shared layers, never through a pairing of tool and animal". Others simply haven't been built yet. We list them so nobody mistakes the plan for the game.

The best example is the pigeon's head. Real pigeons walk with a hold and a thrust: the head stays still relative to the ground for part of each step, then shoots forward. At a run the hold disappears ^{Davies 1988}. Our plan said to replace the game's smooth bob with that curve. The code still draws a sine wave, tied to the step, and it doesn't fade at speed. A study of another bird even found head and legs out of step ^{Hancock 2014}, which argues against tying the bob to the step at all.

The pigeon's head: described, and drawnSketch, not data. Dashed: the body moving steadily. Red: the game’s bob, a smooth wave tied to the step, drawn larger here to be seen.

Over two steps the described head holds still, then shoots forward; the game's head rides a smooth wave around the steadily moving body.

the body moving steadily
hold and thrust, as described
the game's bob

Source: Blue: the hold and thrust described by ^{Davies 1988} and ^{Necker 2007}; the share of each step spent holding is drawn by eye.

The fly's planned jump

About 200 ms of posture before take-off, and a jump halfway between away and forward when the threat comes from the side ^{Card 2008}. Ours leaves at once, straight away.

The fly's grooming order

Eyes, then antennae, abdomen, wings and thorax, each cleaning programme suppressing the next ^{Seeds 2014}. Ours picks a grooming bout at random.

More trail to richer food

Garden ants lay more trail from a rich sugar source ^{Beckers 1993}. Ours lay the same whatever they found.

Pigeons drinking and courting

Pigeons drink by suction without lifting their heads, and court by bowing and wing clapping ^{St John 2010, PigeonWatch}. Our pigeon pecks, walks and flies.

What we don't claim, even where it would sound good

That children learn from it. We have no study of our own showing what children learn from the game. This page says what the design draws on, not what it proves.
That baking soda keeps ants away. In the game, fine powder wipes the scent. No source we found says it repels real ants.
That it's fire-safety advice. The game shows baking soda and a damp towel putting out burning oil, and water making it worse. A fire-service leaflet agrees on water and flour ^MSFCA, but steers away from baking soda, calling it an old practice. It's a game, not guidance.
That our ants lay trail as real ones do. Real garden ants lay about equally both ways, and nothing before food. Ours lay most on the way home, and the scout lays going out.
That pigeons in general alarm the flock. The study was on crested pigeons and their whistle. Ours just make a loud, sudden noise.
That every ignition point is real. Paper's and oil's are defensible. Butter's is contradicted by the one measurement we found, and wax's we couldn't check.

Baking soda stops an ant trail and water makes burning oil spit, from facts every system reads

None of these was written as a pairing. A real figure goes into one shared layer, and whatever reads that layer is affected.

Shared: fine dust on the cell

Sift baking soda across an ant trail

No ant crosses, and the scent under the line is gone.

Animals with their own minds · pinned by a test

Shared: noise + motion layers

Bang the hammer near two pigeons

The near one flies, and its wings send up the other.

Sound and shaking · pinned by a test

Shared: latent heat, 2,260 J/g

Drip water onto burning oil

Steam throws the oil up, and burning drops start new fires where they land.

Heat and fire · pinned by a test

Shared: dust in the flame

Sift flour onto burning oil

It flares up, as a fire-service leaflet warns.

Heat and fire · pinned by a test

Shared: carbonate in the paper → pH

Burn a patch, then paint cabbage juice on it

The juice turns blue in a thin ring round the ash. No cabbage code in the fire.

Chemistry and pH · pinned by a test

Shared: material tags

Put chocolate in the puppy's way

It sniffs, sits, leaves it, and the game says why, once.

Animals with their own minds · pinned by a test

Where we simplify, and why

Every row below starts from a real finding and ends at a number a tablet can run 60 times a second. The second column is what we compute; the third is what that costs.

In real life. Garden ant trail scent lasts about 47 minutes on average ^{Beckers 1993}

In the game. One shared scent layer that fades in under a minute and has a ceiling

Why that's enough. A trail must clear within one sitting. Lost: real persistence; wind doesn't move the scent

In real life. Argentine ants turn by (left − right) ÷ (left + right) ^{Perna 2012}

In the game. The same form with a centre feeler in the total, 7 rad/s, plus wander

Why that's enough. Steady on faint and strong trails alike. Lost: the species' own noise and detail

In real life. A fly plans its jump for about 200 ms, its direction set by where the threat is ^{Card 2008}

In the game. Every animal flees straight away from the alarm, at once

Why that's enough. The direction reads right. Lost: posture, delay and the sideways bias

In real life. Crested pigeons' wings whistle a distinct alarm ^{Hingee 2009}

In the game. Any take-off writes noise and movement into the shared layers

Why that's enough. One take-off sets off the pigeon beside it. Lost: any difference between a calm take-off and a frightened one

In real life. Heat is energy, with real heat capacities for paper, water and oil

In the game. One unitless heat number per spot, anchored on paper catching at 230 °C

Why that's enough. One scale serves everything. Lost: real heat capacities; water warms far too easily

In real life. Yolk sets by time and temperature together, from about 56 °C ^{Vega 2011}

In the game. Egg white turns at 63 °C and yolk at 70 °C, each at a rate

Why that's enough. Fried eggs look right. Lost: the soft-boiled subtleties

In real life. Cabbage dye is several molecules, each changing with pH

In the game. Eight hand-picked colours on a pH estimate

Why that's enough. The order matches printed charts. Lost: exact shades; our pH 5 is too pink

In real life. Wet dogs shake at 4.3–6.8 Hz, swinging up to about 90° ^{Dickerson 2012}

In the game. 7.6 Hz, ±13°, for 1.2 s

Why that's enough. It reads as a shake on a toy-sized dog. Lost: true frequency and swing

Every source, with what it found and what the game did with it

Each entry says what the source found and what became of it in the game. Other chapters link here. Status tags show how each source relates to the code.

Applied: used directly in the code
Informed: shaped a mechanism that shipped
Plan only: shaped a plan that wasn't built
No use: listed in our notes, no traceable use
Found later: found after the code; consistent, or, where the tag says so, not consistent
Not used yet: well-known work we haven't drawn on yet

Ants

Beckers R, Deneubourg J-L, Goss S (1993). Modulation of trail laying in the ant Lasius niger and its role in the collective selection of a food source. Journal of Insect Behavior 6: 751–759.
doi:10.1007/BF01201674Informed
Mean trail pheromone lifetime about 47 minutes; 43 % more trail from a rich source. Our scent fades in under a minute on purpose; the richer-food part isn't built.
Perna A, Granovskiy B, Garnier S, Nicolis SC, Labédan M, Theraulaz G, Fourcassié V, Sumpter DJT (2012). Individual rules for trail pattern formation in Argentine ants (Linepithema humile). PLoS Computational Biology 8: e1002592.
doi:10.1371/journal.pcbi.1002592
In our notes · same form. Turns follow a Weber-type rule, difference over sum; speed largely unaffected. Our steering has the same form; nothing in the code names the paper.
Amorim P (2014). A continuous model of ant foraging with pheromones and trail formation.
arxiv.org/abs/1402.5611
No use. Ants as a continuous population. The game does the opposite: individual ants on a scent grid.
Mathigon, Graphs and Networks: Ants.
mathigon.org/step/graph-theory/ants-1
No use. A popular explainer of laying, following, reinforcing and fading. Background only.
Beckers R, Deneubourg J-L, Goss S (1992). Trail laying behaviour during food recruitment in the ant Lasius niger. Insectes Sociaux 39: 59–72.
doi:10.1007/BF01240531
Found later · not consistent. Laying starts only once food is found, then about equally both ways. Ours lays several times more coming home, and the scout lays going out.
Wittlinger M, Wehner R, Wolf H (2006). The ant odometer: stepping on stilts and stumps. Science 312: 1965–1967.
doi:10.1126/science.1126912Found later
Found later · matches a comment. Desert ants count their steps. A comment in our code says so; the code doesn't count.

Flies

Card G, Dickinson MH (2008). Visually mediated motor planning in the escape response of Drosophila. Current Biology 18: 1300–1307.
doi:10.1016/j.cub.2008.07.094Informed
Flies set their escape direction about 200 ms before take-off and jump away from a looming threat; mean delay 215 ± 42 ms. Ours matches on direction only.
Fotowat H, Fayyazuddin A, Bellen HJ, Gabbiani F (2009). A novel neuronal pathway for visually guided escape in Drosophila melanogaster. Journal of Neurophysiology 102: 875–885.
doi:10.1152/jn.00073.2009
No use. An escape pathway that fires when an approaching object reaches a threshold angular size.
Klapoetke NC, Nern A, Peek MY, Rogers EM, Breads P, Rubin GM, Reiser MB, Card GM (2017). Ultra-selective looming detection from radial motion opponency. Nature 551: 237–241.
doi:10.1038/nature24626
No use. Neurons that respond to things expanding in view, not shrinking or turning.
Seeds AM, Ravbar P, Chung P, Hampel S, Midgley FM, Mensh BD, Simpson JH (2014). A suppression hierarchy among competing motor programs drives sequential grooming in Drosophila. eLife 3: e02951.
doi:10.7554/eLife.02951
Plan only. Eyes, antennae, abdomen, wings, thorax, by suppression. Not built: our fly picks a bout at random.
Mueller JM, Ravbar P, Simpson JH, Carlson JM (2019). Drosophila melanogaster grooming possesses syntax with distinct rules at different temporal scales. PLoS Computational Biology 15: e1007105.
doi:10.1371/journal.pcbi.1007105
No use. Grooming moves from front to back, probabilistically.
Zhang N, Simpson JH (2022). A pair of commissural command neurons induces Drosophila wing grooming. iScience 25: 103792.
doi:10.1016/j.isci.2022.103792
No use. Flies clean one wing at a time with both hind legs. Our fly does put its hind legs over its wings.

Pigeons, a snail and sparrows

Hingee M, Magrath RD (2009). Flights of fear: a mechanical wing whistle sounds the alarm in a flocking bird. Proceedings of the Royal Society B 276: 4173–4179.
doi:10.1098/rspb.2009.1110Informed
Crested pigeons took off to alarm whistles only; the whistle's tempo mattered. Ours: take-off is loud and sudden, by analogy.
phys.org (2009). Press story on the crested pigeon's whistle.
phys.org/news/2009-09-tweet.htmlInformed
Coverage of the paper above.
Birds Online. Fright moult.
birds-online.de/wp/en/birds-online-english/health-and-diseases/plumage-disorders-and-molt/moult/fright-moultInformed
A hobby site naming diamond doves as prone to fright moult; it doesn't mention pigeons. Our pigeon drops a few feathers when startled, more after a bigger fright.
Davies MNO, Green PR (1988). Head-bobbing during walking, running and flying: relative motion perception in the pigeon. Journal of Experimental Biology 138: 71–91.
doi:10.1242/jeb.138.1.71
Plan only. Hold and thrust when walking; no hold when running or flying. Not built: ours is a sine wave.
Necker R (2007). Head-bobbing of walking birds. Journal of Comparative Physiology A 193: 1177–1183. ; web version at
doi:10.1007/s00359-007-0281-3
Plan only. Head still in the hold, rapid forward in the thrust.
Jiménez Ortega L, Stoppa K, Güntürkün O, Troje NF (2009). Vision during head bobbing: are pigeons capable of shape discrimination during the thrust phase? Experimental Brain Research 199: 313–321.
doi:10.1007/s00221-009-1891-5
Plan only. Pigeons tell shapes apart as well in the thrust as in the hold.
Theunissen LM, Troje NF (2017). Head stabilization in the pigeon: role of vision to correct for translational and rotational disturbances. Frontiers in Neuroscience 11: 551.
doi:10.3389/fnins.2017.00551
Plan only. On a moving perch the head is held almost perfectly against rotation.
Hancock JA, Stevens NJ, Biknevicius AR (2014). Elegant-crested Tinamous Eudromia elegans do not synchronize head and leg movements during head-bobbing. Ibis 156: 198–208.
doi:10.1111/ibi.12115
Plan only. Head and legs out of step in a tinamou. Our bob is tied to the step.
Birdorable (2025). Head-bobbing explained.
birdorable.com/blog/head-bobbing-explained-how-pigeons-and-other-birds-stabilize-vision
Plan only. A popular blog post.
St John K (2010). Anatomy: how birds drink.
birdsoutsidemywindow.org/?p=21961
Plan only. Doves and pigeons drink by suction without raising their heads. Not built.
St John K (2024). Pigeons conspicuously court in public.
birdsoutsidemywindow.org/?p=310226
Plan only. Bowing, circling, wing clapping. Courtship not built.
Washington State University news (2026). Dr. Universe explores how birds stay hydrated.
news.wsu.edu/news/2026/07/08/dr-universe-explores-the-surprising-ways-birds-stay-hydrated
Plan only. Pigeons and doves drink by suction.
PBS Nature. Pigeon dating rituals.
pbs.org/wnet/nature/pigeon-dating-rituals-the-strut-the-coo-and-the-kiss/33573
Plan only. Bowing and strutting.
PigeonWatch quick reference (illustrations by Julie Zickefoose).
home.csulb.edu/~rodrigue/geog442/PigeonWatch/QuickReference.pdf
Plan only. Bowing, tail-dragging, billing, wing clapping.
Harkin EL, van Dongen WFD, Herberstein ME, Elgar MA (2000). The influence of visual obstructions on the vigilance and escape behaviour of house sparrows. Australian Journal of Zoology 48: 259–263.
doi:10.1071/ZO00003
Plan only. With their view blocked, sparrows scan more and flee later. There are no sparrows in the game.
Science Insights. How fast can a snail travel?
scienceinsights.org/how-fast-can-a-snail-travel-the-science-explainedInformed
About 1 mm/s, up to 4. A weak page that cites no study. Our snail: about 3.6 mm/s, three times faster on purpose.

The cat and the dog

Raibert MH (1986). Legged Robots That Balance. MIT Press.
Applied
A foot placed at the neutral point keeps a running machine's speed unchanged. Our pets' feet land at home + v·Tstance/2, for animation only.
Dickerson AK, Mills ZG, Hu DL (2012). Wet mammals shake at tuned frequencies to dry. Journal of the Royal Society Interface 9: 3208–3218.
doi:10.1098/rsif.2012.0429
Found later · partly consistent. Dogs at 4.5–8 Hz; Labradors 4.3–4.6, a Chihuahua 6.8; Labradors swing about 90°. Ours: 7.6 Hz, ±13°.
Kogan LR, Grigg EK (2021). Laser light pointers for use in companion cat play: association with guardian-reported abnormal repetitive behaviors. Animals 11: 2178.
doi:10.3390/ani11082178Found later
A survey of 618 cat owners linking frequent laser play with repetitive behaviours. Correlational. Our cat gets to catch the dot.
Quaranta A, Siniscalchi M, Vallortigara G (2007). Asymmetric tail-wagging responses by dogs to different emotive stimuli. Current Biology 17: R199–R201.
doi:10.1016/j.cub.2007.02.008
Found later · partly consistent. Wagging leans one way toward the owner and the other toward a strange dog (as reported in Nature's news coverage). Ours has a small fixed lean, whatever the dog sees.
Merck Veterinary Manual. Chocolate toxicosis in animals.
merckvetmanual.com/toxicology/food-hazards/chocolate-toxicosis-in-animalsFound later
Theobromine and caffeine are the toxic principles; dogs are affected most.
Merck Veterinary Manual. Xylitol toxicosis in dogs.
merckvetmanual.com/toxicology/food-hazards/xylitol-toxicosis-in-dogsFound later
A rapid insulin release in dogs that can cause profound low blood sugar.
Merck Veterinary Manual. Grape, raisin, and tamarind toxicosis in dogs.
merckvetmanual.com/toxicology/food-hazards/grape-raisin-and-tamarind-vitis-spp-tamarindus-spp-toxicosis-in-dogsFound later
Kidney failure in some dogs.

Kitchen science and fire

Wikipedia. Caramelization.
en.wikipedia.org/wiki/CaramelizationApplied
Fructose 105 °C, glucose 150, sucrose 170. Secret ink shows from 105 °C in the next update; 80 °C, unsourced, in the released game.
McGee H (2004). On Food and Cooking, 2nd ed. Scribner.
Applied
Applied, second-hand. The source of the caramel figures above, reached through that page. We haven't read the book itself.
Aprifel. Lemon nutrition sheet (Ciqual 2020 data).
aprifel.com/en/nutritional-sheet/lemonApplied
5.9 g citric acid per 100 g of fruit. Used for the lemon ink's make-up in the next update.
SoupersAge. Milk vs lemon juice (USDA data).
soupersage.com/compare-nutrition/milk-vs-lemon-juiceApplied
Milk 5.1 g sugar and 3.3 g protein, lemon juice 2.5 g sugar, per 100 g. Used for both inks' make-up in the next update.
Wikipedia. Maillard reaction.
en.wikipedia.org/wiki/Maillard_reactionInformed
Faster when alkaline, with pretzels as the example. How much faster per pH unit is our own estimate.
Wikipedia. Autoignition temperature.
en.wikipedia.org/wiki/Autoignition_temperatureFound later
Paper 218–246 °C (the page's own reference for it is a dead link). Ours: 230 °C.
Ramli A, Razali MA, Madon RH, Hushim MF, Khalid A, Ja'at MNM, Salleh H (2018). Influence of material composition on flame spread behaviour over combustible solid of paper/bagasse. International Journal of Integrated Engineering 10(8).
journal.uthm.edu.my/index.php/ijie/article/view/2034Found later
Pure paper, downward spread: 0.81 mm/s. The released game's front: about 1.1 mm/s, never tuned toward it.
Avinash G, Kumar A, Raghavan V (2014). Experimental study of effect of angular orientation on flame spread over thin solid fuels. 10th Int. Conf. on Heat Transfer, Fluid Mechanics and Thermodynamics.
repository.up.ac.za/items/2f2ddf22-cedf-4441-8c1a-2a1a0192d09bFound later
Background on thin-fuel flame spread; no values on the page.
Buda-Ortins K (2011). Auto-ignition of cooking oils. University of Maryland, undergraduate research paper.
drum.lib.umd.edu/items/97dbb870-b9cb-40be-8074-d437cf448062
Found later · oil yes, butter no. Canola 424 °C, vegetable 406, olive 435; canola flame point 379; butter did not auto-ignite. Our oil: 340 and 420. Our butter: 330 and 400.
IPCS International Chemical Safety Card 1457, paraffin wax (fume).
inchem.org/documents/icsc/icsc/eics1457.htmFound later
Found later · flash point only. Flash point 199 °C, closed cup; no autoignition figure. Our wax: 250 and 300.
Engineering Toolbox. Water properties: vaporization heat.
engineeringtoolbox.com/water-properties-d_1573.htmlFound later
2,256.4 kJ/kg at 100 °C. The code types in 2,260.
Vega C, Mercadé-Prieto R (2011). Culinary biophysics: on the nature of the 6X°C egg. Food Biophysics 6: 152–159.
doi:10.1007/s11483-010-9200-1Found later
Yolk can set from 56 °C; texture depends on time and temperature together. Nothing on egg white.
Salvadó Z, Arroyo-López FN, Guillamón JM, Salazar G, Querol A, Barrio E (2011). Temperature adaptation markedly determines evolution within the genus Saccharomyces. Applied and Environmental Microbiology 77: 2292–2302.
doi:10.1128/AEM.01861-10Found later
Baker's yeast grows best at 32.3 °C, up to 45.4. Ours works best at 35, models gas rather than growth.
Wille RL, Lutton ES (1966). Polymorphism of cocoa butter. Journal of the American Oil Chemists' Society 43: 491–496.
link.springer.com/article/10.1007/BF02641273Found later
Six crystal forms; chocolate is normally form V. Our chocolate softens at 32 °C.
Beakawi Al-Hashemi HM, Baghabra Al-Amoudi OS (2018). A review on the angle of repose of granular materials. Powder Technology 330: 397–417.
doi:10.1016/j.powtec.2018.02.003Found later
Found later · no value. The angle depends on how it's measured. We saw no value for sand, so we don't cite it for our sand's 33°.
Minnesota State Fire Chiefs Association, Public Education Committee. Recipe for kitchen safety.
lexingtonmn.gov/page/open/1209/0/MSFCA%20Recipe%20Book%20for%20Kitchen%20Safety.pdf
Found later · partly consistent. Never water ("it will explode"), never flour; put a lid on it. It calls baking soda an old practice. The game agrees on water, flour and lids; in the game baking soda works.
Royal Society of Chemistry. Invisible inks.
edu.rsc.org/resources/invisible-inks/1948.articleFound later
Some heat-activated inks are organic substances that turn brown when heated; fruit juices work. Ours is a browning rule, not an oxidation model.
Gutierrez G. S (2020). Stay-at-home science project: leave secret messages with invisible ink. Popular Science.
popsci.com/story/diy/invisible-inkFound later
Milk browns as its sugars react with proteins. On lemon it disagrees with other sources about the mechanism.
Abakcus. Invisible ink DIY card.
abakcus.com/diy/invisible-inkFound later
Calls its own figures an illustrative toy model; a hair dryer is too cool.

Red cabbage and baking soda

Royal Horticultural Society. Red cabbage indicator colour chart.
rhs.org.uk/getmedia/386dd481-7d8c-43c0-b64e-ff66bf02dbe9/Res-L-Red-Cabbage-Indicator-Colour-Chart.pdfFound later
pH 1 red, 2 pink, 4 violet, 6 purple, 8 indigo, 10 blue, 12 green, 14 yellow. Same order as ours.
Carl R. Woese Institute for Genomic Biology, University of Illinois. Red cabbage pH indicator.
igb.illinois.edu/sites/default/files/uploads/Cabbage%20pH%20indicator%20Activity_0.pdfFound later
pH 2 red, 4 purple, 6 violet, 8 blue, 10 blue-green, 12 greenish yellow.
Science Insights. Does baking soda bring pH up or down?
scienceinsights.org/does-baking-soda-bring-ph-up-or-down
Found later · not consistent. A 0.1 M solution is about pH 8.3–8.5. The released game reads up to 10; the next update stops at 8.3. A weak source for a textbook figure.

Not used yet, and worth reading

van Breugel F, Dickinson MH (2014). Plume-tracking behavior of flying Drosophila emerges from a set of distinct sensory-motor reflexes. Current Biology 24: 274–286.
doi:10.1016/j.cub.2013.12.023
Not used yet. Flies speed up and turn upwind 190 ± 75 ms after meeting a plume. Our animals head upwind while they smell something, a rule written without this paper; the game has none of its timing.
Goss S, Aron S, Deneubourg J-L, Pasteels JM (1989). Self-organized shortcuts in the Argentine ant. Naturwissenschaften 76: 579–581.
doi:10.1007/BF00462870
Not used yet. The classic shortest-path result. Whether our ants can find a shorter path has never been tested.
Dussutour A, Fourcassié V, Helbing D, Deneubourg J-L (2004). Optimal traffic organization in ants under crowded conditions. Nature 428: 70–73.
doi:10.1038/nature02345
Not used yet. To compare with how our ants meet and step aside.
Muijres FT, Elzinga MJ, Melis JM, Dickinson MH (2014). Flies evade looming targets by executing rapid visually directed banked turns. Science 344: 172–177.
doi:10.1126/science.1248955
Not used yet. Escape in flight, where Card & Dickinson studied take-off.
Reis PM, Jung S, Aristoff JM, Stocker R (2010). How cats lap: water uptake by Felis catus. Science 330: 1231–1234.
doi:10.1126/science.1195421
Not used yet. Inertia against gravity sets the lapping rate.
Troje NF, Frost BJ (2000). Head-bobbing in pigeons: how stable is the hold phase? Journal of Experimental Biology 203: 935–940.
doi:10.1242/jeb.203.5.935
Not used yet. For the day the hold and thrust is built.
Bradshaw J, Casey R, Brown S (2012). The Behaviour of the Domestic Cat, 2nd ed. CABI.
Not used yet. The standard book; it could source or correct the numbers in our cat notes.
Bekoff M (1977). Social communication in canids: evidence for the evolution of a stereotyped mammalian display. Science 197: 1097–1099.
Not used yet. The play bow, which our dog does.
Babrauskas V (2003). Ignition Handbook. Fire Science Publishers.
Not used yet. The place to look for a citable ignition point for wax.
Babrauskas V (1983). Estimating large pool fire burning rates. Fire Technology 19: 251–261.
doi:10.1007/BF02380810
Not used yet. Burning rates for pools of 0.2 m and up, far bigger than any pool on the game's paper.

About this source

Real science sets the direction, play sets the pace

We traced 57 sources to the code, and say what each one really did

Our ants steer by the same kind of rule measured in real ants

Startled animals flee straight away, as filmed flies do

The cat and dog borrow a robot's footwork and a vet's food list

One real number, paper's 230 °C, anchors every temperature

Red cabbage follows the printed charts, and baking soda was too strong

Ant trails fade and snails crawl faster than in real life, so a child can watch them

Some research we read, and chose not to build

The fly's planned jump

The fly's grooming order

More trail to richer food

Pigeons drinking and courting

Baking soda stops an ant trail and water makes burning oil spit, from facts every system reads

Sift baking soda across an ant trail

Bang the hammer near two pigeons

Drip water onto burning oil

Sift flour onto burning oil

Burn a patch, then paint cabbage juice on it

Put chocolate in the puppy's way

Where we simplify, and why

Every source, with what it found and what the game did with it

Ants

Flies

Pigeons, a snail and sparrows

The cat and the dog

Kitchen science and fire

Red cabbage and baking soda

Not used yet, and worth reading