Recurrent memory. Chemical signaling. Terrain biomes. Day-night cycles. And claws that changed everything.
Scroll to explore
Part 3 solved body diversity. Every node type earned its cost, organisms grew to 29 nodes, and the ecosystem found a stable seven-type equilibrium that held for 275,000 ticks. But Part 3 had a quiet failure hiding inside its success: mouths were doing two jobs. They ate food, and they attacked enemies. Every organism built three or four mouths because mouths were free weapons. Combat was just more eating.
Part 4 separates those two jobs and sees what happens.
CLAW: a dedicated attack node. Damage scales with node velocity: base five plus node velocity times a factor. Light mass (0.5) so claws move fast at limb tips. Expensive at 0.015 energy per tick. The velocity mechanic creates an implicit musculoskeletal synergy: muscles contract, tendons snap back, bones provide leverage, and the claw at the tip whips forward. No hardcoded bonus needed. The physics engine creates the synergy.
Mouths nerfed to 3 damage. Down from 15. Mouths eat. Claws fight. Organisms now have to decide whether they're building a forager or a predator, and the answer costs something either way.
Nonlinear tendon stiffness. Past 20% stretch, tendons get 5x stiffer. Elastic snap-back, spring-loaded transmission. Distinct from rigid bone and neural-controlled muscle. The three edge types finally mean three different things.
Bone metabolic discount. Bone edges reduce metabolic cost by up to 15%, scaled by bone edge ratio. The skeleton pays for itself directly, not through secondary mechanics.
And beyond the combat redesign: recurrent neural networks with persistent hidden state across ticks, eight memory registers, two neuromodulators. Chemical signaling through SIGNAL nodes that broadcast pheromone vectors. Four terrain biomes that shape where food grows and how fast organisms move. A day-night cycle that drops sensor range to 40% at midnight. Toxic food. Drifting hazard zones. Organism collisions with soft repulsion forces. Two new node types: STOMACH for digestion efficiency and SIGNAL for chemical broadcast.
The world is larger, 4000 by 2250 units, double the footprint of Part 3. The population cap is higher. The plan was to run one million ticks.
The simulation seeds with sixty organisms. Each has five nodes, random neural weights, and no memory of anything. The world is 4000 by 2250 units, nearly nine million square units of terrain, and there are sixty of them. They are specks.
The first hundred ticks are pure noise. Neural networks fire at random inputs, bodies flail in directions that make no physical sense, and the organisms that happen to wander toward food survive while the rest burn through their starting energy and die. By tick 100, the population has grown to 97 as the first generation reproduces. By tick 500 it reaches 118, holding there as reproduction and mortality roughly balance. Bodies still average 5.1 nodes. Nobody has figured anything out yet.
The first meaningful selection happens slowly and then all at once. The organisms that survive long enough to reproduce pass on whatever neural weights led to eating. Over thousands of ticks, sensing and foraging behavior crystallizes. By tick 5,000, the population reaches 250 and bodies average 6.2 nodes — organisms are adding a sixth node as the extra metabolic load becomes affordable once feeding behavior works reliably. Then the carrying capacity asserts itself: tick 10,000 crashes the population to 99. The organisms that overfilled the world die. The 99 that remain average 7.2 nodes. Larger, more efficient, better fed. The crash is the first selection event that matters.
This is how Part 4 begins: not with a dramatic explosion but with a slow sorting. The world is large and nearly empty. The food is scattered across four terrain biomes with different density and accessibility. Toxic food items are invisible until an organism bites them. Day-night cycles drop sensor range to 40% at midnight. Everything about the environment is harder than Part 3, and the first 10,000 ticks are spent just figuring out how to survive in it.
Genesis — Tick 0–2,000
The first 2,000 ticks: 60 organisms seeded into a 4000×2250 world. Bodies average 5 nodes. Neural networks are random. Watch the initial scatter as organisms with no learned behavior wander at random, then the slow consolidation as the first viable foraging patterns emerge and early clusters form around food patches.
The early simulation looks worse than Part 3. Much worse. Bodies average 8 to 10 nodes instead of 15. The population cycles between 120 and 250 instead of holding near 500. Claws, the new expensive node type, appear on fewer than 10% of organisms and nearly go extinct twice before tick 90,000.
This is what the mouth nerf actually means. In Part 3, organisms funded growth with combat income: kill something, absorb 70% of its energy, fund three new nodes. Mouths dealt 15 damage. A well-built organism could sustain itself partly by fighting. In Part 4, a mouth deals 3 damage. The math no longer works. The only reliable income is eating food, and eating food is hard when base digestion efficiency is 60% and toxic items are invisible until you bite them.
The ecosystem responds to austerity with efficiency. Stomach nodes, which improve digestion efficiency by 10% each, spread to universal adoption within 20,000 ticks. Every organism has at least one stomach. Many have three, which brings effective efficiency to 80% and partially recovers what the base-rate reduction took away. Bodies are lean and tight: a core, two or three mouths, two bone struts, a stomach, a sensor. The bare minimum that survives.
One species dominates this lean era with an iron grip. sp_244, which emerges around tick 24,000, holds majority control for over 55,000 consecutive ticks. Its competitors appear, grow to a few dozen members, and then quietly disappear. The top position barely flickers. In a harsh economy, the incumbent has an enormous advantage: established neural patterns, refined body layout, a population large enough to buffer bad luck. Whatever sp_244 is doing, nothing does it better for 55,000 ticks.
The claw situation is instructive. The few organisms that carry claws average 14 nodes and 50 energy, well above the population average. Claws are profitable on the individual level. But claws don't spread. Predation works when there are soft targets. When 95% of the ecosystem is playing the same defensive forager game with similar body sizes, the return per kill isn't enough to fund the offspring that would spread the trait. Claws are frequency-dependent: viable when rare, self-defeating when common. So they stay rare. Barely.
One emergent behavior stands out in this era: the organisms cluster. Same-species groups congregate around food-dense patches and hold them against competitors. This isn't cooperation in any programmed sense — it's a direct consequence of the group sensor bonus, which grants +20% sensor range per same-species neighbor up to a 2× cap. An organism surrounded by three kin can detect food twice as far as a lone forager. The neural networks learn to seek out conspecifics not out of altruism but because proximity to kin is the most reliable path to early detection. What looks like hoarding is actually a perceptual advantage playing out in real time: the pack sees the hotspot first, arrives first, and by the time competitors detect it, the group's density makes it profitable to stay and defend rather than wander.
By tick 79,000, average body size has actually shrunk from 10 to 8.1 nodes. The simulation is doing the opposite of what Part 3 did. Bodies are getting smaller, not larger. sp_244 controls 45% of the population. Signal nodes have gone completely extinct. Claws are down to three organisms out of 177. The ecosystem is converging toward a single lean template and pruning everything that isn't essential.
The Founding Crisis — Tick 35,000–37,500
The lean era: bodies averaging 9 nodes, population below 200. sp_244 (the dominant cluster) holds 40%+ of the population. Notice how same-species groups congregate around food hotspots — the group sensor bonus rewards clustering with kin, so packs see twice as far as lone individuals and reach food first.
The first uncharted territory begins at tick 89,000. Everything before it replayed a crashed earlier run, deterministically identical tick for tick. After it, no data exists. The ecosystem that emerges from that boundary is not the one the previous 89,000 ticks suggested.
At tick 90,000, armor goes from a curiosity to a phenomenon. One lonely armored organism at tick 70,000. Eight at tick 79,000. Then 65 out of 260 at tick 90,000, a jump from 0.7% to 25% adoption in roughly 11,000 ticks. The fastest compositional shift of any part in the project's history, including Part 3's famous armor rush.
The trigger is the species succession. sp_975 has been slowly climbing at the expense of sp_244. By tick 84,000, sp_975 holds 88 members to sp_244's 31. A more homogeneous prey population is easier to specialize against. The few claw organisms, still present at the margins, start hitting the same body plan repeatedly. Armor, which reflects 20% of incoming damage back at attackers, becomes a viable counter. The claw population doesn't need to be large to create selection pressure. It just needs to be consistent.
Bodies start growing again. The shrinking era reverses. Average nodes climb from 8.1 to 9.9, and the maximum reaches 18. Armor nodes are cheap at 0.003 energy per tick, so organisms that would normally hold at 8 or 9 nodes can grow to 11 or 12 while barely increasing their metabolic burden. The first growth spurt in 70,000 ticks is funded by armor nodes that are nearly free to carry.
Stomach overtakes mouth somewhere around tick 95,000. After 95,000 ticks of mouth leading every composition chart in the simulation, the crossover is quiet and almost invisible until you check the numbers. At tick 90,000: mouth 25.7%, stomach 23.7%. At tick 110,000: mouth 20.2%, stomach 27.1%. The simulation started as an eating contest. It just became something else.
The Armor Rush — Tick 91,000–93,500
Armor (purple) spreading from 25% to 51% adoption in real time. Bodies growing for the first time since tick 10,000. The claw organisms at the margins are creating consistent predation pressure on a newly homogeneous prey population, and armor is the counter.
Between tick 110,000 and 115,000, the ecosystem undergoes the fastest transformation in 115,000 ticks of simulation. In five thousand ticks, claws go from 30 organisms to 118. Armor goes from 25% to 51% adoption. Fat, nearly extinct for 90,000 ticks, appears on 25% of organisms. Species diversity hits 103, the highest since before the claw redesign. The three-way equilibrium of the founding era, one dominant species, a handful of challengers, a sea of minor players, shatters completely.
The cause is self-reinforcing. Larger bodies with more energy reserves make better prey. When average energy crosses 46 and bodies average 12 nodes, a successful predator absorbs 70% of a meaningful kill. That energy funds offspring, which inherit claw morphologies, which increase predator pressure, which makes armor more valuable, which makes fat stores more valuable, which makes bodies larger, which makes prey more valuable. The loop closes.
Three distinct strategies become visible for the first time. Armored tanks: large bodies (averaging 18 nodes), heavy armor investment, stomach-heavy for maximum digestion efficiency. Predators: mid-to-large bodies (averaging 16 nodes), claws positioned for maximum velocity, energy reserves well above average. Lean foragers: the original body plan, smaller, no armor or claws, surviving on speed and efficiency. The ecosystem that spent 90,000 ticks converging on a single template has fractured into three.
The old dynasties are gone. sp_244 and sp_975, which between them controlled the population for most of the simulation's history, have both vanished from the leaderboard. The top five species at tick 110,000 control 48% of the population between them with no single species above 13%. Power is distributed and contested in ways it never was when one species held 40-45% for decades.
The Cambrian Moment — Tick 109,000–110,000
The hinge point: stomach just overtook mouth as the most common node type. Claws (red) are breaking out, armor (purple) is spreading. Bodies averaging 12 nodes with representatives from all 10 types for the first time.
The arms race that began at tick 115,000 runs for 22,000 more ticks and doesn't stop until it runs out of things to arm. The numbers trace a trajectory that has no analog in the earlier parts.
At tick 118,000: 62% claws, 79% armor, 65% fat, bodies averaging 17.5 nodes. At tick 121,000: 71% claws, 99% armor, 71% fat. At tick 126,000: 80% claws, 100% armor, 91% fat. At tick 137,000: 100% claws, 100% armor, 100% fat.
Six of ten node types reach universal adoption. Every organism in the simulation carries a core, bone, stomach, armor, fat reserves, and claws. The minimum body size, which was 7 nodes at the lean era's nadir and 14 nodes when the arms race began, reaches 25 nodes when it ends. There is no such thing as a small organism anymore.
The node hierarchy inverts completely. Stomach, which led every composition chart from tick 20,000 onward, falls to fourth behind fat, armor, and bone. The typical body plan in the post-arms-race world: 3-4 fat nodes, 3-4 armor nodes, 3-4 bone struts, 3-4 stomachs, 2-3 mouths, 2 sensors, 1-2 claws, 1 muscle, 1 core. The body budget that once went mostly to mouths and digestion now goes mostly to defense and energy reserves.
The Red Queen runs her full course. Armor at 20% creates pressure to add claws. Claws at 60% create pressure to add armor. Armor at 100% makes predation more expensive and drives larger bodies for more claw velocity. Fat at 100% means every organism survives encounters that would have killed the lean-era versions. Energy climbs: average 40 at tick 90,000, average 150 at tick 137,000. Maximum energy exceeds 600. Predation's 70% energy transfer creates a closed loop where kills recycle wealth between organisms instead of destroying it, and the ecosystem inflates.
Species diversity peaks at 195 during the mid-arms-race chaos, then declines to the 150 range as the mandatory trait list grows. When every organism must carry the same six traits, the design space for distinct body plans shrinks. Differences between species become subtle: how many of each node type, where they are placed, what the neural weights do. The ecosystem moves from competing body plans to competing body implementations.
The Arms Race — Tick 125,000–125,100
Post-arms-race equilibrium: 600 organisms averaging 28 nodes, every body carrying armor (purple), fat (yellow), claws (red), and bone struts (white). Minimum body size is 25 nodes. The ecosystem inflated 3× in mass and 4× in average energy over 47,000 ticks. The frames are dense enough that the viewer loads a snapshot in time rather than a smooth sequence — a reflection of how much information each organism now carries.
The plan was one million ticks. The world is 4000 by 2250 units, the most complex environment in the project's history. Recurrent brains, terrain biomes, ten node types, organism collisions, pheromone signaling. A run of this scope was always going to be slow.
What I didn't anticipate was that it would get slower over time. The physics engine simulates every spring, every mass, every collision in the population. At the lean era's peak of 200 organisms averaging 9 nodes, that's around 1,800 nodes in the world. At tick 137,000 with 600 organisms averaging 28 nodes, it's 16,800 nodes, more than nine times the load. The simulation that ran at 1,400 ticks per hour in the founding era slowed to under 900 by the post-arms-race plateau. One million ticks at that rate would have taken months of continuous compute.
The simulation was stopped at approximately 150,000 ticks.
Fortunately, the story was already complete. The arms race, which is the story of Part 4, started at tick 90,000 and ended at tick 137,000. Everything that happened happened within that 47,000-tick window: three strategic archetypes emerged and collapsed into one, six node types achieved universal adoption, minimum body size tripled, average energy quadrupled. The founding crisis, the armor rush, the Cambrian explosion, the Red Queen cascade, the universalization of every trait. The simulation ran long enough to tell the full story. It just didn't need a million ticks to do it.
Post-Arms-Race Equilibrium — Tick 147,000–147,100
The world at tick 147,000: 600 organisms averaging 29 nodes. Every body carries armor, fat, claws, bone, sensors, and stomach. Species colors still vary, but the silhouettes are almost identical.
This is what the last 13,000 ticks looked like. Six hundred organisms, all roughly the same size, all carrying the same complement of nodes, drifting through a world they have thoroughly saturated. The population hit its cap and stayed there. The species count fluctuated between 140 and 175, but the differences between species had become invisible to any metric except neural weights. Every body was a 26-to-34-node chassis built to the same blueprint: bone scaffolding, armor shell, fat reserves, claw armament, sensor array, stomach lining. The only variation was how the brain used all of it.
Nothing went extinct. Nothing broke through. The composition held to within a percentage point across 13,000 ticks of stasis. Fat and armor both locked at 17%. Bone at 17%. Stomach at 14%. The ecosystem had found its Nash equilibrium: no individual mutation could improve on the universal body plan, and no environmental pressure existed to disrupt it. The remaining 850,000 planned ticks would have been refinement, not revolution. The part that matters is the part that happened.
The composition charts tell the story more clearly than any other metric. Watch the fat and armor lines, both essentially flat near zero for the first 90,000 ticks, then climb in parallel through the arms race. Watch mouth decline from 34% at tick 28,000 to under 12% at the end as body budget shifts from feeding to fighting. Watch stomach rise and then get crowded out by the combat traits it made possible: by enabling energy surplus, stomach funded the arms race that ultimately demoted it.
The species chart shows the arc differently. Long dynasties in the founding era, sp_244 holding 40-45% for 55,000 ticks. Then fragmentation during the Cambrian explosion: no species above 13%, 195 distinct lineages competing. Then slow re-consolidation as the body plan converges and the differences between species become neural rather than morphological.
Signal nodes, which the simulation was partly designed to study, never find their moment. They peak at 7.5% during the arms race, briefly. The group sensor bonus, which should reward clustering, is outweighed by the metabolic cost of broadcasting and by the fact that sensor investment itself collapses under the pressure to carry armor and fat instead. Signal is the part of the experiment that didn't work. Part 5 will need a stronger mechanic to make communication matter.
Ecosystem Timeline
Population (green, left axis) and species count (gold, right axis). The founding crisis plateau, the Cambrian explosion, and the arms-race convergence are all visible in the species line.
The ecosystem timeline is a story in two halves. The first 90,000 ticks are a long plateau: population oscillating between 100 and 300, species count holding between 50 and 100. The lean era. Then at tick 90,000, both lines break upward simultaneously. Population climbs past 400, past 500, to 600. Species diversity explodes past 150, peaks near 200. The Cambrian moment and the arms race are visible as a single coordinated surge. Bigger bodies carry more energy, more energy funds more offspring, more offspring means more species. The two lines decouple only at the very end, when population holds at 600 while species diversity drifts back down. Convergence: different species, same body plan.
Species Dynamics
Streamgraph of species populations over time. sp_244's 55,000-tick dynasty dominates the early era before fragmentation during the Cambrian explosion.
The streamgraph makes the dynastic structure visible in a way the timeline can't. That wide band dominating the first 60,000 ticks is sp_244, the founding dynasty, holding 40-45% of the population for 55,000 consecutive ticks. No challenger comes close. Then sp_975 rises to replace it, holding a similar grip for another 20,000 ticks. These are not just the biggest species; they are the only species that matter. Everything else is noise at the margins.
Then the explosion. Around tick 100,000, the streamgraph shatters into dozens of thin bands. No species holds more than 13%. The visual shift from two-color dominance to a fragmented rainbow is the arms race arriving: when the mandatory trait list expands and bodies diversify, the design space opens and no single lineage can monopolize it. The late-game reconsolidation is visible too, as a few bands widen again at the right edge. But they never reach anything close to sp_244's grip. The era of dynasties is over.
Body Composition Over Time
Total node counts by type. Fat and armor are invisible for 90,000 ticks, then climb steeply through the arms race.
This is the chart that tells the arms race story most directly. For 90,000 ticks, the total node count in the world barely changes. A few thousand nodes spread across a couple hundred organisms, dominated by mouth and stomach. Then the curve goes vertical. Total nodes in the world go from roughly 2,000 at tick 90,000 to over 16,000 at tick 137,000. An eightfold increase in 47,000 ticks. The new mass is almost entirely combat infrastructure: fat (green), armor (purple), bone (gray), and claw (red) stacking on top of the original feeding body in visible layers. The mouth and stomach bands at the bottom barely change in absolute terms. They aren't shrinking. They're being buried under the weight of everything the arms race added on top.
Body Composition — Proportional
Each node type as a percentage of total nodes. Shows how body budget shifted from feeding (mouth, stomach) to combat (armor, fat, claw) over 150,000 ticks.
The proportional view tells a different version of the same story. In the lean era, mouth (gold) and stomach (brown) together account for 50-60% of all nodes. Core and sensor fill most of the rest. The body budget is spent almost entirely on feeding. Then the arms race compresses the feeding traits into a narrow band at the bottom. By tick 137,000, mouth and stomach together are under 25%. Fat, armor, bone, and claw have claimed the majority. The crossover happens around tick 110,000, the same moment the narrative identifies as the Cambrian explosion. Stomach's arc is particularly telling: it rises through the lean era as organisms invest in digestion efficiency, peaks around tick 95,000, then gets squeezed as combat traits crowd the body budget. Stomach enabled the energy surplus that funded the arms race, and then the arms race demoted it.
Species Dynamics — Proportional
Each species as a percentage of total population. sp_244's iron grip on 40–45% through the founding era gives way to genuine competition during the explosion.
The proportional species view is the clearest picture of political power in the simulation. The founding era is a two-party system: sp_244 commands nearly half the population, one or two challengers split the rest, and dozens of minor species survive at the margins. The transition around tick 80,000 is sudden. The incumbent dynasty collapses and nothing replaces it. The proportional chart goes from one dominant color to a patchwork of equal-sized bands, each species holding 5-10% at most. This is what genuine ecological competition looks like: no dynasty, no monoculture, just a shifting mosaic of lineages that rise and fall on timescales of a few thousand ticks. The late-game narrowing, where a few bands widen slightly, reflects morphological convergence rather than competitive dominance. The species that grow are the ones that happened to arrive at the universal body plan first.
Part 3 ended with a question embedded in its conclusion: if the environment demands all of them, evolution supplies all of them. What happens when the environment demands combat instead?
Part 4 answers that. Give evolution a dedicated attack node with velocity-based damage and watch the entire ecosystem reorganize around it. Not because the simulation forced it. Because the arms race logic is self-executing once the first condition is met: if killing is profitable, predators appear; if predators appear, armor appears; if armor appears, fat appears to survive armored combat; if everyone is armored and fat-padded, claws need to be larger and faster to penetrate; if claws need to be larger, bodies need to be larger. The loop runs to completion.
The ecosystem that started as a lean-era economy of 8-node herbivores converging on a single template ended as a world of 28-node armored predators where every organism carries every weapon and every defense. Bodies three times larger. Energy four times higher. A minimum body size that excludes any organism smaller than the average organism of 50,000 ticks ago.
The brain experiment, the recurrent networks, the memory registers, the pheromone signaling, will need a follow-up to properly evaluate. In a world undergoing a total arms race, the behavioral advantages of better memory are hard to measure against the immediate pressure to add another armor node. Signal nodes never mattered. Neural sophistication likely did, but quietly, embedded in the neural weights of the species that won each succession cycle rather than visible in any aggregate metric.
What Part 4 demonstrated, beyond the arms race, is that the design space for evolutionary surprises is far from exhausted. A single new node type, with a single mechanically interesting property, produced 47,000 ticks of dynamics that no amount of prior reasoning would have predicted in detail. The morphological story was complete at tick 137,000. The behavioral story is still unwritten.
Part 4 proved that minds can evolve, but evolution handed roughly the same mind to every organism. By tick 80,000 the arms race had saturated: every species converged on the same mix of claws and fat and sensors, and the simulation settled into the familiar late-game stasis. Part 5 asks whether death can break that equilibrium permanently.
When an organism dies in Part 5, it drops a corpse — a food object proportional to its fat reserves and energy at the moment of death. A fat organism at 150 energy becomes a 100-unit meal for whoever kills it, or whoever arrives first. Killing is now immediately profitable. Scavenging becomes a viable niche. Being fat makes you a target, not just a tank. A single-encounter lethality rule — exceed 30% damage in one exchange and the prey dies outright — ends the war of attrition that has dominated every previous part. Together these mechanics create the conditions for Lotka-Volterra oscillations: genuine predator-prey cycles that have never appeared in this simulation.
Food superpatches drift on predictable sine-wave paths, rewarding organisms that can anticipate rather than merely react. A new Memory node type makes that investment visible in body composition data: instead of invisible weight differences, the charts will show whether evolution actually selects for neural investment when it confers a foraging edge. The signal system is simplified to what it can teach. No population cap — the dynamics enforce it. Two concrete hypotheses, two testable outcomes. The first part of this project where we're not just watching to see what happens.