Findings
- Sweeps
- 22
- Runs finished
- 5,462
- Epochs simulated
- 154,691,331
- Seeds that transitioned
- 472
- Largest replicator census
- 4,000
What the sweeps have shown so far. Every claim is written against its own runs, and every run keeps its seed, so any of them can be re-run. The words these write-ups use are defined in the glossary.
- open
- The sweep is running and no claim is made yet.
- partial
- A reading is stated, but it cannot yet be read as final.
- published
- Every run finished and the claim stands on them.
- negative
- The sweep finished and the effect was not there.
-
Paid for depth, the soup jumps once and stands: no late rise on a ladder without a near top
DESIGN §1.3 sweep 18 asked whether the deepest logic feature a soup holds keeps getting deeper when depth is paid, on a ladder of every function of four inputs, each paid by its minimal NAND count. The 54 meta-stack worlds ran another 100 000 epochs under three rewards: depth paid, depth paid up to five NANDs, nothing paid. All 162 children finished, none died out and none reached the ladder's floor. The rung-4 question is not shown: one child of the paid arm rose late against none unpaid, 1 to 0 with 53 ties (p = 0.5). Against the capped arm it is refuted, 1 to 13: the arm that pays nothing past depth five crept up a rung more often. The re-readings agree. The one late rise is a re-climb to depth 10, where most paid children sat from early on, and offline its code grew by 2 bytes, barely above drift. The soup takes the jumps a reorganisation offers early, then stands. The substrate imports an objective, a primitive, a hereditary channel, the primitive's semantics and a ladder, so rung 4 on Soup stays not shown.
Sweep: Topless rise
162 / 162 runs finished · 0 transitioned (—) -
On growable tapes, a reach of 4 emerges more than ten times as often as radius 1
DESIGN §1.3 sweep 14 asked whether the best reach of the locality finding carries to the growable tapes every rung-4 experiment runs on: the host–parasite sweep's economy-off world at cap 128, tapes of 64 bytes free to grow to 128, at radius 4 on that control's seeds 1–270. All 270 runs finished and every world they kept was read. Radius 4 emerged in 119 of 270 worlds against the control's 11 of 270 at radius 1, 10.8 times the rate (one-sided Fisher p = 1.5e-30), so the pre-registered test is shown. On fixed 64-byte tapes sweep 13 had read radius 4 at four times radius 1. In both arms the dominant replicator fills the 128 bytes it may grow to. 108 of the emerged worlds at radius 4 end with replicators holding at least half the world, against 11 in the control: a pool of parents with room to grow, though no later sweep's rule is set by it.
Sweep: Reach with room to grow
270 / 270 runs finished · 171 transitioned (63%) -
With a stack NAND on a metabolism tape, paid parts assemble the logic rungs that need an input twice, in a sixth of the worlds
DESIGN §1.3 sweep 17 moved the Logic sweep's computing off the copier: a second, 32-byte tape the soup never runs, handed on with each copy and read by the assay with a NAND that keeps its operands. The 18 emerged worlds ran another 40 000 epochs, three seeds a parent, paired with the Logic sweep's children. All 162 children finished and none died out. 9 children of 54, from 7 parents, held XOR or EQU at a tenth of their world, against none in any other arm, so the rung-4 question is shown, 9 pairs to 0 with 45 ties (p = 0.00195). The stepping-stone test against the arm paid for XOR and EQU alone is shown too, 9 to 0: the deep rungs were assembled from paid parts, Lenski's mechanism in a soup. The same tape read by the NAND that writes in place built none. Without the four piloted parents the deep tests read 8 to 0 and rest on two parents. In 14 of 15 plantings a world the deep rung was wiped from re-grew it within 500 epochs: a property of the population more than of a lineage. The ladder tops out at EQU, and the substrate imports an objective, a primitive, a hereditary channel and the semantics that let the deep rungs come, so rung 4 on Soup stays not shown.
Sweep: Meta-stack
162 / 162 runs finished · 0 transitioned (—) -
Paid to compute, replicators take the one-step tasks and stop there
DESIGN §1.3 sweep 15 paid replicators for computing. The 18 emerged worlds of the from-emerged sweep ran another 40 000 epochs each, three seeds a parent, with arithmetic tasks rewarded and, in a paired twin, assayed but unpaid. All 108 children finished and none died out. Pay bought capability: all 54 pairs favour the reward (p = 5.55e-17), and the paid worlds took ECHO, INC and DEC, the tasks one substitution of a copier computes. They went no further. The rung-4 question, whether paid computation climbs to the tasks that need a counted loop, is refuted: no child of either arm held a loop task at a tenth of its world, 0 to 0 with 54 ties. Paid complexity did rise, in 18 pairs against 1 (p = 3.81e-05), and both shown tests hold without the piloted parents. The substrate imports an objective, so none of this moves rung 4 on Soup, which stays not shown.
Sweep: Metabolism
108 / 108 runs finished · 0 transitioned (—) -
Given a NAND, paid replicators climb every logic task that reads its inputs once, and none that needs one twice
DESIGN §1.3 sweep 16 gave paid replicators a NAND. The 18 emerged worlds of the from-emerged sweep ran another 40 000 epochs each, three seeds a parent, with an assay-only NAND byte and Avida's logic tasks as the ladder, in three arms: every rung paid, XOR and EQU alone paid, and nothing paid. All 162 children finished; one paid child died out. Pay bought capability: all 53 measured pairs favour the paid ladder (p = 1.11e-16), and the paid worlds climbed every rung from NOT to NOR, the tasks a NAND circuit can compute reading each input once. None held XOR or EQU, the two that need an input twice, at a tenth of its world. The rung-4 question, whether a soup assembles those deep rungs from paid parts, is refuted, 0 to 0 with 53 ties, and so is the stepping-stone test against the arm paid for XOR and EQU alone. Paid complexity did not rise, 2 pairs against 1 (p = 0.5). The substrate imports an objective and a primitive, so none of this moves rung 4 on Soup, which stays not shown.
Sweep: Logic
162 / 162 runs finished · 0 transitioned (—) -
Emergence peaks at an intermediate reach
DESIGN §1.3 sweep 13 put an exploratory pattern from sweep 12 to fresh worlds under rules fixed first: does a replicator arise more often at an intermediate reach than at the tightest one or in a well-mixed world? All 630 runs finished, 90 fresh seeds at each of seven reaches. Radius 4 emerged in 16 of 90 worlds, against 4 of 90 at radius 1 and 3 of 90 well-mixed (one-sided Fisher, Holm-corrected p = 0.0039 and 0.0027), and a logistic fit over the six finite radii bends down (radius² coefficient −0.114, Wald p = 0.0041) with its peak at radius 4.1. Both pre-registered hypotheses are shown. Four of the seven arms emerged exactly as often as sweep 12's did: different seeds and different worlds, the same totals.
Sweep: Locality and emergence
630 / 630 runs finished · 94 transitioned (15%) -
Under an energy economy, the dominant replicator copies itself faster
DESIGN §1.3 sweep 11 continued emerged worlds for another 20 000 epochs, each paired with a continuation under its own dynamics. A confirmatory reading, written after ten parents had been seen and tested only on the eight that qualified afterwards, asked rung 3's question: under an energy economy, does the dominant replicator get faster at copying itself? It does. Its copy_latency, the step at which a partner first holds a complete image, fell relative to the paired continuation in 20 of 24 pairs at economy 2048 (p = 0.00077) and 21 of 24 at economy 8192 (p = 0.00014). The median child's last-decile latency was 0.82 and 0.75 of its first-decile latency under the two economies, against 1.11 in the continuation, and the result holds in 7 and 8 of the 8 parents. It is the programme's first confirmed adaptation: one substrate, latency read on the dominant tape in isolation, and the mechanism is not identified.
Sweep: From an emerged world
216 / 216 runs finished · 0 transitioned (—) -
From an emerged start, an energy economy does not keep complexity rising
DESIGN §1.3 sweep 11 asked whether an existing replicator keeps getting more complicated once a treatment is switched on: 216 continuations of 18 emerged worlds, 20 000 epochs each, under the parent's own dynamics, two energy economies or host mode. Read as pre-registered, neither economy arm can be read, because each relapses more than the continuation; the rich economy's sign test (12 pairs to 1) rests on two parents. Host mode is refuted, and so is strict persistence: 52 of 54 continuations held their replicators and 2 relapsed. A held-out confirmatory reading on the survivors of 8 unseen parents did not replicate the complexity signal: refuted at economy 2048, not shown at economy 8192 (1 pair to 0). From an emerged start the economy does not keep complexity rising.
Sweep: From an emerged world
216 / 216 runs finished · 0 transitioned (—) -
After emergence, one lineage takes the world at every reach
DESIGN §1.3 sweep 12 asked rung 2's question: once a world has made replicators, does it stay polyphyletic, and do more lineages survive as a cell's reach shrinks? All 360 runs finished and 32 emerged. Read by the pre-registered rule, on lineage tags that follow descent through reverse copies, none of them stayed polyphyletic: 27 read monophyletic, 5 between one lineage and two, and the median effective number of lineages is 1 in every arm, well-mixed to radius 1. The trend toward more lineages at shorter reach is neither shown nor refuted, at a permutation p of 0.079. Emergence itself differed by reach — 3, 16, 9 and 4 runs of 90, well-mixed then radius 4, 2 and 1 — which the sweep was not designed to test and is reported descriptively only.
Sweep: Lineage diversity
360 / 360 runs finished · 61 transitioned (17%) -
Does complexity keep rising when energy is contested?
Complexity rises at emergence and then stops rising on every substrate the programme has tested, because a byte off the copy path costs its tape nothing and no quantity in the world is worth taking. DESIGN §1.3 sweep 9 priced that — instruction energy as a stock, with a steal op that takes what a neighbour saved — and the sweep has finished. Pricing energy did not make life more common: no priced arm emerged significantly more often than the economy-off control at its cap, and the poorest economy held no replicator at all. Theft evolved wherever it was offered. Read arm by arm on dominant_instruction_count and the conserved core, no priced arm the rule reads keeps rising: the one that did at 90 seeds, on one of its two measured runs, reads neither at 270, beside its control at 270, and the rest plateau, read neither or are barren. Two arms that were never extended hold one measured run each, and it rises, too few for the rule to read, so the verdict is unresolved rather than negative. That reading stands on a measured-run rule amended after the data were seen; as registered, the rule reads almost every emerged run as unmeasured and no arm at all.
Sweep: Host–parasite economy
1800 / 1800 runs finished · 41 transitioned (2%) -
Does complexity keep rising when only one partner's code runs?
Complexity rises at emergence and then stops rising on every substrate the programme has tested, and each bet so far changed what a tape can hold or spend rather than what it is to its partner. DESIGN §1.3 sweep 10 changed that — under a host interaction only the first tape's code runs — and the sweep has finished. Under it nothing emerged at all: the host arms are barren at both caps while the concat controls beside them held replicators, so the asymmetry did not lower the plateau, it kept replication from arising. With no host run to read, the pre-registered refutation and the distinct_lineages secondary cannot be evaluated; the concat controls, the same worlds sweep 9's control ran on its seeds 1–90, are read as the reference.
Sweep: Asymmetric execution
360 / 360 runs finished · 9 transitioned (3%) -
Does complexity keep rising?
The open-endedness verdict across the three substrate bets of DESIGN §1.3 — instruction cost, environmental structure and room to grow — read against the default substrate the complexity-plateau finding established. Each of those sweeps carries its own control arm: the cost off, a uniform world, a tape that cannot lengthen, the substrate every earlier sweep ran. This reads each sweep at render time and asks the same two questions of every arm — does the dominant replicator reach a higher complexity than the control arm reaches, and does the world keep more lineages alive — by placing each transitioned run above or below its own control and counting, never fitting a trend. An arm with too few transitioned seeds decides nothing, and the hypothesis it belongs to reads unresolved rather than answered.
Sweeps: Instruction cost · Environmental structure · Room to grow
Weighed against each other, and against the substrate they vary. -
Does copying get cheaper?
Every sample now records the copy cost of the dominant replicator — the interpreter steps the most populous tape that passes the replicator test needs for one byte-exact copy — so a world that got better at copying itself can be seen getting better rather than inferred. This reads that series off every transitioned run the lab has, whichever sweep it came from, and counts how many end cheaper than they started. It is the first direct test of adaptation in the programme and it is not yet the adaptation sweep: the runs were assembled by having transitioned, not designed to answer this, and a falling cost can be a cheaper replicator winning rather than one lineage improving.
Sweeps: Asymmetric execution · BFF positive control · Environmental structure · Host–parasite economy · Instruction cost · Instruction set ablations · Lineage diversity · Locality and emergence · Max steps per interaction · Mutation rate · Mutation rate, long runs · Neighbourhood radius · Reach with room to grow · Room to grow · World size
472 terminal runs that crossed, finished or failed -
Does the replicator keep getting more complicated?
Every sample now records how much tape the dominant replicator is — the compressed length of the most populous tape that passes the replicator test, and how many of its bytes the run's instruction set executes. That is the open-endedness baseline every later substrate is measured against: a world that keeps elaborating what it found reads a rising length, a world that found one recipe and stopped reads a flat one. This reads the series off every transitioned run the lab has and counts how many end more complicated than they started. It measures a tape's size, not its abilities, and the runs were assembled by having transitioned rather than designed to answer this.
Sweeps: Asymmetric execution · BFF positive control · Environmental structure · Host–parasite economy · Instruction cost · Instruction set ablations · Lineage diversity · Locality and emergence · Max steps per interaction · Mutation rate · Mutation rate, long runs · Neighbourhood radius · Reach with room to grow · Room to grow · World size
472 terminal runs that crossed, finished or failed -
Emergence is a state a world can leave
DESIGN §1.2 fixes when a world enters the transitioned state, and the engine reports only that first crossing; nothing said when a world leaves one. Every terminal run now carries a persistence summary read off its own stored samples, and this reads every one of them the lab has, whichever sweep it came from: how many transitioned worlds were still in the state at their last sample, how many climbed back out, how long each held, and how high its replicator census rose. Relapse is not an edge case — the design record's two recorded relapses are only the ones that were noticed — and the outcome is split by whether the census ever saw a colony, since a world can leave a state no replicator was ever counted in. Every number is read from the database at render time and describes the runs the lab happens to have: each was censored by its own budget, so a world that persisted persisted to its last sample and no further.
Sweeps: Asymmetric execution · BFF positive control · Environmental structure · Host–parasite economy · Instruction cost · Instruction set ablations · Lineage diversity · Locality and emergence · Max steps per interaction · Mutation rate · Mutation rate, long runs · Neighbourhood radius · Reach with room to grow · Room to grow · World size
472 terminal runs that crossed, finished or failed -
Emergence is rare, and half of it happens after epoch 20 000
The four rates around sweep 1's transitions, re-run for 60 000 epochs — three times the budget that produced the mutation-rate window finding. It reproduces that sweep run for run and then doubles it: six of 40 runs reach a census-confirmed replicator, and half of those transitions land after epoch 20 000, where the short sweep had already stopped watching. One of its disagreements resolves the same way — the run it recorded as flagged with an empty census counts replicators at epoch 26 140. The 2^-14 arm stays silent at three times the budget, so the lower cutoff is not a censoring artefact.
Sweep: Mutation rate, long runs
40 / 40 runs finished · 8 transitioned (20%) -
Life emerged, but the mutation-rate window did not
All 100 runs finished, and five of them crossed the transition: self-replicating structure arose from a random soup with no fitness function, the first at epoch 5 030. But it arose at 2^-13, 2^-12, 2^-9 and 2^-8 alike, about one seed in ten wherever it arose, so the window of DESIGN §1.3 sweep 1 is not supported at this scale. Only a lower cutoff is hinted at: 0 of 40 runs below 2^-13 against 5 of 60 at or above it, one-sided Fisher p ≈ 0.073.
Sweep: Mutation rate
100 / 100 runs finished · 5 transitioned (5%) -
Does the engine reproduce BFF emergence at all?
The positive control the design record makes mandatory: a well-mixed soup of 2^17 tapes, the shape the published BFF work used. It shows the transition in tape statistics in one mutated seed and the largest replicator census in the lab in another, so the interpreter and the pairing rule are not what the sweeps were measuring. Two of its stored worlds sit on census-positive samples and rescore to the live count exactly, 316 replicating cells and 112, so the census is real; but no stored world brackets the peak itself at this experiment's snapshot cadence, so its height is still read from in-run samples alone. A zero-mutation control collapses entropy with a census of zero: compression alone is not a replicator.
Sweep: BFF positive control
6 / 6 runs finished · 6 transitioned (100%) -
Does time to emergence scale with the number of cells?
DESIGN §1.3 sweep 2 sets two readings against each other: if emergence is a hazard per cell-epoch, a world with more cells buys more tickets and events scale with cell count; if it is a hazard per world, cell count buys nothing. Square worlds from 32² to 256² separate them, and emergence gets more likely as the world gets bigger: 0, 0, 1 and 3 of 10 seeds transitioned, against the 0.06, 0.26, 1 and 3.7 a constant per-cell hazard predicts. Pooled, that hazard is 2.5 × 10⁻¹⁰ per cell-epoch (95% 0.7–6.4). The earliest transition of the whole programme, epoch 4 730, is a 256² run — but four events cannot measure an exponent, and two of them are still resuming from their snapshots.
Sweep: World size
40 / 40 runs finished · 4 transitioned (10%) -
Does spatial locality buy emergence?
All 40 runs finished and 6 transitioned, but locality is not what decided it: the well-mixed arm transitioned as often as the best local arm, 2 of 10, and the tightest arm did worst. The speed half of DESIGN §1.3 sweep 3 is not supported — though 1 or 2 events per arm can only rule out an enormous effect. What the sweep does show is about the soup rather than emergence: among runs that never transitioned the final compress_ratio falls monotonically as a cell's reach widens, 0.951 to 0.859. The diversity half of the hypothesis is untested and stays open.
Sweep: Neighbourhood radius
40 / 40 runs finished · 6 transitioned (15%) -
More compute per interaction is not more life
DESIGN §1.3 sweep 4 expected a floor: below some number of instructions a copy cannot finish, above it the budget should stop mattering. What the four arms show instead is a single positive point. In 20 000 epochs at 128×128 emergence appears only at max_steps 8 192, 2 of 10 seeds, and never at 256, 1 024 or 65 536 — the arm with eight times the compute of the one that works produces nothing, and the extra instructions are spent rather than merely bought: an epoch at 65 536 costs twelve times an epoch at 256. The detector and the census agree in every arm, which is rare in this programme. But a peak found by one arm out of four is located to within a factor of 64, and each silent arm bounds its rate only to about 0–26%.
Sweep: Max steps per interaction
40 / 40 runs finished · 2 transitioned (5%)