A Changing World

What Changed in v0.7 and Why

In v0.6, we encountered an evolutionary trap: easy living degraded minds. On rich, unchanging terrain bathed in eternal sunlight, digital organisms stripped their genomes down to single routines running barely 30 instructions per tick (the simulation's fundamental time step). Only programs starving in the resource-poor crystal belt retained large genomes and complex memory. Grounded in biology's "variability selection" and "cognitive buffer" hypotheses—which argue that intelligence evolves to navigate unpredictable environments and extractive foods—v0.7 overhauled The Grid’s physical laws:

• A Dynamic Climate: Sunlight cycled through regular seasons (multiplying solar energy by ±40% over 4,000-tick years), fractal light zones drifted across the world every ~300,000 ticks, and hydrothermal vents became ephemeral, expiring and reopening elsewhere.
• Extractive Energy Ecology: Beyond passive free energy ("grass") and slow-growing crystal, we introduced woody trees that grew logistically, pulsed nut crops in synchronized "mast" events, and stores for caching energy across lean seasons. Accessing dense nut calories required a multi-step act: finding the nut, stopping, and cracking its shell with mechanical force (ERODE) or slow gnawing.
• Autonomous Wild Game: Independent kingdoms of grazing and browsing animals were released to co-evolve alongside our programs, creating living prey to test the "Red Queen hypothesis"—the evolutionary principle that organisms must continuously adapt just to survive against active biological competitors.
• Cheaper Inheritance: The metabolic cost to copy genetic instructions was slashed from 0.05 to 0.015 energy units per instruction, reflecting the Grid's premise that minds should cost energy to run, not to inherit (brains are metabolically expensive tissue; Aiello & Wheeler 1995).

The Story of the Run (in Eras)

Across 5.71 million ticks and 8,905 lineages, the history of seed 941682414 divided into three distinct evolutionary epochs:

1. The Burst of Promise (0 to 500k ticks)

The run opened with explosive adaptation. Organisms achieved 30 behavioral and sensory innovations in just 114,000 ticks (a feat that took v0.6 half a million ticks). By tick 31,500, programs independently evolved nut-cracking: TRN-137 derived 42% of its energy from nuts, QRA-281 reached 71%, and KRN-160 hit 23%. Sexual reproduction flourished, supplying 38% of all early births.

However, this early dawn was fragile. Mineral rushes sparked wild population surges: CLU-150 peaked at 1,666 individuals before crashing by 52% and dying out. The pioneer nut-crackers left no descendants. By tick 114,265, a population bottleneck left SWT-420 (descended from genesis organism SWT-6) commanding 100% of the Grid. By tick 515,990, rule passed to its direct child TRI-958, while the tree and nut counters plummeted to zero.

2. The Crystal Cycle and Monocultures (500k to 3.7M ticks)

With forests extinct, the world simplified into an austere cycle of mineral exploitation. Because no living organisms initially mined crystal, huge reserves accumulated (reaching 306,000 units by tick 1.65M), fueling massive booms of 500 to 1,187 individuals whenever miners evolved, followed by starvation crashes once deposits were stripped bare.

Political power remained strictly dynastic: TRI-958 begat TRN-1966 (holding 99% of the world at 1.1M ticks), who begat SMT-2224, BIT-2849, TRI-3146, TRI-3907, and MRP-4603. Flashes of large working minds appeared on the fringes: DMN-1012 executed 74 of its 92 instructions per tick at 516k, and crystal miner QRA-3625 executed 79 of its 109 instructions at 2.66M, driving the world average to 49. But both died out without descendants, replaced by lineages like QRA-4000 that hoarded 122 instructions while running only 41.

3. Extreme Bloat and the Ecological Split (3.7M to 5.71M ticks)

At tick 4.24M, ruler CLU-5842 established a 100% monoculture, displaying the run's most extreme bloat: 137 instructions carried, but only 32 run per tick (23%). Mating crashed to 0%.

Yet following massive population spikes, a striking divergence kept recurring (at 2.66M, 4.75M and 5.71M ticks): the ruling line fractured into two distinct forms:

1. A lean, direct-line ruler grazing on free energy with ~31–37 instructions (SWT-7385, KRN-8159, and APO-8830).
2. A heavy crystal miner carrying 120 to 172 instructions and 4 to 8 routines (such as GIB-7418 with 172 carried, or SRK-8733 with 126 carried), executing 42 to 46 instructions per tick.

Each time, the miner line later died out without descendants (why is not yet confirmed), and the world collapsed back to a grazer monoculture (such as KRN-8159, 100% at 5.22M) until the next boom and the next split. In this final era, sex rebounded steadily across three consecutive windows (14% → 23% → 29%), and hunting rose to 398 total kills.

What Worked

Several major interventions proved successful:

• Stability: After tuning, no world crashed. Working minds held at ~35 instructions per tick past 1.4 million ticks, the point where v0.6's carried code had shrunk; v0.6's working minds had also held steady, so this is not evidence that the changing world helped minds.
• Novelty Generation: The run produced 124 recorded innovations (versus 105 in v0.6), including complex routines utilizing up to 24 memory operations.
• Social Cooperation: The GIVE action (voluntary energy sharing) spread to nearly every living lineage by tick 2.15M, and territory marking recurred among rulers (TRN-1966, MRP-4603, APO-8830).
• Niche Differentiation: The recurring split between lean grazers and big-genome crystal miners demonstrated that distinct ecological castes can spontaneously re-evolve from a single ruling lineage.

What Failed and Why

Four critical mechanics faltered:

• The Forest Defect: Between tick 114k and 516k, tree biomass crashed from 29,607 to zero and nuts from 1,710 to zero, eliminating forest foods for the remainder of the run (final whole-run diet: 0.2% nuts, 0.1% browse). The likely cause (unconfirmed) was an ecological flaw: browsing, by the wild browsers and the Grid's own programs, kept saplings clipped down to immortal rootstocks below the browse line. Stunted rootstocks permanently clogged lit cells so falling nuts could not germinate. When drifting sunlight moved away from mature canopies, the forests died unreplaced.
• Wild-Game Immigration: Autonomous animals were propped up by an artificial population floor. Whenever browsers dropped below 9, nine cloned immigrants spawned with 100 units of free energy each, inheriting the living browser lineage BRW-2808. Consequently, BRW-2808 logged 9,288 births and 8,788 starvation deaths while remaining perpetually frozen at generation 2. This injected unearned free energy into the physics-only simulation while forestalling co-evolution (totaling only 398 hunts).
• Genetic Bloat: Genomes carried far more code than they ran: 46–101 instructions on average by era, peaking at 172 in GIB-7418, while the working mind stayed near 35. Cheaper copying may have contributed, but v0.6 showed the same pattern at the old cost.
• Sex and Monoculture: Mating plummeted from 38% early to 0% at 4.24M ticks, averaging 8.6% overall. Our interpretation (not yet tested): in a largely uniform world without persistent enemies, sexual reproduction could not consistently outperform asexual budding. An unbroken dynasty of twelve successive rulers held power, repeatedly bottlenecking the Grid into total monoculture (100% at 4.24M and 5.22M).

The Central Finding: The Clamped Mind

The definitive finding from 11.2 million cumulative ticks of v0.6 and v0.7 is stark: regardless of seasons, drifting light, forests, or game, instructions executed per tick remained locked between 31 and 43.

Looking across million-tick eras, the consistency was absolute:

• v0.6: 40, 36, 33, 31, 37, 41
• v0.7: 38, 36, 43, 35, 34, 36

Large, active minds were not impossible to evolve. Programs running 95 to 115 instructions per tick arose repeatedly in v0.7 (including SWT-3659, CLU-3670, APO-984, and ANO-1647). Yet not a single one ever exceeded 4 living members.

Our engine diagnosis identified the bottleneck: The Grid's virtual machine executed every routine sequentially from top to bottom every tick, with an arbitrary rule that the last routine to set an action won. Consequently, if a mutation produced a new functional routine, it almost inevitably overwrote and broke the vital foraging or movement actions established by earlier routines. Adding functional intelligence destroyed existing competence. Silent mutations—unexecuted copies—were the only safe mutations.

What's Next: v0.8 ("Genes That Listen")

To break through this mechanical ceiling, v0.8 redesigns the engine, world chemistry, and geography:

• Gene Regulation: Genomes will no longer execute top-to-bottom. Routine genes are gaining biological promoters—conditional switches responding to internal states (hunger, health, age), senses, or four decaying hormone signals (h0–h3). Routines will run by evolutionary priority and claim discrete action groups (movement, feeding, fighting). A high-priority "flee" module can take over movement during danger without overriding foraging when calm.
• Cooking and Carrying: Autonomous game kingdoms are removed. In their place, organisms will gain pouches to carry crystal shards to volcanic vents. Heat raises the temperature in crystal's chemistry, so shards carried to a vent release their energy far faster, grounding the "cooking hypothesis": cooking made food easier to extract, helping fund large brains. Pouch theft will spark a direct, peer-to-peer arms race.
• Geography and Repaired Forests: The world expands to 400 units across, divided by mountain ranges with passable bottlenecks, enabling geographical isolation (allopatric speciation) to break the twelve-ruler monopoly. Forests will be revived with mortal saplings and external seed rain.

Data: data/runs/2026-09-28T00-52-08-522Z_seed-941682414