Lean Seasons

What v0.12 Changed and Why

In version 0.11, fruit became an established ecological staple (6 to 7 percent of the diet), supporting specialized frugivores and primitive hoarding. Yet food remained abundant year-round: winter sunlight never dropped below 60 percent of its mean, allowing a single generalist lineage to rule every region for over a million ticks without needing foresight or seasonal adaptation.

Ecological theory identifies seasonality as a powerful driver of cognition. Food-caching birds in harsh winter climates, such as black-capped chickadees, evolve larger hippocampi and sharper spatial memory than southern populations (Pravosudov and Clayton 2002; Roth and Pravosudov 2009). The cognitive buffer hypothesis similarly holds that larger brains provide the behavioral flexibility needed to survive seasonal food collapses (Sol 2009); in primates, seasonal scarcity also limits brain size, offsetting the buffer (van Woerden et al. 2012). Seasonal bottlenecks can also reshuffle communities, punishing complacent generalists.

Version 0.12 introduced a single physical law: the seasonal light curve became max(0, sin θ + cos(π·W)) with W = 0.25, scaled so total annual light remained constant. For one quarter of each 4,000-tick year—roughly 1,000 ticks, spanning most of an organism's lifespan—light is zero. Ambient free energy stops spawning, plants and trees halt growth, and fruit cannot set. In summer, light peaks at 2.3 times the mean. Geothermal vents, crystal deposits, carrion, nuts, and existing stores remain unaffected. Guided by Gemini, the committee adopted a gradual autumn to prevent immediate extinction and preserved vents as winter refuges, deferring changes like faster winter rot.

Validation

Validation benchmarked version 0.12 across eight seeds run for one million ticks in a 400-wide grid against a version 0.11 re-run. Passing required zero reboots, living lineages matching baseline, caching reaching at least 2 percent of diet (or nuts, fruit, and cache combined rising 5 points), and memory ops matching baseline or active working mind exceeding version 0.11 by more than one standard deviation.

Working mind—the volume of virtual-machine routine instructions executed per tick—rose sharply, memory ops reached 2.15, and crystal foraging rose to 31 percent as vents served as refuges. However, caching stayed at zero and lineages did not expand. Because cognitive expansion was confirmed safely without crashes, version 0.12 was merged.

The Story of the Run

Seed 298065341 (world width 400, speed x64) executed for 2,342,740 ticks. Winter immediately reshaped demographics: population swung between 150 in the dark and 1,100–4,650 during summer booms, with midwinter free energy falling to 825 units. Mating reached a record 15 percent by tick 475,237. Innovation was fast: 114 innovations by then (versus 42 by 396k in version 0.11), including fruit specialist BIT-873 (68 percent fruit). Strangely, four cache-sensing eyes evolved (cache kinship here, cache here, cache ahead, cache kinship ahead), mostly in clades TNK-998 and QRA-3961, despite zero caches on the map. Generalist ruler APO-3882 held 87 percent of the world with a lean 145-instruction genome and 7 memory ops.

By tick 852,013, minds expanded. World working mind hit 43 instructions executed per tick out of 144 carried, deploying a record 30 percent of the genome. Programs averaged an active repertoire of 4.6 routines per lifetime, 2.2 routines switched on, 2.5 memory ops, and 1.3 hormone ops per tick. Sharing appeared in ORC-4701 and GIB-5881, which cached or gave energy 11 to 17 percent of the time, and a fifth cache eye (left-right) appeared. Yet map caches stayed at zero, and APO-3882 still held 75 percent dominance alongside crystal specialist GIB-7021 (25 percent) across just two lineages.

Rather than sustaining complex brains, the dynasty subsequently streamlined. By tick 1,381,639, ruler KRN-7405 (descended from APO-3882) captured 79 percent of the world with only 71 instructions and 3.8 memory ops. Big genomes retreated to the crystal clade, where crystal miner KRN-7604 (17 percent share) carried 340 instructions and 36 memory ops. Innovation slowed to 9 innovations over 530,000 ticks.

At tick 2,342,740, the run was halted early for version 0.13. Final ruler APO-10222 (61 percent) preserved the unbroken APO-3882 lineage across six surviving lines. World execution averaged 33 instructions per tick, with diet split between free energy (56 percent), crystal (19 percent), nuts (16 percent), and fruit (5.1 percent). A sixth cache eye (cache kinship left-right) appeared at 1.87M ticks, but caches remained absent.

What It Means

Winter pays for bigger minds, briefly. As the cognitive buffer hypothesis predicted, seasonal darkness initially favored programs executing larger behavioral repertoires and memory routines to manage resource fluctuations. Yet sustaining big brains proved energetically unnecessary once generalists adapted. The ruling line survived not by storing food; its genomes got smaller (71 instructions at 1.38M). How it bridged the winters is not yet measured.

The total failure of food storage stemmed from simulation physics. Under the CACHE law, a gland deposited energy directly into the underlying cell, losing only 0.02 percent per tick. However, normal feeding draws energy directly from the current cell. Because organisms feed nearly every tick, cachers instantly re-ate their own deposits, while passers-by grazed the rest. Caching required a coordinated routine to deposit energy, suppress feeding, and depart—an evolutionary cliff offering no partial reward. Programs evolved six distinct eyes seeking caches, but found nothing.

What's Next

To bridge this gap, the committee designed version 0.13 ("Buried Stores"). Mirroring scatter-hoarding animals like jays and squirrels, cached energy is now buried, out of reach of grazing (cache eyes can still detect it, much as a dog smells a buried bone). Retrieval requires active digging via a negative CACHE activation, enabling spatial memory, recovery, and pilfering to co-evolve.

Validation for version 0.13 (8 seeds × 1M ticks) confirmed stability (zero reboots), active minds (47.2 ± 13.0 run/tick), memory ops (1.95), and hormone ops (1.67). Probes on seed 3 confirmed that buried stores endure, holding 4,461 energy across 5,222 cells at 80,000 ticks. A flaw where digging never ran was caught by Codex and resolved prior to validation.

Still, caches in the diet registered 0.00 percent: programs scattered tiny deposits without learning to return and dig within one million ticks. Version 0.13 is now running live. If digging fails to emerge over long runs, the committee will introduce stepping stones—such as sensory perception of self-made caches—or determine whether storage is viable in The Grid.

Data: snapshots and logs: data/scratch/v012-archive-final.json (seed 298065341, 2,342,740 ticks); decisions D-026 and D-027; validation runs val12-v012.txt and val13-v013.txt.