Fruiting

The Question v0.10 Asked and the One Change It Made

Version 0.9 showed that removing inheritance taxes on routine code allowed large, regulated genomes to dominate, sustaining working minds of 47 instructions executed per tick (the simulation's fundamental time step). Version 0.10 asked: can the environment actively reward that cognitive capacity through memory, timing, and navigation?

This tests the ecological intelligence hypothesis (Milton 1981; DeCasien et al. 2017). Brain size across primates tracks diet: fruit-eaters possess larger brains than leaf-eaters because fruit is patchy in space and time. Trees ripen on independent schedules, rewarding foragers that memorize "traplines"—visiting specific groves at specific intervals.

Version 0.10 added one mechanic:

• Fruiting Sites: Approximately 48 sites (~80 units apart) were seeded on lit ground at genesis. Each operates on an independent cycle (3,000–9,000 ticks), depositing 1,500 energy over a 500-tick window into a 2-cell radius patch (re-tuned from 600 energy over 300 ticks in Decision D-021), rotting at 0.4%/tick. A visual blossom appears 200 ticks prior.
• Short-Range Vision: Fruit and blossoms are visible only within 4 units, regardless of eye range. Finding fruit reliably requires spatial memory (STORE, LOAD, BEAR to stored coordinates) and timing (internal counters or clocks).

Validation

Under Decision D-021, initial validation across 8 seeds (1M ticks, world 400) failed: fruit was 1% of diet (target: ≥5%) and memory ops fell. Following parameter re-tuning (1,500 energy, 500-tick window, 0.4%/tick rot), Decision D-022 benchmarked v0.10 against v0.9:

The committee read the shortfall as fruit rotting unfound (a later measurement showed otherwise; see the correction below). Because sensory traits (+86%) and carried memory (+55%) rose without crashes, the committee merged v0.10 under D-022 to let the live run test whether traplines needed more time.

The Story of the Run

The run began on 2026-09-28 at 07:49 (tag v0.10, commit bdca2b9, speed x64, world 400).

First Cooks and Early Dynasties (Tick 806,635)

Early dominance belonged to ZUS-2828 (58%), a lean wanderer (79 carried / 32 run) living on free energy and nuts across 5 of 6 regions. Fruit accounted for 1.9% of world income (free energy 58%, crystal 20%, nuts 13%, browse 5%). Fleeting fruit foragers emerged—GIB-660 (22% fruit at 28k ticks) and ANO-1472 (13% at 289k)—while a blossom sense appeared at 50k (42 carriers), but traplining never materialized. Unexpectedly, the run logged its first cooks on record: TRN-1147 (tick 197,500) derived 68% of its energy from cooked crystal: it carried crystal shards in its pouch to volcanic vents, where the heat releases their energy faster. NIO-1879 later cooked 11%. Neither founded a lasting line, leaving cooking at 0.2% of world diet. Meanwhile, giant crystal miner TRI-2623 (16%) ran 53 instructions/tick across 10 routines with 33 memory ops.

Swelling Genomes and First Caches (Tick 1,668,867)

Carried code leaped from 98 to 263 instructions, while working minds remained lean at 37 run/tick (14%). Hormone ops peaked at 3.2/tick, and 98% of routines were conditional. Ruler BIT-4313 (71%, descending from ZUS-2828) led 5 regions with 272 carried instructions but only 35 run. Fruit held at 1.9%, despite new eyes for fruit here (1.06M ticks) and ahead (1.34M). Concurrently, caches appeared for the first time in a live run: 8 storage caches formed, feeding newborn lineages ANO-5038 and NIO-5039 (28–38% of their energy). New eyes for caches and for cache kinship appeared around the same time (1.45M); whether they were used to find caches is unconfirmed.

Record Genomes and the Fruit Specialist (Ticks 2,476,432 to 3,222,226)

By tick 2.48M, carried code averaged 419 instructions (working mind 40 run/tick). Crystal miners reached massive proportions: QRA-6360 carried 611 instructions in 22 routines with 62 memory ops. A burst of fruit foragers arose between 2.05M and 2.15M ticks (MRP-6714 reached 30% fruit), but all collapsed or reverted.

At tick 2.61M, a true fruit specialist appeared: ZUS-7892 drew 49% of its energy from fruit, peaking at 805 programs. Yet its lineage abandoned the niche; descendants BIT-8994 (ruler at 63%) and SMT-9499 reverted to free energy and nuts. When halted at tick 3,222,226, world fruit was flat at 2.0%. SMT-9499 set an all-time Grid record with 650 carried instructions (24 routines), while the world averaged 544 carried code and 38 run.

What It Means: A Negative Result and Latent Capabilities

The run delivers an instructive negative result: a resource that rewards memory is not enough if nothing leads there step by step.

Under v0.10's physics, traplining requires a coordinated sequence: an organism must record site coordinates (SELF x, SELF y, STORE), maintain an internal clock, and later recall the target and steer toward it (LOAD, BEAR, TURN). This takes roughly seven instructions across two routines. Individually, none of these instructions provides a fitness advantage; storing coordinates gives no return until navigation routines are already wired to read them. Without intermediate payoffs, evolution cannot bridge the gap. Programs evolved eyes for fruit and blossom, but those see only 4 units and sites are 80 units apart, so they found fruit only by chance wandering.

Simultaneously, the run proved that free routine inheritance allows enormous latent genomes to accumulate without metabolic drag. Average carried code reached 544 instructions (95% of routines conditional), yet the world's working mind stayed at 36–40 instructions per tick. Organisms carry vast genetic libraries while running compact, energy-efficient programs. Finally, the emergence of crystal cooking and food caches demonstrates that complex latent behaviors exist within the physics, awaiting selective scaffolding.

Correction: The Fruit Was Never There

After this report was drafted, a direct measurement overturned the committee's reading. Every unit of fruit that grew was eaten, and at many moments the standing crop was zero. The fruit was not rotting unfound; it was barely growing. Fruiting sites were placed on lit ground at genesis, but the Grid's light drifts, and a site in the dark bears nothing. In one test world, 39 of 48 sites were lit at tick 5,000 and only 4 by tick 60,000. For almost the whole run, v0.10's fruit share was capped by supply, not by the programs' wits. The 2% was close to all the fruit there was. The negative result above therefore says less than it seemed: the traplining test was never really run. Fixing the defect (Decision D-024) comes first in v0.11: a site stranded in the dark dies and a new one takes root on lit ground.

What's Still Missing

• Navigational Stepping Stones: A sensory bridge connecting 4-unit sight across the 80-unit gap between fruiting sites.
• Goal-Directed Memory: Clades carried up to 81 memory operations (YRI-9373), but what that memory does is not yet known. It did not build routes to fruit.
• Ecological Diversity: Dynasties (ZUS-2828 → BIT-4313 → APO-7063 → BIT-8994) dominated all regions, while mating remained depressed at 1–2%.

What's Next: v0.11 "Scent"

With fruit flat at 2.0%, Decision D-022 triggers version 0.11 ("Scent").

As Lenski et al. (2003) demonstrated in Avida, complex features evolve by stepping stones, and only when simpler functions are also rewarded. In nature, frugivores use smell as well as memory: ripe fruit releases volatile compounds that guide fruit bats (Hodgkison et al. 2007), and fermentation ethanol may guide primates (Dudley 2000; Dominy 2004).

Version 0.11 introduces scent:

• Fruiting sites emit a scent gradient decaying over 30 units (far beyond 4-unit vision, well short of 80-unit spacing), strengthening as fruit ripens.
• A new log-scale scent eye layer allows organisms to climb gradients using 1–2 steering instructions.

This builds an evolutionary ladder: scavenge overripe leftovers by scent, arrive earlier, and eventually evolve memory to harvest fresh fruit first.

Data: 2026-09-28T14-49-50-758Z_seed-186260746 (archive snapshot: data/scratch/v010-archive-final.json)