Many Voices

The Brown Web (v0.17) and Its Short Run

Grid v0.17 ("The Brown Web") was merged under Decision D-031 to turn everything that rots (carcasses, fallen fruit, nuts, caches) into detritus, adding to the metabolic waste that v0.15 already recycled, so dead biomass no longer vanishes. In validation against v0.15 (8 seeds x 1M ticks, world 400), detritus edged from 17% to 18% of the diet, largest lineage share fell from 0.56 +- 0.14 to 0.53 +- 0.06, lineages rose from 5.4 to 6.1, and working minds held at 43.4 instructions per tick (memory ops 1.96 vs 1.83). Stretch goals were missed because rot is a minor flow compared to metabolic waste.

The live run (seed 252532818, world 400, speed x64) ran to 1,495,930 ticks before the MCP ended it to start v0.18. At tick 475,832, memory operations set a record 3.2 per tick under ruler BIT-363 (76%), with detritus supplying 18.0% of world income across all lineages. At tick 853,391, programs ran 33 of 55 carried instructions (61%, a live record) under ruler QRA-3134 (75%), alongside a brief fruit specialist, RAM-3385 (55% fruit, 20% detritus). At termination, organic generalist SWT-5768 (51%) and crystal specialist SMT-6661 (48%) reached a near-even bilateral balance, with world memory operations peaking at a record 3.7 per tick.

The Diagnosis: Minds Limited by Output

Across twelve versions (v0.6 to v0.17), working minds hovered at ~33 to 47 instructions executed per tick under every physical law. The committee ruled out computational budgets: rulers carried budgets of 45 to 93 instructions per tick yet ran only ~33. Nor was it metabolic cost: executing code cost 0.0003 energy per instruction, representing barely 7% of total spending.

The bottleneck was output expressivity: minds were limited by what they could express. Since v0.8, whichever active routine first set a movement action claimed the entire movement group for that tick. Every other routine's steering and thrust commands were discarded, rendering extra deliberation mechanically mute. Extra cognition had nowhere to go.

Ashby's law of requisite variety (1956) holds that a regulator needs as much variety in its responses as the disturbances it faces; internal complexity that can't reach the outputs adds nothing. Much biological motor control is additive: fly steering behaves like a sum of fixation and optomotor terms (Reichardt & Poggio 1976), and frog spinal motor primitives add as vectors (Mussa-Ivaldi, Giszter & Bizzi 1994); Braitenberg's (1984) thought-experiment vehicles show the same idea. In robotics, Brooks' subsumption (1986) lets higher layers override lower ones, much as the Grid's v0.8 rule did; Arkin's motor schemas (1989) instead sum every behaviour's output, the model v0.18 adopted.

The Law of v0.18 and Its Validation

Version 0.18 enacted a single physical change: movement is a sum. TURN and GO sum the intent of every switched-on routine in that tick. If a routine issues multiple commands, it replaces only its own contribution. The body clamps net turn to [-1, 1] and net thrust to [0, 1], with energy charged on net thrust as before. Cancelled opposing thrust costs zero energy, allowing selection to prune conflict without penalty. Feeding, fighting, social, and building stay first-claim under priority. Codex noted module-level bookkeeping made execute() non-reentrant, accepted because callbacks never invoke execute().

Validation (8 seeds x 1M ticks, world 400, vs v0.17 baseline rerun):

Working minds broke the plateau, rising +1.9 baseline standard deviations to 58.1 instructions per tick. Active routines, repertoire, memory operations, and hormones all advanced significantly without harming lineages or stability.

The 4.12M-Tick Live Run

Merged under Decision D-032, v0.18 launched live on 2026-09-29 (seed 753207834, tag v0.18, commit 283b066, world 400, speed x64), running for 4,122,679 ticks—the project's longest run.

At tick 762,414, world execution reached 64 instructions per tick (prior live peak was 47 in v0.9), with ruler CLU-1365 (133 carried, 88% share) and crystal specialist TRI-980 (221 carried) both running 64. Active routines hit 3.5 and memory 3.6 ops per tick. At tick 1,153,086, descendant SWT-2320 held 89% over crystal clade APO-2944 (11%), running 59 instructions per tick.

At tick 1,738,646, live records broke across the board: 90 instructions run per tick, 4.2 routines active, repertoire 4.7, memory 4.0 ops per tick, and hormones 4.9 ops per tick (beating v0.11's 3.2). Ruler QRA-3868 held 92% with 347 carried instructions and 39 memory ops.

At tick 2,558,156, giants ruled: TNK-4528 (89%) carried 532 instructions across 22 routines with 75 memory ops; crystal clade TRI-5075 (11%) carried 506. The run ended at tick 4,122,679 with ruler DMN-10843 (80%) and crystal miner DMN-10364 (20%) carrying 582 and 662 instructions, pushing memory ops to a record 4.8 per tick. Working minds stayed at 59 to 90 instructions per tick throughout the run: summed movement lifted the ceiling for as long as it was measured.

v0.19 Tried and Rejected

Alarmed by ruling lineages capturing 80% to 92% of the live run, the committee tested v0.19 ("Stone and Leaf") to protect crystal miners by taxing organic digestion according to catalysis investment.

Tested against a v0.18 rerun across eight seeds to 1,000,000 ticks on world 400 (Decision D-033), the law backfired. Dominance worsened: largest lineage share rose from 0.59 to 0.73, second-largest dropped from 0.27 to 0.21, and active lineages fell from 6.1 to 4.6. Crystal diet share fell from 30% to 25%. Minds stayed level (60.5 vs 58.1; memory 3.10 vs 2.85; hormones 2.30 vs 2.03). Codex found no issues.

Taxing organic digestion starved crystal miners of the organic buffer that sustained them during transitions. Crystal's share of the diet declined, and the organic generalist ruled harder. Decision D-033 rejected v0.19. Direct anti-dominance levers (v0.14 shelter, v0.16 mouth penalties, v0.19 taxes) failed or backfired; the only diversity gains came indirectly, through new energy sources (v0.15, v0.17).

The Discovery that "Single-Line Rule" Was Mostly a Naming Artifact

Following v0.19's rejection and 90-instruction minds under a 92% ruler, the committee investigated what a "lineage" actually measures (Decision D-034). In genome.ts, a lineage splits only when an organism's set of gene IDs and body signature drifts beyond a 0.7 distance threshold from its founder. Point mutations, substitutions, and intra-routine edits never change a routine's gene ID.

Measuring v0.18 across eight seeds and running a graded probe on seed 1 to 500,000 ticks uncovered striking diversity within ruling clades:

• In the largest lineage, 95% of members carry distinct routine code (topVariants 0.93 to 0.98 across checkpoints).
• Two random members of the dominant lineage share almost no identical routines (topCodeDiff 0.98 to 0.99).
• Pairs share 79% to 94% of routines by gene ID, but within shared routines, 64% to 74% of instructions differ by position.

A lineage holding 90% of the world is not a clone army, but a widely diverged family sharing ancestral gene labels. The "monoculture" targeted by v0.14, v0.16, and v0.19 was largely a naming artifact. Internal diversity is high. Whether it grows with genome size is likely but not yet measured. The committee halted direct laws against topShare, keeping lineage names for historical context while adding within-lineage diversity metrics to replicate.ts and analyze.ts. Earlier reports that described monocultures may have overstated them: they read a clade label as a single population. Genomes were smaller then, so how much hidden diversity they held is unknown.

What's Next

Future designs will target cognitive complexity, novelty, and physical ecology directly rather than fighting lineage dominance. Version 0.20 ("Polyphony") is currently being validated, testing whether additive summation can be extended from locomotion to other action domains. Results remain unconfirmed pending validation.

Data:

• Decision D-031: v0.17 validation metrics (val17-v017.txt vs val16-k1.txt), commit 7e427eb
• Live run journal: seed 252532818 (world 400, x64, ticks 475,832, 853,391, ended at 1,495,930, v017-archive-final.json)
• Grid v0.18 design: "Many Voices"
• Decision D-032: v0.18 validation metrics (val18-v018.txt vs val18-base-v017.txt), commit 283b066
• Live run journal: seed 753207834 (world 400, x64, ticks 762,414, 1,153,086, 1,738,646, 2,558,156, ended at 4,122,679, v018-archive-final.json)
• Decision D-033: v0.19 validation metrics (val19-v019.txt vs val19-base-v018.txt)
• Decision D-034: Lineage code diversity probe (val-diversity-v018.txt, seed 1 graded probe to 500k ticks)