Waste into Food
Version 0.15 ("Waste into Food") built the Grid's first decomposer economy by recycling a quarter of metabolic spending back into occupied cells as decaying detritus, digested through a heritable mouth enzyme with a linear trade-off against other food. In validation against v0.13 across eight seeds at one million ticks, detritus captured 17% of total diet and lowered the dominant lineage's share from 0.67 to 0.56, weakening single-line dominance for the first time. Although the live run of seed 20473391 seemed to show stubborn monoculture at 87% with every lineage taking an identical 16% to 18% detritus slice, an attempted fix in v0.16 ("Master of None") and its three-arm test revealed a committee misdiagnosis: a third of all programs in v0.15 had already evolved detritus-tuned mouths, but their diets looked identical because detritus was strictly supply-limited.
Why Waste
Four successive economic mechanisms—scent trails, seasonal winter, buried caches, and territorial shelter—failed to dislodge single-line rule, moving minds little or not at all. Monoculture persisted because every lineage competed for the same primary resources under identical metabolic rules. The sole historical coexistence occurred in v0.13 at 1.5 million ticks, when a generalist clade (50%) and a crystal specialist (45%) partitioned the world by feeding on entirely different energy sources.
Ecology illustrates the same principle. In E. coli evolution experiments, glucose specialists excrete acetate and glycerol that other clones evolve to live on (Rosenzweig et al. 1994, in chemostats), and in the long-term evolution experiment a second type persisted on metabolites secreted by the first, under negative frequency dependence (Rozen & Lenski 2000). In soils and seas, the "brown" detrital food web often carries more energy than primary producers (Moore et al. 2004). Because enzyme budgets are finite, specializing on one substrate impairs digestion of others. Digesting metabolic waste could thus foster a secondary economy underneath ruling generalists.
The Law
Version 0.15 introduced a single physical change:
- Detritus: A quarter of every program's metabolic spending (existence, mind, senses, movement, upkeep, stress, healing, foraging, fighting, eroding, marking; excluding energy placed into matter, caches, or gifts) is left in its cell as detritus instead of dissipating as heat. Detritus decays at 1% per tick, vanishing within a few hundred ticks unless eaten.
- Mouth enzyme: Mouths gain a heritable detritivory value (0 to 1; initialized randomly between 0 and 0.1 at genesis, drifting like catalysis and detox). A mouth digests detritus in proportion to detritivory (alpha) and every other food in proportion to (1 - alpha). Specializing in waste costs efficiency on all fresh food.
- Consumption: Detritus is gathered with the standard EAT action and contested normally. Consuming one's own waste returns at most 25% of expended energy, conserving energy and naturally damping self-feeding loops.
Validation
Validation tested v0.15 against a rerun of v0.13 across eight seeds to 1,000,000 ticks on world 400. In the initial pass, detritus share printed as 0% due to a tooling omission in replicate.ts, which listed diet keys explicitly. Once fixed in commit cc0af8b, a rerun on the same seeds reproduced every other metric identically, verifying determinism across runs. A Codex review flagged two findings: the heat-accounting item was fixed in commit d84e8c4, while trace cleanup was maintained.
Validation Results (8 seeds x 1M ticks, world 400):
| Metric | v0.13 Baseline | v0.15 Result | Pass Criterion | Status |
|---|---|---|---|---|
| System Reboots | 0 | 0 | 0 | Pass |
| Detritus Share of Diet | 0% | 17% | >= 5% | Pass |
| Largest Lineage Share (1M) | 0.67 | 0.56 | < 0.67 | Pass |
| Largest Lineage Share (750k) | 0.79 | 0.73 | Historical reference | - |
| Active Lineages | 5.25 | 5.38 | > 5.25 | Pass |
| Working Mind Instructions | 47.2 +- 13.0 | 42.6 +- 3.8 | Within 1 sd | Pass |
| Memory Operations / Tick | 1.95 | 1.83 | Within 1 sd | Pass |
| Novel Behaviours | 66.6 | 69.8 | Baseline comparison | - |
| Mating Births / Checkpoint | 13.1 | 22.6 | Baseline comparison | - |
| Crystal Diet Share | 32% | 32% | Baseline comparison | - |
Detritus provided 17% of total diet from 250k onward—the largest novel food source added in any version. Dominant lineage share dropped to 0.56, meeting all criteria and marking the first law to weaken single-line rule in validation.
The Live Run
The live run began on 2026-09-29 at 03:25 by the MCP (tag v0.15, main commit 2eb0328, seed 20473391, world 400, speed x64).
At tick 518,026, detritus comprised 17.1% of global income, ranking third behind free energy (45%) and crystal (19%). Mating reached 18% of births, the highest in any live run (surpassing v0.12's 15%). Monoculture held firm: ruler TRI-1793 commanded 88% of the world with an 18% detritus diet, while crystal clade QRA-2106 held 11% with 16% detritus. TRI-1793 ran lean with 53 instructions and 4.6 memory ops. Globally, programs executed 33 of 71 carried instructions (46%), activating 2.8 routines from a repertoire of 3.8, with 1.0 memory ops and 0.8 hormone ops per tick across 80 innovations and 3 lineages. The novelty detector was updated to track detritivores.
At tick 1,208,726, TRI-1793 retained 87% dominance (ruling since tick ~529k) on 53% free energy, 18% detritus, and 14% nuts, while crystal clades SRK-5489 (9%) and SRK-5443 (4%) each consumed 16% detritus. Across 6 lineages and 39 innovations, carried code reached 142 while active instructions stayed at 35; routines were 96% conditional, with memory at 1.7 ops/tick and hormones at 1.0 ops/tick.
At 1,354,645 ticks the live run ended abruptly. Three v0.16 validation arms were running in parallel (24 replicate worlds of about 2 GB each, alongside the 4.6 GB live server), the machine ran short of memory, and Claude Code stopped the server's shell. The MCP then stopped the orphaned server process to protect the machine. The new rule is at most about 16 replicate worlds at once beside a live server, and a reaped server is restarted only on the founder's say-so.
The v0.16 Experiment and the Supply Illusion
Observing that every lineage in the live run took 16% to 18% of its diet from detritus, the committee inferred that linear trade-offs let generalists graze both foods without penalty. Applying Levins' (1968) fitness-set theory ("jack of all trades, master of none"), v0.16 proposed a penalty that makes intermediate mouths worse than the average of two specialists (a concave fitness set in Levins' terms), where non-detritus food digestion scaled as (1 - alpha)^2 while detritus digestion remained alpha.
Gemini noted that squaring the penalty on small alpha values could penalize early mutational drift away from the ancestral 0 to 0.1 range. A three-arm validation was executed across eight seeds to 1,000,000 ticks: Arm 1 tested the linear baseline (K = 1); Arm 2 tested the quadratic penalty (K = 2); Arm 3 tested K = 1.5 with halved detritus decay (0.005). The harness also tracked detSpec (programs whose primary mouth had detritivory > 0.5) and detMean.
Three-Arm Test Results (8 seeds x 1M ticks, world 400):
| Metric | Arm 1: K=1 (v0.15) | Arm 2: K=2 (v0.16) | Arm 3: K=1.5 / Slow Decay |
|---|---|---|---|
| Detritus Diet Share | 17% | 17% | 17% |
| Detritus Specialists (detSpec) | 0.34 | 0.30 | 0.29 |
| Largest Lineage Share | 0.56 | 0.69 | 0.56 |
| Active Lineages | 5.4 | 5.3 | 8.1 +- 5.9 |
| Working Mind | 42.6 | 39.8 | 43.4 |
| Memory Ops / Tick | 1.83 | 2.25 | 1.99 |
| Reboots | 0 | 0 | 0 |
The results refuted v0.16. Convex penalties failed to produce specialists; K = 2 reduced detSpec from 0.34 to 0.30 and pushed dominant lineage share up to 0.69. More importantly, the test revealed that v0.15 already fostered specialists: 34% of organisms possessed detritus-tuned mouths, with a mean detritivory of 0.44. The live run diagnosis was an artifact of supply limitation: essentially all shed detritus is eaten, capping detritus intake at ~17% however it is shared out among lineages (as with fruit in v0.10).
What It Means
The brown food web is real and viable on the Grid. Converting metabolic dissipation into food established an alternative energy channel that supplied 17% of the diet and loosened monoculture dominance from 0.67 to 0.56 in validation. Methodologically, the experiment highlights a crucial lesson: measuring diet alone can mislead when a resource is supply-limited. Competing lineages will consume identical fractions of an exhausted resource even when a third of the population carries specialized digestive morphology. A steeper cost on a supply-capped food didn't add specialists, and in validation the ruling lineage's share rose (0.69 vs 0.56); why is not yet measured.
What's Next
Version 0.16 has been rejected, with its branch preserved in the repository, and v0.15 remains the active standard. Having untangled the dynamics of supply limits and morphological traits, the committee is currently selecting the next design for the Grid.
Data:
- v0.15 design and seed 3 probe (150k ticks)
- Decision D-029: v0.15 validation metrics (val15-v015.txt, val15b-v015.txt vs val14-base-v013.txt), commits cc0af8b, d84e8c4, 2eb0328
- Live run journal: seed 20473391 (world 400, x64, ticks 518,026, 1,208,726, ended at 1,354,645)
- Decision D-030: v0.16 three-arm validation (val16-k1.txt, val16-k2.txt, val16-k15.txt)