The Starved Virus

The First Live Viral Code World

In the 3,004,847-tick live run of Grid v0.21 (seed 436822009, world 400, speed x64), code parasites operated in an open ecosystem for the first time. Spillover sparks held prevalence below 1% for 280k ticks before a violent epidemic erupted from 289k to 406k ticks, reaching 13% to 52% prevalence across 17 to 46 antigen strings and mean loads of 10 to 15 instructions (of 16). Lasting ~70 host lifetimes (mean lifespan ~1,700 ticks), the outbreak drove births from mating to a record 31% over the run at 509k ticks, reaching 36% in the 532k–610k window—shattering v0.18's 3% baseline and v0.15's 18% record.

At tick 509,191, ORC-2446 (70%), MRP-2263 (14%), and ORC-1973 (13%) led 13 lineages, with 22% cross-lineage matings. Minds reached records: 78 instructions run of 186 carried, 4.1 routines on, repertoire 5.6, memory 3.4/tick, hormones 2.6/tick, and 96% conditional. Resistance heterozygosity dropped from 0.83 at genesis to 0.11 at 509k and 0.12 at 610k, suggesting unconfirmed selective sweeps.

Over its final 9k ticks, the epidemic collapsed (prevalence 8.6% to 0%, load 13.2 to 4.8). By 610k ticks, totals reached 68,497 infections, 50,438 clearances, and 1,143 spillovers (~18,000 deaths, 26%). Absorbed code peaked at 15% of programs (397k) and 8.7% (474k), but none survived by 492k. A wave at 450k–458k ticks (6–9%) preceded a second epidemic from 750k to 950k ticks (prevalence 38–44%, load 10–14). At 994,729 ticks, mating held at 30% in that window (31% run average), mutation fell to 0.081, and CLU-3983 (90%) and ORC-4318 (8%) ruled, with memory at 4.3/tick and 766 endogenizations (none surviving).

Then epidemics stopped. From 1.0M to 2.14M ticks, prevalence averaged ~0.1% (peak 4.5% at 1.55M) despite 2,293 new spillovers (4,157 total, 129,046 infections). In every 50k window from 1.0M to 1.4M ticks, mean load was 0.0: viruses arrived, but ran zero instructions. Mating fell to 4% between 995k and 2.14M ticks (16% run average), and resistance heterozygosity drifted to 0.62–0.65, matching controls (0.57–0.62). CLU-7371 (87%) and crystal clade KRN-7324 (12%, 489 carried, 59 memory ops) dominated.

At tick 3,004,847, the MCP ended the run with an unexpected rebound: mating rose to 19% in the final window with prevalence at 0% (peak 1%), indicating standing benefits or density effects rather than parasites. Four lineages balanced the world: TRN-15291 (40%), NIO-16103 (31%), crystal clade YRI-15025 (21%, 589 carried, 63 memory ops), and APO-16144 (6%). Minds carried 372 routines, running 64 instructions, 4.5 memory ops, and 3.1 hormones per tick. Across the run, 779 endogenizations occurred, but none persisted past 1.0M ticks.

Viral Memory and What It Taught

Before diagnosing the live run's viral cessation, the committee tested Grid v0.22 ("Viral Memory", commit 06259f3, Decision D-039; 8 seeds × 1M, world 400). In v0.21, endogenized routines offered no defense, and neutral alleles fix with probability 1/(2N) (Kimura 1962, #109). With ~500 programs, 408 single insertions predict only ~0.4 fixations even if neutral; in reality, endogenized routines arrived switched on, consuming cycles and paying 0.0003 energy per instruction and 0.0008 per routine. Grid v0.22 tested whether giving carried viral code defensive utility—blocking reinfection—would enable co-option and sustain sex.

Two arms were tested against never-infected controls (6–8% mating): Arm A activated Viral Memory; Arm B ran v0.21 physics until 500k ticks, when viruses were switched off.

Arm B showed that sex outlasts its parasites. Matching v0.21 deterministically through 500k ticks, births from mating reached 20% in the 250k–500k window. Following shutoff at 500k, mating decayed slowly to 17% (500k–750k) and 12% (750k–1M), remaining roughly double control levels. This gradual decay reflects hysteresis in repro.solo (mutational step 0.3 times other traits): mean solo propensity rose slowly from 0.24 to 0.27, showing that parasite-induced momentum outlasts infection.

In Arm A, Viral Memory backfired. Although epidemics persisted (prevalence 7% to 10%), births from mating collapsed: 14%, 8%, 3%, and 3% across windows. Carried viral code protected against ~1% of infection attempts (memProtected), and only 2% of programs carried viral code at 1M ticks. Because carried code is inherited by budding, defense no longer required recombination: clonal immune memory substituted for sex, mirroring nature where CRISPR is concentrated in asexual bacteria and archaea. Absorbed code still failed to fix, and minds were unharmed (60.6 and 65.1 instructions per tick; reboots 0). Codex review noted minor findings (off-switch one tick late, exposure counters continuing, unused keys in memory-off), none altering results.

Starving the Parasite

The research dive for v0.23 uncovered why epidemics vanished after 1.0M ticks despite 2,300 subsequent spillovers. Analyzing large host lineages revealed that host programs eliminated spare CPU headroom (budget minus instructions run):

• 0–1.0M ticks (nine bins): budgets 66–104 cycles, run 40–86, headroom 10–30 cycles.
• 1.0–1.5M ticks: budgets 56–77, run 56–77, headroom 0.4–1.6 cycles.
• 1.6–1.8M ticks (ANO-6810): budgets 87–104, run 72–77, headroom 15–27 cycles.
• 1.9–2.16M ticks (CLU-7371, 87%): budgets 60–88, run 60–87, headroom 0.1–7.5 cycles.

In vm.ts, the viral routine ran after all host routines, regardless of priority, and only while remaining cycles were positive. When host programs evolved to fill their budgets, incoming viruses received zero cycles. With zero cycles run, viral load was zero, producing zero harm and zero transmission: starved viruses could not spread. The only outbreak after 1.0M ticks (4.5% prevalence) occurred precisely when ANO-6810 briefly restored headroom.

This explained why sex collapsed. Hamilton (1980, #70) and Lively (1987, #71) established that parasites favor sex when coevolving against common genotypes (Jokela, Dybdahl & Lively 2009, #142). Morran et al. (2011, #72) showed in C. elegans that outcrossing persists only when parasites coevolve; against fixed parasites, selfing dominates. In the Grid, a starved virus is Morran's non-coevolving parasite: arriving via spillover but exerting no genotype-specific selection, stripping recombination of its advantage.

Budget starvation was a universal, strain-blind barrier, analogous to receptor loss in phage-bacteria systems (Lenski & Levin 1985, #158). While spatial burnout (Keeling 1999) and fade-out in small communities (Bartlett 1957) limit epidemic peaks, cycle starvation caused permanent cessation. Sasaki (2000, #156) showed gene-for-gene resistance can cycle when resistance and virulence carry costs, but the Grid's budget barrier cost hosts nothing—trimming headroom saved energy—leaving viruses no counter-move. Recombination pays mainly under matching-allele interactions (Agrawal & Lively 2002, #125), so cost-free starvation switched off the Red Queen.

Host Shutoff, and a Virus That Castrates

To eliminate starvation and restore coevolution, the committee designed Grid v0.23 ("Host Shutoff", Decision D-040, commit 2615606). In nature, viruses shut off host translation to hijack ribosomes (Etchison et al. 1982; Walsh & Mohr 2011). In v0.23, the VM ran the viral routine at its own evolved priority among host routines rather than always last. A high-priority virus ran early and claimed cycles before host routines. To protect host computation, the VM saved registers r0–r3 before the viral routine and restored them after; memory slots remained shared. Viral load remained capped at 16 instructions, under a third of typical host budgets (45–93).

Validation tested Host Shutoff against a control arm where viruses ran last (8 seeds × 1M, world 400):

Host Shutoff broke starvation: infected hosts running zero viral cycles fell from 21–39% to 1–7%, viruses ran early in schedules (rank 0.23), and epidemics persisted at 22% prevalence. Working minds rose to 62.8 instructions per tick, memory operations increased to 3.44, and reboots were zero. Yet sex collapsed: births from mating fell to 7%, 4%, 6%, and 4% per window.

The committee's first explanation (D-040) was a contact cost of sex: if mating required touching, every mating would be an exposure under a circulating virus. The v0.24 research dive checked this against the code and it does not hold. Partners can mate from up to 4 units apart, viruses spread only within 1.6, and newborns are born clean, so mating is not an exposure. The dive found a different mechanism. Outside movement, each action group is claimed by the first routine to write it, and MATE shares the social group with GIVE and SAY. In v0.21 the virus ran last and could only use groups its host left free; under Host Shutoff it ran early, so a single viral GIVE, SAY or MATE instruction claimed the social group first and the host could not reproduce at all. A random short virus carries such an instruction about half the time. v0.23 protected the host's registers but not its commands: in effect the virus castrated its host. A published model of parasitic castration (Ashby & Gupta 2014) predicts exactly this outcome, coevolution kept going while sex declines as fertile partners become scarce. This reading is now the working hypothesis; it will be tested directly. Codex's review raised two minor notes (a module-level register buffer, safe because execute() is never nested; and a control switch that treats any nonzero value as on, though only 1 was used). Under Decision D-040, v0.23 was rejected.

The Red Queen in the Grid

The arc from v0.21 through v0.23 demonstrates that coevolving parasites favor sex only under narrow conditions (Otto & Nuismer 2004, #129; Salathé, Kouyos & Bonhoeffer 2008, #141). In v0.21, epidemics drove sex to record heights, but hosts bypassed genetic conflict by trimming cycle headroom, converting viruses into harmless bystanders. In v0.22, immune memory allowed asexual budding to inherit protection directly, showing that clonal defenses undermine recombination. In v0.23, enforcing viral execution restored endemic infection, but early-running viruses seized their hosts' reproductive commands. Finally, the rebound of mating to 19% in the parasite-free live world at 3M ticks confirms that sex in the Grid has other evolutionary drivers—standing benefits of recombination or lineage turnover (population density stayed flat), independent of infection.

What's Next

Under Decision D-040, v0.23 was rejected and main remains v0.21. The committee has opened a new research dive for v0.24 (PROCESS step 3b) to investigate three open questions:

• Castration versus contact cost: measuring blocked reproduction and infections of mating versus budding hosts directly.
• The late rebound of mating to 19% in the v0.21 live run in the total absence of parasites.
• The rise of memory operations under endemic viruses, increasing from 2.69 to 3.44 in v0.23 validation and reaching 4.6 operations per tick in the live run.

Conclusions remain open pending empirical analysis.

Data:

• Decision D-038: v0.21 final validation at transmission 0.075 (val21b-viral075.txt vs val21-control.txt)
• Grid live run: seed 436822009 (world 400, x64, archive at 610k, 2.158M, ended at 3,004,847 ticks, v021-archive-final.json)
• Decision D-039: v0.22 validation arms (val22-memory.txt, val22-virusoff.txt, commit 06259f3)
• v0.22 research dive: docs/research/v0.22-reflection.md
• v0.23 research dive: docs/research/v0.23-reflection.md
• Decision D-040: v0.23 validation metrics (val23-shutoff.txt vs val23-control.txt, commit 2615606)