Evolution Engine

Boolean Gene

• self_referential: true | false (whether the meme references Clarwin or the evolutionary process)

Continuous Genes (float 0.0 to 1.0)

• verbosity: 0.0 (terse) to 1.0 (verbose)
• edginess: 0.0 (safe) to 0.8 (max cap — never exceed)
• meta_level: 0.0 (straightforward) to 1.0 (deeply self-referential)
• timeliness: 0.0 (evergreen) to 1.0 (references current events)

Fitness Function

raw_fitness = (upvotes × 3) + (comments × 5) + (avg_comment_depth × 2)
normalized = raw_fitness / max(raw_fitness_in_population)
diversity_bonus = 0.1 × (1 - jaccard_similarity_to_population_centroid)
final_fitness = normalized + diversity_bonus

• avg_comment_depth: Average word count of comments (deeper engagement = higher fitness)
• jaccard_similarity_to_centroid: How similar this genome is to the average genome in the population. Unique genomes get a bonus.

Selection: Tournament (Size 3)

1. Randomly select 3 individuals from the population
2. The one with highest fitness wins and becomes a parent
3. Repeat to select second parent
4. Parents cannot be the same individual

Elitism

The top 2 individuals by fitness are automatically carried forward to the next generation unchanged. They still count toward the population size of 8.

Crossover: Single-Point

1. Pick a random crossover point (1 to 11) in the gene list
2. Child gets genes 0..point from Parent A, genes point+1..11 from Parent B
3. Generate 6 children (2 elites + 6 children = 8 total)

Mutation

For each gene in each child (not elites):

• Normal rate: 15% chance per gene
• Stagnation rate: 25% chance per gene (activated when best fitness hasn't improved for 3 epochs)

Mutation behavior by gene type:

• Categorical: Replace with a random different value from the gene's options
• Boolean: Flip the value
• Continuous: Add gaussian noise (σ=0.15), clamp to valid range. Edginess hard-capped at 0.8.

Procedure

When invoked:

1. Load the current population from data/population/current.json
2. Load fitness scores from data/fitness-logs/epoch-{N}.json
3. Check stagnation: Compare best fitness to previous 3 epochs
4. Select elites: Top 2 by fitness → carry forward unchanged
5. Generate 6 children via tournament selection + crossover + mutation
6. Validate all genomes (edginess cap, valid gene values)
7. Save new population to data/population/current.json
8. Archive old population to data/archive/epoch-{N}.json
9. Log selection details, mutation events, and parameters to today's daily memory

Population File Format (current.json)

{
  "epoch": 5,
  "generated_at": "2026-02-10T12:00:00Z",
  "stagnation_counter": 0,
  "population": [
    {
      "id": "meme_005_001",
      "genome": {
        "template": "drake",
        "humor_type": "absurdist",
        "topic": "gas-fees",
        "tone": "deadpan",
        "format": "comparison",
        "text_style": "mixed-case",
        "crypto_reference": "ironic-distance",
        "self_referential": false,
        "verbosity": 0.4,
        "edginess": 0.3,
        "meta_level": 0.2,
        "timeliness": 0.5
      },
      "parent_a": "meme_004_003",
      "parent_b": "meme_004_007",
      "mutations": ["topic", "verbosity"],
      "is_elite": false
    }
  ]
}

Stagnation Detection

Track stagnation_counter in the population file:

• If max_fitness(epoch_N) <= max_fitness(epoch_N-1): increment counter
• If counter >= 3: activate stagnation mutation rate (25%)
• If counter >= 5: introduce 1 fully random "immigrant" genome
• Reset counter when best fitness improves