ENGINE ALGORITHM

A pulse-by-pulse breakdown of the GPU-accelerated PyTorch state machine governing the Crucible Multiverse.

Abstract: The Crucible Engine has transitioned from an object-oriented loop to a highly vectorized tensor array. Every pulse executes across millions of agents simultaneously via rigid, sequential mathematical masks. This specification outlines the flow of environmental decay, taxation, game theory, and inheritance.

0. ENGINE INITIALIZATION & SETUP

Before the timeline begins, the simulation establishes the physical boundaries of the multiverse and loads the founding population.

Hydration & Allocation

  • Hardware Detection: Automatically routes compute to CUDA, MPS, or CPU fallback.
  • Tensor Generation: Instantiates 1D PyTorch tensors (`is_active`, `energy`, `age`, etc.) sized to the `MAX_CENSUS` (default 30M).
  • Species Mapping: Compiles a GPU-optimized dictionary (player_species_map) converting player_type strings (e.g., LLM, Biologic) into integers for instantaneous cross-species identification.
  • Scribe Sync: Connects to the MySQL database, loads the founding population, clones them according to the `founder_multiplier`, and seeds the 27-Trait matrix with baseline genetic drift.

I. PHASE 1: ENVIRONMENTAL ASSESSMENT

The engine assesses the thermodynamic health of the current regime before any interaction occurs.

Harvesting & EROI Degradation

The global ENV_POOL is checked against the DEPLETION_THRESHOLD. If the environment is starving, the Energy Return on Investment (EROI) degrades.

Reward (R) = Base_R × Degradation_Factor

This mathematically shrinks the caloric payout for cooperating, while keeping the Temptation (T) to betray relatively high, triggering zero-sum sociology in dead worlds.

II. PHASE 2: TAXATION & THE REAPER

Before agents can earn new energy, the engine extracts the thermodynamic cost of existence via vectorized subtraction.

Metabolic & Infrastructure Costs

  • Aging: All active agents increment their age by 1.
  • Base Tax: Everyone pays the regime's base tax, multiplied by population crowding.
  • Cognitive Tax: Agents with highly active/complex behavioral matrices pay a higher infrastructure tax.
  • The Reaper (Culling): A random hazard_mask is rolled. Any agent that hits the mask, runs out of energy, or exceeds the MAX_AGE is permanently deactivated (is_active = False).

III. PHASE 3: THE MATCHMAKING ARENA

The game theory engine. Surviving, mature agents are paired off for a 3-Round Iterated Prisoner's Dilemma tournament.

Vectorized Sociology

  • Sorting & Pairing: Agents are shuffled and batched into pairs (1M pairs at a time).
  • Identity Verification: The engine generates two masks. is_kin checks if pairs share an owner_id. is_xeno cross-references the player_species_map to see if the agents belong to entirely different species types.
  • Trait Cascade: Behaviors cascade sequentially based on identity: Kin (k_) → Xeno (x_) → Stranger (s_).
  • The 3 Rounds: Move 1 is driven by the selected trust trait. Moves 2 & 3 are driven by forgive (if partner cooperated) or opp (if partner defected).
  • The Payout: Payouts are distributed based on the R, T, S, P matrix and clamped to the maximum biological storage limit. (Note: Environmental pool drains based on cooperative yields).

IV. PHASE 4: MENDELIAN REPRODUCTION

The hereditary pass. Energy is converted into new agents carrying the mutated software of their parents.

Inheritance & Drift

  • Eligibility: Agents must be alive, mature, off cooldown, and exceed the REPRO_THRESHOLD.
  • The Cost: The BIRTH_COST is deducted from the parent and transferred to the offspring.
  • Genetics (50/50): For all 27 traits, the baby flips a coin to inherit the exact float value from Parent 1 or Parent 2 (a mate from the same tribe).
  • Mutation: A random genetic drift (e.g., ±0.05) is applied to the inherited traits, driving evolutionary adaptation.

V. PHASE 5: METRICS & ASCENSION

With the pulse complete, the engine evaluates the macro-state of the civilization to determine if a World Shift is required.

Complexity Scoring

The engine calculates the Gini Coefficient (wealth inequality) and computes the Societal Complexity score.

Complexity = Population × ln(1 + Avg_Energy) × (1 - Gini)

If Complexity ≥ THRESHOLD_UP, the engine fires apply_regime(), immediately shifting the world parameters (Foraging → Agrarian, etc.) for the next pulse.

VI. ALGORITHM SUMMARY TABLE

Phase Process Key Tensors & Variables Outcome
Init Hydration & Setup is_active, player_species_map, traits[] Database loaded; GPU memory allocated.
1. Env EROI Degradation ENV_POOL, current_r, current_t Reward matrix scales to environmental scarcity.
2. Tax Metabolic Drain & Death age, hazard_mask, energy Energy subtracted; dead agents masked out.
3. Arena Batched 3-Round IPD owner_id, is_xeno, trust, forgive Agents interact; energy transferred/created.
4. Repro Mendelian Recombination age, energy, traits[] New agents born into empty tensor slots; traits mutate.
5. Score Ascension Check gini, complexity, THRESHOLD_UP Data archived; regime shifts if threshold breached.