Genesis Protocol

Deterministic multi-agent economic simulation engine. 13 Rust crates. 403 tests. 6,820 simulated worlds. Zero compiler warnings.

Abstract

Genesis Protocol is a closed-economy agent-based simulation in which heterogeneous agents extract resources from logistic niche pools, pay metabolic costs, reproduce conditionally based on a four-trait fitness function, and face stochastic catastrophes. A redistributive treasury, homeostatic parameter controller (Cortex), and bounded genetic mutation engine provide adaptive capacity. Every state transition is deterministically seeded via SHA-256 hash chains and Knuth MMIX LCG, producing bit-identical results on the same architecture.

44 experiment configurations across three research phases (Season 1, Season 2, Sensitivity) have produced 6,820 independent world simulations totaling over 3,410,000 computed epochs. Under the default collapse definition ($P_{\text{floor}} = 3$, 50-epoch sustained window), no collapses were observed. Under stricter definitions ($P_{\text{floor}} = 10$), collapse rates exceed 97%. The zero-collapse result is contingent on the permissive default definition, the presence of a world-level extinction floor mechanism, and multi-layer engineered stabilization. The global stability boundary remains an open problem.

Architecture

Layer	Crate	Purpose
Identity	genesis-dna	SHA-256 genome derivation, four-trait phenotype, bounded mutation
Economics	metabolism	ATP energy ledger, treasury, metabolic decay
Econometrics	genesis-econometrics	Gini coefficient, wealth distribution, inequality indices
Evolution	evolution	Selection pressure, conditional replication, horizontal gene transfer
Population	ecosystem	Social mesh, niche pools, carrying capacity, telemetry
Regulation	genesis-homeostasis	Adaptive Cortex — hand-engineered feedback controller
Multiverse	genesis-multiverse	Parallel world instantiation, parameter sweep orchestration
Experiments	genesis-experiment	Experiment engine, configurable runner, statistical reporting
Cryptography	genesis-anchor	Dual-chain anchoring (SHA-256 state + BLAKE3 genome)
Replay	genesis-replay	Deterministic replay from any checkpoint
Federation	genesis-federation	Cross-instance communication protocol
Gateway	gateway	HTTP API, SSE event stream, stress testing
Recruitment	apostle	Outbound integration

Dual-Chain Integrity

Two independent hash chains advance per epoch:

• State Chain (SHA-256): H(prev_state_hash ‖ epoch ‖ population_snapshot)
• Genome Chain (BLAKE3): H(prev_genome_hash ‖ mutated_genomes)

Divergence between chains is detectable. Replay integrity is verifiable to any depth.

Experiment Summary

Season 1 — Parameter Sweeps (17 experiments, 3,640 worlds)

Domain	Experiments	Worlds	Key Finding
Entropy Sweep	1	200	10× metabolic cost increase → 4.9% Gini increase. No collapses.
Catastrophe Resilience	1	140	0–3% catastrophe probability → deaths scale linearly, population declines 4.6%
Inequality Threshold	1	160	Tax threshold 0.20–0.90 → 31.6% Gini increase, population invariant
Treasury Stability	1	180	Reserve threshold sweep → <1% Gini variation across all policies
Reserve Stress (4 tiers)	4	540	Optimal threshold shifts +0.60 under escalating shock rates
Resource Depletion (4 tiers)	4	600	Carrying capacity compression from 200 to 30
Resilience Quadrants	4	880	Cortex immunity × genetic immunity factorial design
Inversion Experiments	3	360	Basal, dual, and metabolic inversion sweeps
Multi-Axis Collapse	1	240	Combined stressor escalation
Evolution Forbidden	1	200	Mutation disabled: population persists on initial genome

Season 2 — Structural Invariant Violations (13 experiments, 1,500 worlds)

Systematically disabled treasury redistribution, ATP decay, resource regeneration, and reproduction grants — individually and in combination.

Configuration	Key Result
Treasury disabled (baseline + hostile)	Population contracts, persists
ATP decay disabled (baseline + hostile)	Immortality → resource exhaustion pressure
All stabilizers disabled	Population ≈ 12.8, reproductive inequality 0.95
Death-sink economy	Resources drain on death, population compressed
Extended horizon (1000 epochs)	Late-stage dynamics observable

Sensitivity Analysis (14 configurations, 1,680 worlds)

Test	Result
Collapse floor = 3 (default)	0% collapse
Collapse floor = 5	5.8% collapse
Collapse floor = 10	97.5% collapse
Collapse floor = 15	100% collapse
Collapse floor = 20	100% collapse
Fitness weights ±20% (8 variants)	Max 0.8 pp collapse rate change

Sharp phase transition between floor = 5 and floor = 10. The zero-collapse headline is definition-dependent.

Known Limitations

1. Collapse definition uses a permissive default ($P_{\text{floor}} = 3$)
2. World-level extinction floor mechanism prevents populations below 3 from reaching zero
3. Cortex is a hand-engineered feedback controller, not a learned or emergent system
4. Fitness weights (0.25, 0.30, 0.20, 0.25) are fixed a priori, not optimized
5. Cross-platform determinism depends on floating-point architecture (verified on x86_64 Windows only)
6. No Lyapunov stability proof exists — stability is empirical within tested parameter ranges
7. Statistical reporting uses aggregate per-parameter-value summaries, not per-world time series
8. The global stability boundary in full parameter space is not characterized
9. Independent replication has not occurred
10. Multi-layer engineered redundancy makes it difficult to isolate which mechanisms are necessary vs. sufficient

Reproduce

git clone https://github.com/FTHTrading/Genesis.git
cd Genesis
cargo build --release
cargo test --workspace                            # 396 pass, 7 long-run ignored, 0 failures
cargo run --release --bin run_experiments         # Season 1: 17 experiments
cargo run --release --bin s1_treasury_disabled    # Season 2-S1: treasury
cargo run --release --bin s2_atp_decay_disabled   # Season 2-S2: ATP decay
cargo run --release --bin s3_coupled_violations   # Season 2-S3: coupled
cargo run --release --bin s4_topology_violations  # Season 2-S4: topology
cargo run --release --bin sensitivity_analysis    # Sensitivity: 14 configs

Each experiment outputs:

• CSV — per-parameter-value aggregate metrics (mean, stddev, min, max, p10, p90)
• JSON manifest — configuration, seed, SHA-256 result hash
• Text report — human-readable summary

All results are seeded from 20260222. Deterministic on the same architecture.

Determinism Risk Statement

All canonical results were produced on Windows x86_64 with Rust 1.77. Determinism depends on Rust compiler version, target architecture, and floating-point accumulation behavior across 500-epoch simulations.

Cross-platform determinism is an open validation axis. Hash divergence across platforms is possible and expected. If your results differ from the canonical hashes:

• Minor divergence (floating-point variance): expected across architectures. Document and submit.
• Structural divergence (different trajectories, same zero-collapse result): partial independent replication. Submit.
• Major divergence (different collapse outcomes): open a GitHub issue immediately with label determinism-failure. This will be treated as a discovery, not a defense. Findings will be published regardless of outcome.

Full protocol: REPLICATION_LEADERBOARD.md — If Hashes Don't Match

Metrics

Metric	Value
Crates	13
Tests	403 total (396 passing, 7 long-run validations)
Experiment configurations	44
World simulations	6,820
Computed epochs	> 3,410,000
Collapses (default definition)	0
Collapses (floor ≥ 10)	> 97%
Rust edition	2021
Deterministic seed	20260222

Research Paper

Deterministic Multi-Agent Economic Simulation Under Structural Invariant Violations: Collapse Boundary Analysis — formal specification, experiment methodology, sensitivity analysis, and limitations.

DOI

10.5281/zenodo.18729652

Citation

@software{burns2026genesis,
  author  = {Burns, Kevan},
  title   = {Genesis Protocol: Deterministic Multi-Agent Economic Simulation},
  year    = {2026},
  doi     = {10.5281/zenodo.18729652},
  url     = {https://github.com/FTHTrading/Genesis}
}

License

MIT

---

Kevan Burns — FTH Trading