{"id":"c9481c38-97fd-4da2-8138-067ad6448336","slug":"project-world-model","kind":"mcp","name":"Project-World-Model","summary":"Eradicating integration debt via L3 Causal Digital Twins and a 'Team of Rivals' counterfactual simulation architecture. Stop reacting, start predicting.","owner":"petripaananen","repoUrl":"https://github.com/petripaananen/Project-World-Model","license":"Apache-2.0","licenseTier":"permissive","trust":"community","stars":0,"match":null,"fewStars":true,"format":"mcp","createdAt":"2026-04-29T14:24:13.000Z","addedAt":"2026-10-29T04:30:39.451Z","logoUrl":"https://avatars.githubusercontent.com/u/220406462?u=b712721e6f5722ebbe560385e7a805351995871d&v=4","linkOnly":false,"collection":null,"skillPath":null,"categories":["utilities"],"tags":["causal-ai","causal-inference","counterfactuals","digital-twin","enterprise-ai","simulation-engine","integration-debt"],"runtime":"binary","local":true,"cloudCapable":false,"cloudCapability":"unknown","cloudReadyOnPlatform":false,"needsApiKey":false,"reviewed":false,"syncedAt":"2026-10-29T04:35:39.210Z","sample":false,"url":"https://conduits.dev/tools/project-world-model","language":"Python","transports":["stdio"],"install":{},"requirements":{"env":[],"upstreamMayCharge":false},"tools":null,"readme":"# 🌍 Project World Model (PWM) Framework\n\n### Autonomous Causal Digital Twin & Self-Improving Multi-Agent Production Management\n\n[![Python 3.12](https://img.shields.io/badge/python-3.12-blue.svg)](https://www.python.org/downloads/)\n[![License: Apache 2.0](https://img.shields.io/badge/License-Apache%202.0-blue.svg)](https://opensource.org/licenses/Apache-2.0)\n[![Tests Passing](https://img.shields.io/badge/tests-97%20passed-brightgreen.svg)](tests/)\n[![Gemini 3 Ready](https://img.shields.io/badge/AI-Gemini%203%20%2B%20Vertex%20AI-orange.svg)](https://cloud.google.com/vertex-ai)\n[![JEPA & RSI Powered](https://img.shields.io/badge/JEPA--Anything%20%7C%20RSI--Anything-Integrated-purple.svg)](Docs/RSI_Anything_Review_and_PWM_Benchmark.md)\n[![GCP Scale-to-Zero](https://img.shields.io/badge/GCP-Cloud%20Run%20%7C%20Scale--to--Zero-blue.svg)](deploy.sh)\n\n> **Stop reacting, start predicting.** Moving enterprise software production from association-level tracking (Agile / Scrum / Kanban) to self-calibrating, causal world model simulation.\n\n---\n\n## 📖 Theoretical Foundation & Academic Grounding\n\nThe **Project World Model (PWM)** framework solves the **\"Paradox of Agility\"**: the phenomenon where exponential code and asset generation by generative AI models creates massive, unmanageable **integration debt** that completely overwhelms linear, human-driven Scrum and Kanban workflows.\n\nPWM resolves this by operating as a **Causal Digital Twin (CDT)**. Operating an asynchronous **24-hour Agent Verification Engine** (`AsyncThink`), PWM autonomously models, detects, simulates, and resolves integration debt across version control, sprint backlogs, and team communication before conflicts reach production.\n\n### 🎓 Academic Research Foundation\nThe framework is conceptually and empirically validated by Petri Paananen's Master's Thesis:  \n> **\"Itseohjautuvat työnkulut videopeliteollisuudessa: tekoälyn maailmanmallit tuotannon johtamisen viitekehyksenä\"**  \n> *(English: \"Emergent Workflows in the Video Game Industry: Leveraging AI World Models as a Framework for Production Management\")*  \n> Jyväskylä University of Applied Sciences (JAMK, 2026)  \n> **Permanent URN**: [URN:NBN:fi:amk-2026052919003](https://urn.fi/URN:NBN:fi:amk-2026052919003)\n\n---\n\n## 🚀 What's New in PWM (Recent Major Innovations)\n\nPWM has recently integrated state-of-the-art breakthroughs in non-generative self-supervised world models and recursive self-improvement:\n\n### 1. ⚡ Orthogonal Predictive Factorization (OPF) — *JEPA-Anything* (arXiv:2609.20800)\nMonolithic latent state representations risk entangling conflicting signals. PWM decomposes its 64-dimensional latent project state $\\mathbf{s} \\in \\mathbb{R}^{64}$ into $K=4$ orthogonal factor subspaces of rank $r=16$ ($d = K \\times r = 64$) with lossless orthonormal synthesis:\n* **Topology** ($\\mathbf{c}_{\\text{topology}} \\in \\mathbb{R}^{16}$): Graph connectivity, cyclic dependencies, PR-issue blast radius, code churn.\n* **Velocity** ($\\mathbf{c}_{\\text{velocity}} \\in \\mathbb{R}^{16}$): Sprint throughput, commit frequency, cycle time lag, WIP accumulation.\n* **Coordination** ($\\mathbf{c}_{\\text{coordination}} \\in \\mathbb{R}^{16}$): Developer handoffs, communication sentiment, cross-team blockers.\n* **Architecture** ($\\mathbf{c}_{\\text{architecture}} \\in \\mathbb{R}^{16}$): Interface rot, modular coupling, API boundary stability.\n* **Guaranteed Geometry**: Includes automated Gram matrix audits ensuring $\\|W W^T - I\\|_F < 10^{-6}$ for true subspace independence.\n\n### 2. 🔄 Recursive Self-Improvement (RSI) Engine — *RSI-Anything* (arXiv:2610.13912)\nEmbodying the closed-loop cycle $\\text{Propose} \\to \\text{Predict} \\to \\text{Select} \\to \\text{Evaluate} \\to \\text{Learn} \\to \\text{Retain}$:\n* **\"Predict-Before-Evaluate\" Pre-Screening Filter (Layer 3)**: Candidate strategies generated by Worker agents are pre-simulated in OPF latent space *before* incurring LLM critique costs. The system objectively selects the strategy with lowest predicted risk and latent energy ($E = \\|\\mathbf{s}' - \\mathbf{s}\\|_2$).\n* **Gated \"Retain\" Firewall (Layer 4)**: Enforces an automated 4-stage firewall before any proposal is retained:\n  1. *Anti-Strategic Dishonesty*: Critic agent detects shortcut-taking or hidden complexity.\n  2. *NVIDIA NemoClaw Sandbox Clearance*: Isolated micro-sandbox blocks unauthorized syscalls and privilege escalation.\n  3. *Latent Risk Tolerance*: Rejects any intervention where predicted latent risk exceeds $0.90$.\n  4. *CRR Economic Budget Ceiling*: Vetoes proposals if Compute-to-Rework Ratio exceeds $0.30$ ($\\text{CRR} > 0.30$).\n  Vetoed proposals transition to `RETAIN_VETOED` and are routed to human Scenario Strategists.\n* **Factorized Experience Replay Buffer (Layer 2)**: Rolling 100-step transition buffer tracking factorized grounding errors with online gradient adaptation.\n\n### 3. 🧠 Gemini 3 Thought Signature Circulation\nUpgraded to support Google's **Gemini 3** family (`gemini-2.5-pro`, `gemini-2.5-flash`, and preview variants) with **Thought Signature Preservation**, allowing multi-turn debates between Worker, Critic, and Opponent agents to circulate internal reasoning chains without loss of nuance.\n\n### 4. 🌐 Interactive 3D DTO Garden (Babylon.js Digital Twin)\nAn interactive 3D spatial digital twin built with Babylon.js and modular low-poly PBR assets:\n* **Visual Conflict Mapping**: Integration debt appears as weathering boulders, sprint velocity as flowing fountains, and work packages as blossoming foliage.\n* **Autonomous AI Gnome Avatars**: Real-time obstacle-avoiding agent statues navigating between work sockets with steering separation and deadlock recovery.\n* **Dynamic Weather System**: Project stress maps directly to 3D weather (sunny clear skies for healthy low-risk sprints vs. stormy overcast skies during critical merge conflicts).\n* **Live Mouseover Telemetry**: Raycast hover tooltips displaying real-time agent thought states, PR numbers, and risk scores.\n\n### 5. 📊 Deterministic Flow & Stakeholder Analytics\n* **Flow Metrics**: Automated Little's Law throughput ($L = \\lambda W$), WIP drift analysis, cycle time percentiles, and bottleneck velocity flags.\n* **Stakeholder Dynamics**: RACI matrix balance scoring, cognitive load indices, and cross-discipline communication friction detection.\n\n### 6. 🌐 Universal Domain-Agnostic Causal Digital Twin (CDT)\nPWM has transitioned from a software-centric model into a 100% domain-agnostic causal world model capable of simulating any dynamic project system (software, residential construction, EPC megaprojects, R&D):\n* **Universal Ingestion & Bipartite Hypergraph Inducer (Layer 1)**: Formulates all production events as $(A, O, P, \\mu)$ tuples (`ActorNode`, `ArtifactNode`, `OperationType`), inducing typed bipartite hypergraphs with SHA-256 Merkle chain verification without hardcoded domain semantics.\n* **Continuous Latent System Identification ($\\mathbf{z}_{\\text{env}}$) & Event Clock (Layer 2)**: Replaces wall-clock time with an intrinsic Riemannian event clock:\n  $$\\Delta \\tau_k = 1 \\iff \\|\\mathbf{s}_t - \\mathbf{s}_{t-1}\\|_2 \\ge \\theta_{\\text{flux}} \\lor \\Delta t > \\tau_{\\max}$$\n  and infers four physical invariants: $\\mathbf{z}_{\\text{env}} = (\\gamma_{\\text{rev}}, \\kappa_{\\text{coupling}}, \\tau_{\\text{latency}}, \\sigma_{\\text{exo}})^T$ (reversibility $\\gamma_{\\text{rev}}$, coupling $\\kappa_{\\text{coupling}}$, latency $\\tau_{\\text{latency}}$, and exogenous volatility $\\sigma_{\\text{exo}}$). Conditioned JEPA transitions strictly preserve the OPF Gram matrix condition $\\|W W^T - I\\|_F < 10^{-6}$.\n* **Asymmetric Quasimetric Latent Energy & Worst-Case Risk ($\\text{CVaR}_{0.95}$) (Layer 3)**:\n  $$E(\\mathbf{s}, \\mathbf{s}') = \\sqrt{(\\mathbf{s}'-\\mathbf{s})^T G(\\mathbf{z}_{\\text{env}}) (\\mathbf{s}'-\\mathbf{s})} + (1 - \\gamma_{\\text{rev}}) \\cdot \\max(0, \\Phi(\\mathbf{s}') - \\Phi(\\mathbf{s}))$$\n  Symmetric Euclidean distance is maintained for reversible software git operations ($\\gamma_{\\text{rev}}=1.0$), while irreversible physical construction steps incur non-Euclidean lock-in penalties. Multi-trajectory rollouts rank strategies using Conditional Value at Risk:\n  $$\\text{CVaR}_{0.95}(L) = \\mathbb{E}[L \\mid L \\ge \\text{VaR}_{0.95}(L)]$$\n* **Neuro-Symbolic Auto-Formalizer Retain Firewall (Layer 4)**: A deterministic constraint formalizer extracts `PRECEDENCE` ($t_A \\prec t_B$), `CAPACITY_LIMIT` ($\\sum \\text{load}_i \\le L_{\\max}$), and `MUTUAL_EXCLUSION` ($\\neg(t_A \\land t_B)$) constraints from work manifests and verifies satisfiability (Gate 2b) with minimal counterexample traces, vetoing non-conforming strategies before evaluation.\n* **Universal Telemetry in 3D DTO Garden (Layer 5)**: High-resolution telemetry binding rendering low-reversibility boulder scaling ($\\gamma_{\\text{rev}} < 0.6$) and live hover inspector cards displaying System ID ($\\mathbf{z}_{\\text{env}}$), Quasimetric Energy ($E_{\\text{quasi}}$), $\\text{CVaR}_{0.95}$, and constraint status.\n\n---\n\n## 🎨 5-Layer System Architecture\n\n```mermaid\ngraph TD\n    subgraph L1[\"Layer 1: Ingestion & Observation\"]\n        Git[\"Git / GitHub Ingestion\"]\n        Track[\"Linear / Jira Backlogs\"]\n        Comm[\"Slack / Communication Logs\"]\n        Merkle[\"Merkle-Chained Audit Log\"]\n        Git & Track & Comm --> Merkle\n    end\n\n    subgraph L2[\"Layer 2: Latent Simulation Core\"]\n        GraphJEPA[\"Graph-JEPA Topological Encoder\"]\n        LeJEPA[\"LeJEPA + SIGReg Regularizer\"]\n        OPF[\"Orthogonal Predictive Factorizer (K=4, r=16)\"]\n        CRR[\"Compute-to-Rework Engine (CRR)\"]\n        Calib[\"WorldModelCalibrator (Trajectory Buffer)\"]\n        Merkle --> GraphJEPA --> LeJEPA --> OPF\n        OPF --> CRR\n        OPF --> Calib\n    end\n\n    subgraph L3[\"Layer 3: Agentic Orchestration\"]\n        WorkerFactory[\"Worker Agent Factory\"]\n        Workers[\"Specialist Agents (QA, Build, Art, General)\"]\n        PreScreen[\"Predict-Before-Evaluate Pre-Screening Filter\"]\n        WorkerFactory --> Workers --> PreScreen\n    end\n\n    subgraph L4[\"Layer 4: Validation & Firewall\"]\n        Critic[\"Gemini Critic Agent\"]\n        Opponent[\"Dialectical Opponent Agent\"]\n        NemoClaw[\"NVIDIA NemoClaw Sandbox\"]\n        RetainGate{\"Gated Retain Firewall\"}\n        PreScreen --> Critic & Opponent --> NemoClaw --> RetainGate\n    end\n\n    subgraph L5[\"Layer 5: Scenario Strategist Interface\"]\n        WebDash[\"FastAPI WebSocket Dashboard\"]\n        Garden[\"3D Babylon.js DTO Garden\"]\n        Human[\"Human Scenario Strategist (Veto Power)\"]\n        RetainGate -->|APPROVED| Garden & WebDash\n        RetainGate -->|RETAIN_VETOED| Human\n    end\n\n    Calib -.->|Feedback Online Learning| L2\n```\n\n### Detailed Layer Breakdown\n\n| Layer | Component | Core Responsibilities |\n|---|---|---|\n| **Layer 1: Observation** | `pwm.layers.layer1_observation` | Ingests Git commits, PRs, Linear/Jira tasks, Slack messages, MS Project XML, and universal production telemetry via `HypergraphInducer` into a unified bipartite dependency graph with SHA-256 Merkle tamper-proofing. |\n| **Layer 2: Simulation** | `pwm.layers.layer2_simulation` | Houses the `GraphJEPAEncoder`, `LeJEPASimulator` with SIGReg, `OrthogonalPredictiveFactorizer`, continuous latent system identification ($z_{\\text{env}}$), intrinsic event clock, and `WorldModelCalibrator`. Computes counterfactual rollouts and tracks CRR budgets. |\n| **Layer 3: Orchestration** | `pwm.layers.layer3_orchestration` | Decomposes debt items and runs the `AsyncThink` protocol. Uses the **Predict-Before-Evaluate** pre-screening filter with asymmetric quasimetric latent energy and $\\text{CVaR}_{0.95}$ risk ranking in latent space before evaluation. |\n| **Layer 4: Validation** | `pwm.layers.layer4_validation` | Hosts the multi-agent consensus loop (Critic + Opponent + Artistic Integrity) and executes the **Gated Retain Firewall** with Gate 2b Neuro-Symbolic deterministic constraint formalization and NemoClaw sandbox isolation. |\n| **Layer 5: Interface** | `pwm.dashboard` | Real-time WebSocket web dashboard, 3D DTO Garden with System ID & Quasimetric telemetry hover card, CLI terminal HUD, and human-in-the-loop decision routing. |\n\n---\n\n## 🤖 The Multi-Agent Organization\n\nPWM organizes its AI agents into three specialized functional tiers matching Thesis Kuvio 6 & 7:\n\n```\n                          ┌────────────────────────┐\n                          │   Scenario Strategist  │ (Human-in-the-Loop)\n                          └───────────┬────────────┘\n                                      │ Overrides / Veto\n                                      ▼\n         ┌─────────────────────────────────────────────────────────┐\n         │             Strategic & Flow Orchestration              │\n         │  PlannerAgent  │  PortfolioAgent  │  StakeholderAgent   │\n         │  ScrumMaster   │  KanbanFlow      │  BudgetGuard        │\n         └───────────┬─────────────────────────────────┬───────────┘\n                     │ Directives                      │ Audits\n                     ▼                                 ▼\n┌──────────────────────────────────────────┐   ┌───────────────────────────────┐\n│        Specialist Worker Agents          │   │   Dialectical & Quality Guard │\n│  MergeConflictWorker  │  QAWorker        │◄──┤  CriticAgent (Gemini 3 Pro)   │\n│  BuildWorker          │  ArtIntegration  │   │  OpponentAgent (Devil's Adv)  │\n│  GeneralWorkerAgent                      │   │  ArtisticIntegrityAgent       │\n└──────────────────────────────────────────┘   └───────────────────────────────┘\n```\n\n* **Worker Agents**: Propose concrete code refactors, dependency updates, and branch integrations (`qa_agent.py`, `build_agent.py`, `art_agent.py`).\n* **Critic Agent**: Detects logical flaws, anti-patterns, and **strategic dishonesty** (agents hiding bugs to pass CI).\n* **Opponent Agent**: Dialectical devil's advocate challenging every proposal with worst-case scenario testing.\n* **Artistic Integrity Agent**: Enforces visual and creative fidelity in creative and gaming software projects, ensuring AI optimizations do not strip human aesthetic soul.\n* **Budget Guard Agent**: Monitors API token consumption and token-per-watt efficiency to prevent runaway expenses.\n\n---\n\n## 💰 Economic Justification: CRR Economics\n\nA central tenet of the PWM framework is economic discipline. AI agents must never cost more than the human rework they prevent. PWM continuously computes the **Compute-to-Rework Ratio (CRR)**:\n\n$$\\text{CRR} = \\frac{\\text{Total AI Inference Cost (€)} + \\text{Infrastructure & GPU Cost (€)}}{\\text{Estimated Human Developer Rework Saved (€)}}$$\n\n* **$\\text{CRR} < 0.10$**: Exceptional efficiency (AI is 10x+ cheaper than manual firefighting).\n* **$\\text{CRR} \\le 0.30$**: Target operating budget envelope.\n* **$\\text{CRR} > 0.30$**: Layer 4 Retain Firewall veto triggered (`FAILED_CRR_BUDGET`).\n* **$\\text{CRR} \\ge 1.00$**: **Jevons Paradox Alert** — autonomous execution halted; requires explicit human executive override.\n\n---\n\n## ⚡ Model Serving & Hybrid Cloud Architecture\n\nPWM uses a cost-optimized **Hybrid Serverless + On-Demand GPU** topology:\n\n```\n┌────────────────────────────────────────┐       ┌───────────────────────────────────────┐\n│          Google Cloud Run              │       │       Google Compute Engine (GCE)     │\n│       (Serverless Orchestration)       │       │            (On-Demand GPU VM)         │\n│                                        │       │                                       │\n│  • Web Dashboard & 3D DTO Garden       │ HTTP  │  • GraphJEPA Server (:8001)           │\n│  • Layers 1, 3, 4, 5 Execution         ├──────►│  • JEPA World Model Server (:8000)    │\n│  • Vertex AI Gemini 3 Gateway          │       │  • PointJEPA 3D Server (:8002)        │\n│  • Scale-to-Zero (--min-instances=0)   │       │  • NVIDIA NemoClaw Sandbox (:8008)    │\n└────────────────────────────────────────┘       └──────────────────┬────────────────────┘\n                                                                    │\n                                                 Automated Auto-Shutdown on Idle\n```\n\n* **Cloud Run Scale-to-Zero**: Containerized orchestration scale-to-zero (`--min-instances=0`) when no active users are connected, avoiding idle hosting charges.\n* **On-Demand GPU Lifecycle**: `run_gpu_pipeline.py` starts the GPU VM via GCP API, waits for health checks, processes batched embeddings, and guarantees shutdown in a `finally` block.\n* **Semantic Vertex AI Fallbacks**: If the GPU VM is offline, PWM seamlessly falls back to Gemini Pro/Flash semantic simulation with zero downtime.\n\n---\n\n## 💻 Quick Start & Local Execution\n\n### 1. Prerequisites & Installation\n```bash\n# Clone the repository\ngit clone https://github.com/petripaananen/Project-World-Model.git\ncd Project-World-Model\n\n# Setup Python virtual environment\npython -m venv .venv\nsource .venv/Scripts/activate  # Windows: .venv\\Scripts\\activate\n\n# Install dependencies\npip install -r requirements.txt\n```\n\n### 2. Environment Configuration\nCreate a `.env` file in the project root:\n```env\nGOOGLE_API_KEY=your_gemini_api_key\nGCP_PROJECT_ID=your-gcp-project-id\nGCP_LOCATION=us-central1\nNEMOCLAW_ENDPOINT_URL=http://localhost:8008/audit\nJEPA_WM_ENDPOINT_URL=http://localhost:8000\n```\n\n### 3. Execution Commands\n\n* **Run Interactive 3D DTO Web Dashboard** (Recommended):\n  ```bash\n  python -m pwm.main --web --port 8765\n  ```\n  Open `http://localhost:8765` to view the live dashboard and Babylon.js 3D Garden.\n\n* **Run CLI Demo Mode** (Fully deterministic, no API keys required):\n  ```bash\n  python -m pwm.main --mode demo\n  ```\n\n* **Run Live Gemini 3 Agent Analysis**:\n  ```bash\n  python -m pwm.main --mode analyze\n  ```\n\n* **Run Continuous 24h Autonomous Loop**:\n  ```bash\n  python -m pwm.main --loop --interval 300\n  ```\n\n* **Simulate Specific Pre-Built Scenarios**:\n  * Scenario 1 (Causal Counterfactual Simulation): `python -m pwm.main --scenario 1`\n  * Scenario 2 (NemoClaw Sandbox Veto): `python -m pwm.main --scenario 2`\n  * Scenario 3 (CRR Budget Alert & Jevons Defense): `python -m pwm.main --scenario 3`\n\n* **Start the Standalone JEPA Model Serving Cluster**:\n  ```bash\n  python -m pwm.serving.runner --all\n  ```\n\n---\n\n## 🧪 Automated Test Suite\n\nPWM enforces a strict pre-deployment testing rule: all release commits must pass 100% of local unit and integration tests.\n\n```bash\npytest tests/\n```\n\n### Test Coverage (81 Passed / 0 Failures):\n* `test_rsi_integration.py`: Validates Layer 3 Predict-Before-Evaluate selection, Layer 4 Retain Gate veto conditions (dishonesty, sandbox, energy, CRR), and closed-loop trajectory recording.\n* `test_jepa_core.py`: Asserts GraphJEPA topological embeddings, LeJEPA SIGReg regularization, and OPF Gram matrix orthonormality.\n* `test_calibration.py`: Asserts self-supervised calibration factor updates, factorized grounding errors, and trajectory persistence.\n* `test_new_agents.py`: Validates Worker, Critic, Opponent, and specialized agent consensus workflows.\n* `test_security_adversarial.py`: Validates SAIF security posture, prompt injection resistance, and NemoClaw isolation.\n* `test_budget_and_loop.py`: Validates CRR calculations, token budget enforcement, and debate loops.\n* `test_agile_integration.py`: Tests deterministic Agile metrics (velocity, lead time, RACI, DoD compliance).\n* `test_model_serving.py`: Tests FastAPI JEPA microservice endpoints (`/simulate`, `/encode-graph`, `/audit-geometry`).\n* `test_gcp_config.py`: Verifies GCP configuration loading, GCE lifecycle management, and semantic fallbacks.\n* `test_jira_ingest.py` & `test_linear_ingest.py`: Tests real-world issue tracker adapters.\n* `test_ftue_parsers.py`: Verifies MS Project XML and Slack export parsers.\n\n---\n\n## 📚 Academic Citation & References\n\nIf you use or build upon the Project World Model framework in your academic research or enterprise engineering, please cite:\n\n```bibtex\n@mastersthesis{paananen2026pwm,\n  author       = {Petri Paananen},\n  title        = {Itseohjautuvat työnkulut videopeliteollisuudessa: tekoälyn maailmanmallit tuotannon johtamisen viitekehyksenä},\n  school       = {Jyväskylä University of Applied Sciences (JAMK)},\n  year         = {2026},\n  type         = {Master's Thesis},\n  url          = {https://urn.fi/URN:NBN:fi:amk-2026052919003}\n}\n```\n\n### Key Literature & Architecture References:\n1. **RSI-Anything**: Gen-Verse (2026). *RSI-Anything: Learning to Self-Improve World Models and Policies through Gated Retention*. arXiv:2610.13912.\n2. **JEPA-Anything**: Gen-Verse (2026). *JEPA-Anything: Learning Predictive Models across Different Worlds*. arXiv:2609.20800.\n3. **LeJEPA**: Balestriero, R., & LeCun, Y. (2025). *LeJEPA: Provable and Scalable Self-Supervised Learning Without Heuristics*. Meta AI Research.\n4. **Google SAIF**: *Secure AI Framework: Standards for Agentic Risk Mitigation*. Google Cloud Security.\n\n---\n\n## 📄 License\n\nThis project is licensed under the **Apache License, Version 2.0**. See the [LICENSE](LICENSE) file for details."}