Skip to content

super_sprint_plan.md

Latest commit

 

History

History
51 lines (37 loc) · 3.98 KB

super_sprint_plan.md

File metadata and controls

51 lines (37 loc) · 3.98 KB

Strategify Super Sprint Plan: Completed "Cognitive Core" & Next Steps

Status: "Cognitive Core" Sprint (Phases 15-18) Completed.

✅ COMPLETED: The "Cognitive Core" Sprint

Phase 15: LLM Orchestration & Symbolic Grounding

  • Epistemic Prompt Injection: strategify/reasoning/llm.py now queries StrategicBridge for actual beliefs and knowledge before hitting the LLM.
  • Clojure Counterfactual Forecaster: Integrated ClojureBridge.branch_timelines to provide the LLM with simulated Lisp futures.
  • The CognitiveActorAgent: Created CognitiveActorAgent substituting Nash matrix arrays with highly contextual LLM calls.

Phase 16: Prolog Deep Epistemology & Deception

  • Disinformation Mechanics: Added believes/2 overrides in traits.pl for propaganda spread based on gullible vs analyst traits.
  • Propagating Facts: Integrated StrategicBridge inside GeopolModel.step() to log physical state changes (combat, posture) as Prolog beliefs (action(region, posture)).
  • Information Warfare: Added disinformation_game to crisis_games.py (propaganda vs censorship).

Phase 17: Clojure Monte Carlo Tree Search (MCTS)

  • Expand Clojure State Map: Expanded GameStateRecord to include :economic-reserve, :stability, and :un-resolutions.
  • Fast Path Combat Resolution: Delegated kinetic combat calculations in conflict.py to (resolve-combat state) in Clojure.

Phase 18: The Grand Strategy Tournament

  • Tournament Runner: Established tournament_runner.py comparing traditional Nash agents against CognitiveActorAgent across metrics like global tension and escalations.

🚀 NEW SPRINT: UI & Interactive MLOps Integration

Objective: Move the simulation out of headless mode by properly exposing the newly created Cognitive architectures to the React frontend. We need to allow humans to see the "thoughts" of the agents, interact with them, and expand our economic fidelity.

Phase 19: Full-Stack API Integration

Goal: Tie the Mesa simulation backend to the React/Vite frontend.

  • Task 1: FastAPI Controller Layer: Add a fastapi app running alongside or wrapping the Mesa model. Expose endpoints for POST /api/simulation/start, GET /api/simulation/state, and POST /api/simulation/step.
  • Task 2: Frontend Data Hookup: Populate frontend/src/api/ and frontend/src/hooks/ with Axios fetchers. Replace the hardcoded sampleRiskData in Dashboard.jsx and Simulation.jsx with live simulation data.

Phase 20: Explainable AI (XAI) UI layer

Goal: Visualize the "brain" of the CognitiveActorAgent.

  • Task 1: Epistemic Graph View: Create a PyVis/React component showing a node-graph of what an agent "knows" vs "believes" dynamically queried from Prolog.
  • Task 2: Lisp Branching UI: On the Simulation view, add a "Timeline Prediction" tree showing the Clojure MCTS branch projections that the LLM is currently evaluating.
  • Task 3: Live LLM Logs: Add an inspector panel for agents revealing the raw prompt and response (XAI transparency for the user).

Phase 21: RL Sandbox & Human-in-the-Loop

Goal: Allow users to step into the role of a faction and play against the Cognitive AI or RL Agents.

  • Task 1: Interactive Play Mode: Modify the PettingZoo wrapper so that one agent can be "human".
  • Task 2: Frontend Command Terminal: Add a UI module allowing the user to submit actions (Escalate, SpreadFakeNews, etc.) for their mapped nation during a live turn.

Phase 22: Deep Supply Chain Economics

Goal: Increase the fidelity of the "economic" variable to multi-commodity supply chains.

  • Task 1: Commodity Ledger: Update the economic models to track Oil, Semiconductors, and Food across the network graph.
  • Task 2: Vulnerability Analysis: Implement tools to detect critical supply chain choke points using NetworkX betweenness centrality and feed these into the Prolog epistemology potential_gain/risk_level facts.