features_overview.md

Features Overview

This page is the “map” of the major systems in FBA-Bench and where to find the implementation.

Core Benchmark Modes

• Prompt Battery (prompt): fast, model-only evaluation (no tools/memory implied).
• Agentic Simulation (agentic): long-horizon, stateful simulation for full agent systems.

See: docs/benchmark_philosophy.md

Simulation Fidelity

• World state + arbitration: docs/simulation_services.md (WorldStore)
• Market demand + sales processing: docs/simulation_services.md (MarketSimulationService)
• Supply chain disruptions + black swans: docs/simulation_services.md (SupplyChainService)
• Fees and unit economics: docs/simulation_services.md (FeeCalculationService)
• Trust/reputation signals: docs/simulation_services.md (TrustScoreService)

Security & Adversarial Testing

• Red Team Gauntlet (adversarial injections + scoring): docs/red_team_gauntlet.md

Memory & Agent Modes

• Per-day long-term memory consolidation + competition awareness modes: docs/cognitive_memory.md

Consumers & Demand Modeling

• Utility-based autonomous shoppers: docs/consumer_utility_model.md

Constraints & Fairness

• Token/cost/call budgets and enforcement: docs/budget_constraints.md

Reproducibility

• Deterministic seeding + response caching + golden masters: docs/reproducibility.md
• Audit and replay (artifacts, playback paths, and audit layers): docs/audit_and_replay.md

Extensibility

• Plugin framework (scenarios/agents/tools/metrics): docs/plugin_framework.md

Scoring

• Multi-axis metrics suite (finance/ops/trust/stress/adversarial/cost): docs/metrics_suite.md

Observability

• Logs, metrics, tracing, ClearML hooks: docs/observability_stack.md

Learning / Research

• Learning modules and post-run analysis tooling: docs/learning_systems.md

Visualization / Observer Mode

• Godot Simulation Theater (observer-friendly GUI) and recording workflows:
  • docs/RUNBOOK_SIM_THEATER.md
  • docs/press/promo_video.md
  • scripts/record_godot_demo.ps1