Ecosystem coverage¶
Agentic Arena connects concepts, implementation patterns, and evidence. Choose a question before a product. Coverage here describes this repository, not the maturity of a domain.
| Domain | Decision | Starting point | Coverage |
|---|---|---|---|
| Models/gateways | Can my transport express the task? | Choose | Shared mock/native API and subscription functional paths |
| Loops/frameworks | Who owns control and state? | Learn | Adapters, architecture, and educational build path |
| Tools/protocols | Does a boundary preserve meaning? | Connect | Local tools; scoped protocol lessons |
| Context/memory | What should be retained and retrieved? | Build | Fixed RAG and context-growth evidence; scoped lessons |
| Reliability | What happens when work fails? | Debug | Retry, approval and restart findings |
| Execution/security | What authority and access does work have? | Secure | Approval evidence; controlled boundary fixtures |
| Evaluation/research | What would establish the claim? | Evaluate | Mechanical scorers, comparison methodology, research profiles |
| Observability | Can the evidence explain an outcome? | Evaluate | Run records and educational independent-oracle examples |
| Operations | Can I bound, recover, and clean up work? | Deploy | Local recipes; no production certification |
| Coding/interaction | How does a complete harness perform? | Coding harnesses | DeepSeek Harness source review; executable and general interaction benchmarks remain open |
| Optimization/training | Can improvement avoid evaluation leakage? | Research backlog | Research gap |
| Skills/supply chain | Can extensions be trusted and updated? | Research backlog | Research gap |
External projects and review depths live in the repository catalog. Our planned production harness is separate and receives the same evidence standards as other implementations.