Competitive Position
vs Centralized Assistants
| Aspect | ChatGPT / Claude / Gemini | Eirel |
|---|---|---|
| Control | Single vendor | Decentralized, emission-driven |
| Competition | None — use what the vendor ships | Continuous, multiple competing agents |
| Deep capabilities | One vendor's implementation | Competition-winning specialists per family |
| Verification | Self-evaluated | Owner-verified, hidden tests, ensemble judges |
| Extensibility | Vendor-controlled plugins | Open MCP ecosystem (12,000+ tool servers) |
| Transparency | Closed models, opaque eval | Open SDK, published scoring methodology |
vs Other Bittensor Subnets
| Aspect | Typical LLM Subnets | Eirel |
|---|---|---|
| Evaluation | Single-dimension scoring | Three-pillar, multi-dimensional, owner-verified |
| Gaming resistance | Limited | 12+ detectors + claim verification |
| Miner trust | Miner-hosted endpoints | Owner-hosted containers from submitted archives |
| Task complexity | Single-turn chat | Streaming conversation + autonomous multi-day projects |
| User experience | API-only | Streaming chat interface with conversation memory |
Honest Trade-Offs
- Owner centralization at launch — mitigated by decentralization roadmap (validator-run evaluation by Q3 2026)
- Higher evaluation complexity — full scoring breakdown published per epoch
- Unproven at scale — phased rollout, calibration gates require cross-epoch consistency
- LLM judge dependency — ensemble dual-judge with disagreement detection + deterministic fallback