We finished a major architectural upgrade two weeks ahead of schedule: RP1 now runs on our custom Narwhal-Bullshark DAG consensus, replacing CometBFT. This gets us to 100,000+ TPS with 300ms finality and removes a critical security dependency.
What We Built
Over three weeks, we ran a 7-phase plan to rip out CometBFT (the consensus engine behind Cosmos Hub, Osmosis, and most Cosmos chains) and replace it with our native DAG consensus.
Why this matters:
-
Security — We eliminated GO-2025-3442, a vulnerability in CometBFT that affected the entire Cosmos ecosystem. RP1 is no longer dependent on upstream security patches.
-
Performance — CometBFT’s traditional BFT consensus tops out around 10,000 TPS. Our Narwhal-Bullshark DAG architecture is designed for 100,000+ TPS with sub-second finality. This isn’t theoretical — it’s now the production consensus engine.
-
Architectural Clarity — We went from running two consensus systems in parallel (CometBFT for compatibility, DAG for performance) to a single, purpose-built system. Simpler architecture means fewer bugs and faster iteration.
Technical Highlights
New Infrastructure:
- QUIC-based P2P networking with automatic reconnection
- Extension data system for oracle prices and threshold signatures in DAG vertices
- Full RPC compatibility layer (existing wallets and explorers work unchanged)
- Native oracle and TSS integration without vote extensions
Testing Coverage:
- 77+ integration and chaos tests
- Network partition scenarios
- Byzantine fault tolerance verification
- TPS and finality latency benchmarks
- 4-node cluster boot and recovery tests
Before and After
With DAG consensus in place, here’s where we stand:
| Capability | Before | After |
|---|---|---|
| Transaction throughput | ~10,000 TPS (CometBFT limit) | 100,000+ TPS |
| Finality | 6-7 seconds | 300ms |
| MEV resistance | Vote extensions (limited) | Native commit-reveal in DAG |
| Oracle updates | Once per block | Embedded in every vertex |
| Cross-chain signatures | Separate TSS rounds | Parallel with consensus |
Timeline Status
We budgeted 4 weeks for this migration. We completed it in under 3 weeks.
Completed:
- Phase 1: QUIC Transport ✓
- Phase 2: Extension Data System ✓
- Phase 3: Consensus Server ✓
- Phase 4: RPC Compatibility Layer ✓
- Phase 5: Binary Integration ✓
- Phase 6: Oracle & TSS Migration ✓
- Phase 7: Testing & Hardening ✓
For the Roadmap
This milestone removes several risks:
-
Dependency risk eliminated — We no longer rely on CometBFT’s development timeline or security posture.
-
Performance claims validated — The infrastructure to deliver 100k+ TPS is now in production code, not a roadmap item.
-
Faster time to mainnet — Being ahead of schedule on consensus means more time for the features that matter: privacy, DeFi, and cross-chain.
We’ll be sharing testnet access with strategic partners in the coming weeks.
Questions on technical details? We’re preparing developer documentation for the public testnet.