research: DAG-based consensus exploration (Narwhal / Bullshark / Mysticeti)

[satyakwok]

Investigate whether DAG-based consensus (as in Aptos / Sui / Mysten Labs work) is a fit for Sentrix's next-major-version consensus rewrite.

Motivation

Current Voyager BFT (Tendermint-style 3-phase) has a hard floor of ~3 × RTT per block. With Sentrix's geographic-distributed validators (~65ms avg RTT vps3↔vps6), this caps mainnet bt at ~200ms theoretical + state-apply + BLOCK_TIME_SECS floor. DAG consensus separates data dissemination from total order, achieving sub-second finality even with global validator sets.

Key references

• Narwhal + Tusk: https://arxiv.org/abs/2105.11827
• Bullshark: https://arxiv.org/abs/2201.05677
• Mysticeti: https://arxiv.org/abs/2310.14821

Scope

• Read the papers + understand the trade-offs vs Tendermint
• Map to Sentrix's slashing + state-apply model (DAG consensus is consensus-only; state still applies linearly)
• Prototype against existing libp2p + revm infrastructure
• 6-12 month workstream if green-lit

Out of scope

• Same-session decision; this is a research issue, not a planning issue
• Replacing Voyager in v3 — that's a separate RFC if/when this research lands

Acceptance

• Written summary of pros/cons for Sentrix specifically
• Prototype repo (separate from sentrix-labs/sentrix)
• Go/no-go recommendation