Minimmit

• Name: Minimmit
• URL: https://arxiv.org/abs/2508.10862
• Category: two-round-finality BFT consensus design / low-latency view-progression primitive / Commonware research protocol
• Summary: Minimmit is worth cataloging not just as a Commonware blog post or a footnote in the new two-phase-consensus wave, but as a distinct BFT design that makes one unusually clean tradeoff explicit: it separates the quorum needed to move the network forward from the quorum needed to finalize a block. In the reviewed spec and paper, replicas advance views after a 2f + 1 / ~40% proof and finalize only after an n - f / ~80% vote threshold, all under the stricter n >= 5f + 1 fault model. That makes Minimmit a useful comparison point for HotStuff-2, HotShot, Kudzu, Alpenglow, and ChonkyBFT because the real mechanism is not just fast finality, but a lower-latency view-progression rule that intentionally gives up the usual one-third Byzantine tolerance to cut block-time overhead.
• What it does:
  • Defines a responsive, leader-based partially synchronous SMR protocol that aims for two-round finality: one proposal round plus one vote round
  • Separates quorum thresholds so the network can move to the next view on a smaller proof while requiring a larger quorum to actually finalize a block
  • Uses notarization and nullification certificates as the core control surfaces for progress, parent selection, and safe skipping of prior views
  • Lets leaders build on the most recent notarized block that is connected to the current view by intervening nullification proofs, rather than forcing direct reproposal of every prior candidate block
  • Adds an explicit nullify by contradiction path so replicas that already voted for one block can still guarantee a proof exists for the view once they know that block can no longer finalize
  • Surfaces practical extension hooks such as threshold signatures, erasure-coded block dissemination, fast-skipping of offline leaders, rebroadcast-on-stall behavior, and equivocation slashing
• Key claims:
  • Minimmit clears the corpus bar because it exposes a lower consensus design choice that later performance-stack discussions often flatten together: whether fast chains reduce latency mainly through better networking, through smaller progress quorums, or through slower-path fallbacks around a fast path.
  • The formal paper’s main claim is that view progression and transaction finality can safely use different quorum thresholds. In the reviewed materials, that becomes the concrete split between M = 2f + 1 for progress proofs and L = n - f for finalization.
  • The arXiv abstract claims experiments on a globally distributed 50-processor network achieved a 23.1% reduction in view latency and a 10.7% reduction in transaction latency relative to the compared state of the art. That performance claim is important, but the reusable insight is the threshold split itself.
  • The Commonware blog is especially useful because it states the intended deployment tradeoff plainly: Minimmit optimizes for view latency instead of the stronger fault assumptions or fallback robustness emphasized by other recent two-phase designs.
  • The protocol is analytically valuable precisely because it does not hide its tradeoff. Safety is framed around the stricter < 20% Byzantine tolerance regime (n >= 5f + 1), so Minimmit should be compared against HotStuff-2 and newer fast-finality systems as a deliberately narrower low-latency point in the design space rather than as a generic better BFT claim.
  • The strongest caveat from this pass is maturity. The Commonware post explicitly says Minimmit was not yet fully implemented when released, and the formal paper plus repo spec are better treated as mechanism artifacts than as evidence of production deployment.
• Whitepaper: Yes. The canonical primary artifact is the arXiv paper Minimmit: Fast Finality with Even Faster Blocks, supported by Commonware’s explanatory blog post and the public repository specification; see ../whitepapers/minimmit-primary-sources-2026-05-14.md.
• Sources:
  • https://arxiv.org/abs/2508.10862
  • https://arxiv.org/pdf/2508.10862.pdf
  • https://commonware.xyz/blogs/minimmit
  • https://github.com/commonwarexyz/monorepo/tree/main/pipeline/minimmit
  • https://raw.githubusercontent.com/commonwarexyz/monorepo/main/pipeline/minimmit/minimmit.md

Internal linkages

• Best reads from here: hotstuff-2, alpenglow, and chonkybft.

• Minimmit is the clean quorum-split specimen, not a note that needs to wander into every low-latency ordering or middleware cousin.

• Last reviewed: 2026-05-14 UTC