Areon: Latency-Friendly and Resilient Multi-Proposer Consensus

TL;DR

Areon presents a DAG-based PoS framework that enables multiple proposers per slot and uses a sliding-window, CCALocal fork-choice with a new Tip-Boundedness invariant to bound frontier width. This combination, along with a rigorous DG/DQ/DCP set of properties, yields persistence and liveness with an explicit $(k,\varepsilon)$-finality bound under partial synchrony. The authors formalize both an idealized synchronous protocol (Areon-Ideal) and a practical bounded-delay instantiation (Areon-Base) with VRF-based eligibility, and they prove corresponding finality results, then validate performance via a discrete-event simulator against Ouroboros Praos. The work demonstrates bounded-latency finality and reduced reorg frequency at matched block-arrival rates, showing robustness to network delays and adversarial stake configurations. Practically, Areon offers a scalable path to high-throughput, low-latency finality in PoS by local, windowed, and weight-driven fork-choice in a concurrent DAG setting.

Abstract

We present Areon, a family of latency-friendly, stake-weighted, multi-proposer proof-of-stake consensus protocols. By allowing multiple proposers per slot and organizing blocks into a directed acyclic graph (DAG), Areon achieves robustness under partial synchrony. Blocks reference each other within a sliding window, forming maximal antichains that represent parallel ``votes'' on history. Conflicting subDAGs are resolved by a closest common ancestor (CCA)-local, window-filtered fork choice that compares the weight of each subDAG -- the number of recent short references -- and prefers the heavier one. Combined with a structural invariant we call Tip-Boundedness (TB), this yields a bounded-width frontier and allows honest work to aggregate quickly. We formalize an idealized protocol (Areon-Ideal) that abstracts away network delay and reference bounds, and a practical protocol (Areon-Base) that adds VRF-based eligibility, bounded short and long references, and application-level validity and conflict checks at the block level. On top of DAG analogues of the classical common-prefix, chain-growth, and chain-quality properties, we prove a backbone-style $(k,\varepsilon)$-finality theorem that calibrates confirmation depth as a function of the window length and target tail probability. We focus on consensus at the level of blocks; extending the framework to richer transaction selection, sampling, and redundancy policies is left to future work. Finally, we build a discrete-event simulator and compare Areon-Base against a chain-based baseline (Ouroboros Praos) under matched block-arrival rates. Across a wide range of adversarial stakes and network delays, Areon-Base achieves bounded-latency finality with consistently lower reorganization frequency and depth.

Figures (14)

• Figure 1: Legend of colors used in Figures \ref{['fig:dg']}--\ref{['fig:liveness']}. Black = blocks; green = honest or finalized prefix; blue = frontier region; gray = common past; red = broadcast block event or inclusion.
• Figure 2: DAG Growth (DG).Over any $\ell$-slot interval, with probability $1-\mathsf{negl}\xspace(\kappa\xspace)$ at least $\tau_D\ell$ honest blocks become ancestors of an honest party's fork-choice tip (i.e., lie on some honest preferred subDAG) by the end of the interval. Black dots represent honest blocks produced during the interval.
• Figure 3: DAG Quality (DQ).In any $\ell$-slot window, the fraction of honest blocks among those entering the advancing frontier is at least $\mu_D$ with probability $1-\mathsf{negl}\xspace(\kappa\xspace)$. In the illustration, the blue band marks the advancing frontier; black dots represent all blocks that enter this frontier, and green dots highlight the subset created by honest parties.
• Figure 4: DAG Common Past (DCP). The gray region shows the common past that all honest parties agree on. For two honest parties $P$ and $Q$ at times $t_1\le t_2$, let $\mathrm{Past}\xspace_P(t_1)$ and $\mathrm{Past}\xspace_Q(t_2)$ be the ancestor sets of their preferred frontiers (Def. \ref{['def:DCPprime']}). The blue bands depict the most recent $k_D$ layers adjacent to these frontiers. After removing these $k_D$ frontier layers, the earlier trimmed past is contained in the later one, i.e., $\mathsf{Trim}\xspace(\mathrm{Past}\xspace_P(t_1), k_D) \subseteq \mathsf{Trim}\xspace(\mathrm{Past}\xspace_Q(t_2), k_D)$, with probability $1-\mathsf{negl}\xspace(\kappa\xspace)$.
• Figure 5: Persistence (Safety).All but the last $k$ blocks in the ledger at time $t_1$ remain immutable and appear in every honest ledger at any later time $t_2\ge t_1$, except with probability $1-\mathsf{negl}\xspace(\kappa\xspace)$. The green region shows the $k$-deep finalized prefix at $t_1$, which remains unchanged in later ledgers.

Theorems & Definitions (84)

• Definition 2.1: Window and Short-Reference Window
• Remark 2.2: Default meaning of tips
• Definition 2.3: Closest Common Ancestor (CCA)
• Definition 2.4: Conflict Predicate $\mathrm{Conflicts}$
• Definition 2.5: Conflicted-tips resolution
• Definition 2.6: Fork-choice preference and preferred frontier
• Remark 2.7: Notation: structure vs. preference
• Remark 2.8: Protocol vs. analysis layer
• Definition 2.9: DAG Growth
• Definition 2.10: DAG Quality