Geospatially Aware Proof of Stake

• GPoS is a blockchain consensus mechanism that weights validator power by combining normalized stake with geographical diversity.
• It rebalances voting power using a geospatial diversity index based on validators’ relative distances and stake distribution.
• Empirical evaluations show GPoS can reduce geospatial centralization by up to 45% while maintaining throughput and security thresholds.

Geospatially aware Proof of Stake (GPoS) is a modification of Proof of Stake (PoS) in which validator voting power depends not only on stake but also on geospatial diversity. In its canonical blockchain formulation, GPoS addresses the observation that PoS systems may be decentralized economically while remaining geographically concentrated, so that a small number of regions dominate consensus influence. The design introduced in "GPoS: Geospatially-aware Proof of Stake" rebalances voting weights at epoch boundaries using normalized stake and a geospatial diversity signal, while retaining the underlying validator set and BFT-style consensus structure (Motepalli et al., 3 Nov 2025). Related work on geospatial concentration and minority-validator disadvantage provides the empirical and methodological background, and a separate PoS-based Wide Area Visual Navigation framework in robotics shows that PoS ideas can also be adapted to spatially grounded coordination outside financial blockchains (Motepalli et al., 2023, Paykari et al., 21 May 2025).

1. Problem domain and conceptual motivation

GPoS is motivated by the claim that PoS blockchains are often not geospatially decentralized even when stake ownership is distributed. In the empirical study underlying the protocol proposal, five major PoS systems—Aptos, Avalanche, Ethereum, Solana, and Sui—were analyzed, and the reported result was that the top three countries account for over 33% of total stake across all blockchains. The highlighted concentration regions include the U.S., Germany, Finland, the Netherlands, Ireland, Singapore, Japan, and France. Country-level Gini coefficients were all above 0.5, with reported values of about 0.88 for Ethereum, about 0.71 for Avalanche, about 0.77 for Solana, about 0.57 for Sui, and about 0.51 for Aptos. On the paper’s geospatially weighted influence metric, the Gini coefficient of eigenvector centrality ranged from roughly 0.527 to 0.941, with Ethereum around 0.941 and Avalanche around 0.804 (Motepalli et al., 3 Nov 2025).

The protocol-level concern is not limited to descriptive geography. The literature frames geographic concentration as a source of correlated-failure risk, regulatory vulnerability, and unequal participation conditions. If influential validators are concentrated in the same region, the same outage, disaster, network partition, cloud-provider policy, sanctions regime, or legal intervention can affect a large fraction of consensus influence. Geographic proximity can also create latency advantages that improve the ability to capture MEV, front-run transactions, or win proposer selection, thereby producing uneven access to rewards and influence. A key argument in the GPoS literature is therefore that conventional decentralization metrics such as the Nakamoto coefficient, entropy, or vanilla Gini can miss a distinct operational dimension of decentralization: the physical distribution of validators and transaction processing (Motepalli et al., 3 Nov 2025, Motepalli et al., 2023).

Earlier work sharpened this point by showing that minority validators—validators located in geospatially distant regions relative to the dominant cluster—can be disadvantaged in standard BFT-style PoS operation. That work reported, for example, that 53.11% of Ethereum validators were in the US and 18.29% in Germany, for a combined 69.91% in two regions. In 0L, the same paper reported that at epoch 317, out of 39 validators, 58.9% were in the US and 28.2% were in Finland, and that by epoch 348 the set had grown to 43 validators with 51.1% in the US and 30.2% in Finland. The paper interpreted the persistence of such patterns as evidence of a structural centralization pressure rather than an accidental arrangement (Motepalli et al., 2023).

2. Geospatial metrics and formalization

A central contribution of the GPoS line of work is the formalization of geospatial diversity as a validator-level and system-level quantity. In the earlier geospatial-distribution study, the Geospatial Diversity Index (GDI) of a validator was first defined as the sum of Haversine distances from that validator to all others in the validator set, and then refined to a quorum-aware form that sums distances only to the closest two-thirds of validators needed for quorum. The blockchain-level GDI was defined as the average of validator-level GDI values over the entire validator set. That formulation directly connected a validator’s physical remoteness from quorum to its expected liveliness in HotStuff-like operation (Motepalli et al., 2023).

The later GPoS protocol paper redefined GDI in explicitly stake-aware terms. Its validator-level Geospatial Diversity Index is the minimum distance from validator $v_k$ to the closest set of validators whose combined normalized stake reaches quorum:

$$\text{GDI}_k = \min_{V_C \subseteq \mathcal{V} \setminus \{v_k\}} \left\{ \sum_{v_j \in V_C} \Delta_{k, j} \mid \sum_{v_j \in V_C} s_j + s_k \geq \frac{2}{3} \right\}.$$

Under this definition, a validator has a higher GDI if the nearest quorum-supporting validators are farther away, and a lower GDI if enough voting power is geographically close. Quorum remains the usual two-thirds threshold. This preserves the BFT quorum intuition while embedding it in a spatial model of stake distribution (Motepalli et al., 3 Nov 2025).

For system-wide evaluation, the GPoS paper uses the Gini coefficient of eigenvector centrality (GEC) as its main geospatial decentralization metric. The method constructs a stake-proximity weighted graph, computes eigenvector centrality by solving $A \cdot x = \lambda x$, and then applies the Gini coefficient to the resulting centrality scores. The appendix argues that eigenvector centrality was the only tested centrality measure that simultaneously captures stake sensitivity, spatial awareness, and recursive influence relevant to quorum formation. This makes GEC a metric of geospatially weighted consensus influence rather than a simple count of validator locations (Motepalli et al., 3 Nov 2025).

The same literature also distinguishes majority and minority geospatial roles. In the earlier study, the minority set $\mu$ was operationally defined as validators above the 67th percentile of GDI, so that minority size is at most one-third of the set, and a majority location was defined as a location containing at least two-thirds of validators. This vocabulary is important because it recasts distance from quorum as a structural property of the validator set rather than an incidental deployment detail (Motepalli et al., 2023).

3. Voting-power reweighting in GPoS

In ordinary PoS, consensus voting power is determined by normalized stake. GPoS replaces stake-only voting power with a linear combination of normalized stake and normalized GDI:

$$\omega_i = f(s_i, GDI_i) = \lambda \cdot s_i + (1-\lambda) \cdot GDI'_i,$$

with

$GDI'_i = \frac{GDI_i}{\max(GDI)}.$

Final voting power is then normalized:

$$\rho_i^* = \frac{\omega_i}{\sum_{v_j \in \mathcal{V}} \omega_j}.$$

The parameter $\lambda \in [0,1)$ controls the trade-off between conventional stake weighting and geospatial diversity; $\lambda = 1$ gives ordinary PoS, and lower $\lambda$ gives more weight to geospatial diversity (Motepalli et al., 3 Nov 2025).

The paper recommends $$\text{GDI}_k = \min_{V_C \subseteq \mathcal{V} \setminus \{v_k\}} \left\{ \sum_{v_j \in V_C} \Delta_{k, j} \mid \sum_{v_j \in V_C} s_j + s_k \geq \frac{2}{3} \right\}.$$0 so that stake remains the dominant factor. The stated reason is adversarial: if geography were weighted too heavily, an adversary might game the system by spinning up geographically dispersed Sybil validators with little stake. This makes GPoS a constrained reweighting of stake-based influence rather than a geography-first design. A plausible implication is that the protocol is intended to preserve the security logic of stake while dampening regional concentration in the assignment of voting power (Motepalli et al., 3 Nov 2025).

Protocol integration is deliberately conservative. At epoch boundaries, the system reads stake and coordinates, determines eligible validators deterministically, computes each validator’s GDI, computes $$\text{GDI}_k = \min_{V_C \subseteq \mathcal{V} \setminus \{v_k\}} \left\{ \sum_{v_j \in V_C} \Delta_{k, j} \mid \sum_{v_j \in V_C} s_j + s_k \geq \frac{2}{3} \right\}.$$1, normalizes to $$\text{GDI}_k = \min_{V_C \subseteq \mathcal{V} \setminus \{v_k\}} \left\{ \sum_{v_j \in V_C} \Delta_{k, j} \mid \sum_{v_j \in V_C} s_j + s_k \geq \frac{2}{3} \right\}.$$2, and commits the updated validator set and voting powers for the next epoch. Validator identity is therefore not fundamentally changed; voting weights are rebalanced. The paper further states that GPoS is composable with Proof-of-Location systems but is not itself a PoL protocol. This places GPoS in the category of consensus-layer reweighting schemes rather than validator-discovery or location-attestation systems (Motepalli et al., 3 Nov 2025).

4. Empirical findings and performance evaluation

The empirical analysis in the GPoS paper used validator stake and location data from Aptos, Avalanche, Ethereum, Solana, and Sui, with “Ethereum nodes” considered as a separate dataset in parts of the analysis because of a different collection methodology. Public explorer APIs, web scraping, and IP-based geolocation were used; validators with missing locations were excluded; validators within a 20 km radius were merged to reduce local noise from neighboring data centers; and stakes were aggregated when nearby validators were merged. The paper also reports country-level concentration patterns in which Ethereum was dominated by the United States, Netherlands, Finland, and Germany; Solana by the United States, Germany, Netherlands, Japan, and UK; Avalanche by the United States, Germany, Ireland, Japan, and Singapore; Aptos by Germany, United States, Singapore, Ireland, and Netherlands; and Sui by United States, Germany, UK, Lithuania, and Netherlands (Motepalli et al., 3 Nov 2025).

The principal experimental claim is that transitioning from ordinary PoS with $$\text{GDI}_k = \min_{V_C \subseteq \mathcal{V} \setminus \{v_k\}} \left\{ \sum_{v_j \in V_C} \Delta_{k, j} \mid \sum_{v_j \in V_C} s_j + s_k \geq \frac{2}{3} \right\}.$$3 to GPoS with $$\text{GDI}_k = \min_{V_C \subseteq \mathcal{V} \setminus \{v_k\}} \left\{ \sum_{v_j \in V_C} \Delta_{k, j} \mid \sum_{v_j \in V_C} s_j + s_k \geq \frac{2}{3} \right\}.$$4 produced an average 45% reduction in the Gini coefficient of eigenvector centrality, with individual reductions ranging from about 41.38% to 49.72%. Because lower Gini indicates less inequality, the paper interprets this as a substantial improvement in geospatial decentralization. Sensitivity analysis varied $$\text{GDI}_k = \min_{V_C \subseteq \mathcal{V} \setminus \{v_k\}} \left\{ \sum_{v_j \in V_C} \Delta_{k, j} \mid \sum_{v_j \in V_C} s_j + s_k \geq \frac{2}{3} \right\}.$$5 from 0.5 to 0.9 in steps of 0.1 and showed the expected trend: lower $$\text{GDI}_k = \min_{V_C \subseteq \mathcal{V} \setminus \{v_k\}} \left\{ \sum_{v_j \in V_C} \Delta_{k, j} \mid \sum_{v_j \in V_C} s_j + s_k \geq \frac{2}{3} \right\}.$$6 produced more geospatial diversification, whereas higher $$\text{GDI}_k = \min_{V_C \subseteq \mathcal{V} \setminus \{v_k\}} \left\{ \sum_{v_j \in V_C} \Delta_{k, j} \mid \sum_{v_j \in V_C} s_j + s_k \geq \frac{2}{3} \right\}.$$7 moved the system closer to standard PoS. The paper also evaluated an exponential alternative,

$$\text{GDI}_k = \min_{V_C \subseteq \mathcal{V} \setminus \{v_k\}} \left\{ \sum_{v_j \in V_C} \Delta_{k, j} \mid \sum_{v_j \in V_C} s_j + s_k \geq \frac{2}{3} \right\}.$$8

which yielded an average 30% reduction in Gini when $$\text{GDI}_k = \min_{V_C \subseteq \mathcal{V} \setminus \{v_k\}} \left\{ \sum_{v_j \in V_C} \Delta_{k, j} \mid \sum_{v_j \in V_C} s_j + s_k \geq \frac{2}{3} \right\}.$$9 moved from 1 to 0.5 (Motepalli et al., 3 Nov 2025).

Security-threshold analysis examined how much stake an attacker would need to violate BFT thresholds of 33% for liveness and 66% for safety. In the reported experiments, under GPoS the minimum adversarial stake required to reach these thresholds was equal or greater than in standard PoS, assuming the attacker optimally distributes stake geographically. The paper therefore states that GPoS does not obviously weaken threshold security in the tested setting (Motepalli et al., 3 Nov 2025).

Performance was evaluated by emulating HotStuff and CometBFT on 64 VMs using NetEm, wide-area latency based on real-world server locations, a fixed message size of 128 bytes, and multiple runs per configuration. HotStuff throughput stayed at about 160,000 TPS across configurations, which the paper interprets as essentially no throughput degradation from GPoS in that setup. HotStuff latency also remained mostly stable across configurations. CometBFT was more sensitive to topology and gossip communication patterns, and its latency varied more significantly, especially for Aptos and Ethereum-node-like distributions. The reported conclusion is that GPoS can improve geospatial decentralization with minimal overhead, while the magnitude of performance impact depends on the BFT protocol (Motepalli et al., 3 Nov 2025).

5. Precursor results, minority-validator protection, and unresolved issues

The principal precursor to GPoS showed that geography affects validator participation in classical BFT-style PoS systems such as HotStuff, Tendermint, and systems built on them, including Ethereum 2.0-style PoS, Cosmos, Diem/Libra, Aptos, and 0L. That work formalized validator liveliness in epoch $A \cdot x = \lambda x$0 as

$A \cdot x = \lambda x$1

with validators falling below threshold $A \cdot x = \lambda x$2 treated as crash-liveness failures. In 0L, the threshold was reported as $A \cdot x = \lambda x$3. The study’s core claim was that minority validators often have very low liveliness, sometimes zero, despite being healthy and active, because geographically clustered majorities can form quorum without them or can time out when a distant minority validator is proposer (Motepalli et al., 2023).

This claim was supported by emulation through GeoDec, a HotStuff-based blockchain consensus emulator that takes a geographic validator distribution and injects network latencies using real-world city-to-city ping data. In an experiment with 16 validators, majority clusters in San Jose or Helsinki, and minorities in locations such as Singapore, Melbourne, Bangalore, Hong Kong, Montreal, Munich, Paris, Tokyo, and Vancouver, the paper reported 167 runs of 5 epochs each. Minority validators often had mean liveliness $A \cdot x = \lambda x$4, and box-plot medians for 1, 2, and 3 minority validators were reported as 0, 0.46, and 4.61 respectively, all below the threshold $A \cdot x = \lambda x$5. The reported Pearson correlation between quorum-aware GDI and liveliness was $A \cdot x = \lambda x$6 with p-value $A \cdot x = \lambda x$7, which the authors interpreted as strong evidence that distance from quorum depresses participation (Motepalli et al., 2023).

The mitigation proposed in that precursor paper was not a full geospatially aware consensus redesign, but an adjunct mechanism: GeoDecLite and the GDIExceptionContract. The contract provides nominate(), vouch(), and getMinorities(), allowing geographically remote validators to self-nominate, receive endorsements, and be exempted from jailing or slashing if they are validated as minority validators. The paper reports that this solution can safeguard all minority validators from being jailed. This is reactive rather than proactive: it protects distant validators after identification but does not itself rebalance voting power or quorum incentives (Motepalli et al., 2023).

The GPoS paper leaves several issues unresolved. It explicitly notes that formal safety and liveness proofs under the modified weighting are left for future work; location attestation must be reasonably accurate; validators can dispute false coordinates via a dispute process; and location spoofing is a real concern. Geographic data and IP-based geolocation can be imperfect. These caveats matter because GPoS depends on reliable coordinates but is not itself a location-proof system. The paper’s recommendation that $A \cdot x = \lambda x$8 further indicates an unresolved tension between rewarding geospatial diversity and preventing low-stake, geographically dispersed Sybil strategies (Motepalli et al., 3 Nov 2025).

6. Navigation-oriented and geospatially grounded PoS adaptations in robotics

A distinct but related use of PoS concepts appears in "Integrating Robotic Navigation with Blockchain: A Novel PoS-Based Approach for Heterogeneous Robotic Teams," which presents a work-in-progress PoS-inspired coordination framework for heterogeneous robot teams in Wide Area Visual Navigation (WAVN). The setting is large-scale, GPS-denied environments in unstructured domains such as agriculture and forestry, where visual cues replace GPS, terrain changes, and robots differ in sensors, capabilities, and roles. In this framework, stake is interpreted as a robot’s navigation reliability rather than financial stake, and PoS is used to prioritize robot data and select the best contributors for cooperative navigation (Paykari et al., 21 May 2025).

The framework is geospatially grounded through shared landmarks, robot positions, and the quality of landmark matches. Its stake weight function is

$A \cdot x = \lambda x$9

where $\mu$0 is robot $\mu$1’s stake. Its consensus score with PoS is

$\mu$2

with

$\mu$3

Its navigability function is

$\mu$4

Here $\mu$5 is the weight of landmark $\mu$6, and $\mu$7 is the relative importance or trustworthiness of robot $\mu$8 to robot $\mu$9, derived from past transactions between robots. The result is stake-weighted visual agreement over spatial features rather than stake-weighted voting over blocks (Paykari et al., 21 May 2025).

The paper reports that the consensus score over all robot pairs and landmarks has complexity $$\omega_i = f(s_i, GDI_i) = \lambda \cdot s_i + (1-\lambda) \cdot GDI'_i,$$0, and that the navigability function for one robot is $$\omega_i = f(s_i, GDI_i) = \lambda \cdot s_i + (1-\lambda) \cdot GDI'_i,$$1, approximated as $$\omega_i = f(s_i, GDI_i) = \lambda \cdot s_i + (1-\lambda) \cdot GDI'_i,$$2. Stake management remains unresolved: the paper states that stake represents navigation reliability inferred from navigation performance, landmark recognition quality, and history of cooperative behavior, but it does not provide a formal update rule. It also caps the analysis to the top 10 transactions when assigning inter-robot importance levels (Paykari et al., 21 May 2025).

Preliminary results used 10 robots and 20 landmarks in a $$\omega_i = f(s_i, GDI_i) = \lambda \cdot s_i + (1-\lambda) \cdot GDI'_i,$$3 world over 10 loops with randomly changing robot positions. The outcome included 45 blocks and 458 transactions. The maximum number of common landmarks shared between two robots was 5, and the minimum was zero. Some robots generated more blocks than others, ranging from 9 to 1. The reported “valley” in navigability when landmark quality drops, followed by recovery when quality is restored, indicates that navigability is sensitive to environmental context and perception quality. The paper does not use the term GPoS or define geospatial awareness formally, so identifying it as a form of “Geospatially Aware Proof of Stake” requires caution. A precise characterization is that it is a PoS-based WAVN framework in which stake weights, consensus scores, and navigability functions are grounded in robot behavior in space (Paykari et al., 21 May 2025).