Solana (SOL) published an analysis of a geo-aware leader schedule on Oct. 14, 2026, arguing that reordering the existing stake-weighted leader list by geography could reduce handoff latency without changing how many leader windows each validator receives. The proposal, SIMD-0675, is aimed at shortening the time between consecutive leaders, but Solana’s own simulations say the trade-off is higher geographic persistence and more short-horizon concentration.
What the proposal changes
According to Solana’s post dated Oct. 14, 2026, the current leader schedule assigns four consecutive slots to each validator every epoch and selects those windows according to stake. The geo-aware version does not replace that random selection. Instead, it takes the same schedule as input and changes only the order, grouping compatible leaders into bins so that nearby validators tend to follow one another. Solana said every validator keeps exactly the same number of leader windows; only the sequence changes.
The post says the design is controlled by two parameters: stake floor, or σ, and run length, or γ. In the analysis, Solana defined σ as the fraction of total stake used to determine a location’s radius and γ as the number of compatible leaders emitted together before a bin is shuffled and written out as a contiguous block.
Latency gains versus geographic persistence
Solana’s simulations covered 6,250 synthetic networks built on top of 124 observed validator locations across six macroregions. The company said the geography was kept fixed because it had no evidence of additional locations ready to host validators today, and it warned that conclusions are conditional on that setup.
On handoff latency, Solana reported that the mean gain rises with γ and falls with σ. At σ=10%, the improvement ranged from 15.8 milliseconds at γ=2 to 33.9 milliseconds at γ=50. The company said γ=3 captures 65% of the gain seen at γ=50, γ=5 captures 79%, and γ=10 captures 91%, suggesting diminishing returns from larger bins.
That gain comes with longer same-region runs. Solana measured excess geographic persistence in leader windows and said larger γ was the dominant driver. At γ=50, the excess persistence fell from about 61 windows at σ=2% to about 23 windows at σ=20%. The company also said short-horizon regional concentration increases quickly even when long-horizon regional composition remains broadly similar to random scheduling.
Why σ centers near 10%
Solana said σ faces a hard upper bound: if a macroregion holds less stake than σ, its bins stop being local and begin to mix heavily with other regions. Under the current geography, Asia is the smallest macroregion that still needs local bins, and Solana said its stake share sat at about 10.4% in the latest snapshot and moved between 10.4% and 11.6% across epochs 1000–1041. On that basis, the company chose σ=10% as the largest value below the locality threshold.
The post also said a higher σ raises the stake needed to profit from a fictitious-location attack. In Solana’s analysis, an actor declaring an isolated location can accumulate contiguous control if its stake is large enough relative to σ. The company said the incentive to misreport location increases once attacker stake reaches σ, making the stake floor part of the security boundary as well as the locality boundary.
The suggested run-length range
For γ, Solana said there is no single parameter value that maximizes latency gains while minimizing all risks. Using a Pareto-style analysis, it identified a knee near γ=7 for the regional trade-off at σ=10%. At that point, Solana said the schedule captures 29.1 milliseconds of handoff gain, or 86% of the gain at γ=50, while adding a mean cumulative persistence score of 5.4 extra windows, equivalent to 4.3 seconds on the analysis’s conversion.
The company said a more conservative choice is γ=3, which keeps 65% of the achievable handoff gain and is better if the main concern is exposure to smaller validators below the stake floor. Solana described γ between 3 and 7 as the rational range, with the precise choice depending on how much extra persistence is acceptable for each millisecond of handoff improvement.
Limits and caveats
Solana repeatedly noted that its findings depend on the observed geography and synthetic stake distributions used in the analysis. The company said the simulation did not add new locations, model regional expansion, or account for future migration beyond the observed support. It also said the knee point for γ depends on how far the frontier is extended, meaning the selected value is partly grid-dependent rather than a universal optimum.
The analysis also distinguishes the geo-aware schedule from consensus-finality changes. It concerns leader ordering, not a protocol redesign of finality. That matters because a slot is only a production opportunity, and skipped slots do not necessarily translate into block height.
Solana did not include a date for any activation of SIMD-0675. The current verified near-term network event remains the Alpenglow upgrade scheduled for Oct. 21, 2026, which is separate from this scheduling proposal.
Be the first to comment