# Solana Analysis Models 29.1-Millisecond Handoff Gain at γ = 7 By Simon Yoon Canonical URL: https://www.tokenpost.com/news/technology/32392 Published: 2026-10-15T08:44:02.000Z Updated: 2026-10-15T08:44:02.000Z Section: Technology > The recommended γ range is three to seven; at γ = 7, the model also added 5.4 leader windows to regional persistence. Solana published an analysis of SIMD-0675 on Oct. 14, examining a proposed change to validator scheduling. The recommended settings put the stake floor, σ, near 10% and the run length, γ, between three and seven. At γ = 7, modeled handoff gains reached 29.1 milliseconds, alongside a measure of longer regional persistence. SIMD-0675 would preserve stake-weighted random selection of block leaders, then reorder them so geographically compatible validators are more likely to lead in sequence. Each validator would keep the same number of leader windows; the current schedule assigns each validator four consecutive slots at a time. The stake floor sets the geographic threshold for grouping leaders, while γ determines how many compatible leaders are placed together. The recommended γ range followed tests across a wider set of values, including γ = 2, 10 and 50, with runs tested as high as γ = 200. At a 10% stake floor, γ = 7 produced 86% of the modeled handoff gain at γ = 50. Shorter runs also captured much of the modeled gain: γ = 3 reached 65%, γ = 5 reached 79% and γ = 10 reached 91% of the γ = 50 result. The gains came with a regional-persistence trade-off. At γ = 7, the mean cumulative persistence score rose by 5.4 leader windows, which converts to 4.3 seconds. In 38-window intervals, or about 30 seconds, the most represented region’s share exceeded the random-schedule baseline by an average of 0.5 percentage points at γ = 3, 5.9 points at γ = 10 and 29.5 points at γ = 50. Over 2,250 windows, about 30 minutes, the modeled regional shares mostly overlapped with the random schedule. The point at which persistence began rising more sharply shifted with the maximum run length tested, so three to seven is a recommended range rather than a universal optimum. The 10% stake floor was chosen in part to account for Asia, the smallest macroregion targeted for local grouping. Asia represented 10.4% of stake in the snapshot used for that choice; its share ranged from 10.4% to 11.6% across epochs 1000–1041. Longer regional runs can also extend exposure to a regional outage, censorship or network partition. It built 6,250 network scenarios by assigning simulated stake across the locations of 124 observed validators in six macroregions; the model introduced no additional locations. It estimated delay from distance alone, assuming fiber routes 50% longer than great-circle distance and signal travel at about two-thirds the speed of light. The estimates depend on the observed validator geography and simulated stake distributions, and may change if locations or regional stake shift. A separate, earlier simulation of the proposal put mean handoff latency at 36.2 milliseconds before the change and 17.0 milliseconds after it, a 53% reduction. That simulation used epoch-1038 stake data, 661 validators and a different measurement method, so its results are not directly comparable with the newer distance-based estimates. The [earlier simulation results]() used different inputs and methods. SIMD-0675 remained open on Oct. 15, and its activation timing was undetermined. ## Links in this article - [earlier simulation results](https://www.tokenpost.com/news/technology/25965)