Refreshing the Travel-Time Map Behind Lyft’s Marketplace: Rebuilding Neighborhood Reachability…

This post details the rebuilding of Lyft's Neighborhood Reachability Signal dataset. Key technical changes include: 1. Incorporating 'drivable geohashes' derived from road network data to ensure only navigable cells are included, fixing issues like over-water ETAs. 2. Implementing a gentler pruning strategy for geohashes, retaining all geohashes with at least one rider session in established regions and using the full set of drivable geohashes for new/expanding regions to improve coverage. 3. Adding new fields to the dataset, such as rider-session density, to allow downstream consumers to shape the data without maintaining separate patches. The pipeline's stages (geohash universe decision, pair formation, ETA estimation via historical data or routing simulation, and file emission) are also described.