A route is not just a distance target. A real cyclist doesn't ask only for 60 kilometers. They ask for 60 kilometers to the south-east, away from the city, on quieter roads, with the right character for the day. Until now, most routing engines (and most AI wrappers on top of them) were better at matching the number than preserving the feeling. BikeScout now changes that.
Today we introduce a major routing upgrade: Directional Route Bias with Multi-Seed Candidate Selection. This system allows BikeScout to interpret requests like "keep it south and east", "avoid urban streets", and "prefer countryside roads" as real tactical constraints, not just decorative language around a distance request.
The Problem: Distance Was Winning Too Often
Traditional route generation is biased toward what the engine can easily assemble. Around a dense city, this often means a route will drift toward urban networks simply because they offer more connection options and make it easier to hit a target mileage exactly.
That creates a familiar problem for riders: you ask for a road loop with open countryside character, but the engine quietly pulls you back toward busier roads, denser grids, and less desirable terrain just to satisfy the number on the screen.
The New Logic: Route Character Becomes a First-Class Constraint
BikeScout now supports a set of explicit routing preferences that shape route generation before the final result is chosen:
- Direction Bias: steer the route north, south, east, west, or combinations like south-east.
- Avoid Urban: push selection away from denser city sections when possible.
- Prefer Rural: favor quieter, lower-density countryside riding.
- Distance Flex: allow the route to deviate slightly from exact mileage when preserving ride quality matters more.
- Priority Mode: choose between balanced planning, exact distance matching, or ride-character-first logic.
The Tactical Engine: Directional Anchors
To make directional requests real, BikeScout now generates a virtual anchor point in the requested quadrant. If a rider asks for a loop that stays south and east of town, the system computes a tactical anchor in that direction and uses it as a target reference during route selection.
User intent → virtual anchor point → multiple route candidates → select candidate with best anchor proximity and acceptable distance error
This is the key difference: BikeScout no longer treats direction as a vague textual hint. It turns it into geometry.
The Selection Upgrade: Multi-Seed Candidate Search
Generating a single loop is not enough when the rider has strong spatial intent. So BikeScout now runs multiple round-trip candidates using different routing seeds, then scores them based on two tactical criteria:
- Distance Error: how far the result is from the requested target mileage.
- Anchor Proximity: how closely the route passes through the requested directional zone.
This means the selected loop is no longer just "valid." It is the best available compromise between geography and mileage.
Smarter Flexibility: When 60 km Should Not Mean 60.00 km at Any Cost
Exact distance matching is not always the most important part of planning. If the rider says "about 60 km" or signals that the ride character matters more than precision, BikeScout now relaxes the target distance within a configurable tolerance window.
That flexibility is critical. It allows the engine to reject awkward urban detours and instead choose cleaner, more coherent loops that better reflect the requested experience.
Real-World Impact
This feature is especially powerful for riders living near cities or mixed urban-rural boundaries. A request such as:
now gives BikeScout the tools to preserve the intended direction of travel and the desired ride atmosphere, rather than simply optimizing for a mathematical loop closure near the start point.
What This Unlocks Next
This release is the first major step toward truly intent-aware route generation. The next tactical upgrades will build on the same system with:
- urban density penalties based on map topology
- road-class preference scoring for quiet roads versus arterial links
- stronger rural/countryside inference
- surface-aware directional selection for road, gravel, and MTB modes
In other words: BikeScout is moving from route calculation toward route interpretation. The engine is no longer asking only, "Can I build a loop?" It is starting to ask, "Is this the loop the rider actually meant?"