How Flow routes around chargers you can actually use

Planning a long electric ride means trusting that every charger in your route will actually accept your plug. Car fast-chargers outnumber every other type by ratio in most databases — but they're useless to electric unicycles, e-scooters, e-bikes, and onewheels. Flow's charger pipeline is built around one rule: a charger only appears in your route if it has a connector your vehicle can use.

The connector-authoritative rule

Flow evaluates every charger at the connector level, not the station level. A site with a DC fast-charger (CCS, CHAdeMO, or NACS Supercharger) and a J1772 is not a "DC station" — it is a station that happens to have a J1772, and the J1772 is the part that matters for you.

The principle: Flow keeps any charger that has at least one connector a PEV can use, even when that charger is co-located with DC fast equipment.

This matters because data aggregators often classify a whole site based on its highest-power connector. A site with a single DC fast unit and a J1772 pedestal gets stamped as a "DC fast" site in most databases — and a station-level filter would drop it entirely, eliminating a perfectly usable J1772 stop. Flow evaluates each station's full connector list and keeps it whenever it includes at least one AC or RV connector.

The Laytonville example

A site called GF4EVA near Laytonville, CA had two source records merged by the deduplicator: one for the L2 J1772 and one for the DC fast unit. After merging, the combined record inherited LEVEL_3_DC (the highest level present). An earlier version of the filter excluded all LEVEL_3_DC sites — which silently dropped the usable J1772 from routing and produced a false "47-mile gap" warning on the Mendocino corridor. The fix replaced the bare level-check with the connector-authoritative rule, and Laytonville became a valid stop.

What connectors work for PEVs

EUCs, e-scooters, e-bikes, and onewheels charge from AC mains — the same socket that runs your dryer or clothes iron. They cannot accept the high-voltage DC that car fast-chargers deliver.

Charger tiers in the routing score

Not all usable connectors are equally useful for route planning. Flow assigns each stop a tier based on the best connector it offers:

• Tier 1 — J1772. A standard Level 2 AC pedestal at a hotel, trailhead, or RV park. Typically 6–7.2 kW. A 2-hour stop gets you significant range.
• Tier 2 — RV hookups (NEMA 14-50, TT-30). Common at campgrounds and state parks. Slower — 3.6–6 kW — but often the only option in rural corridors.
• Tier 3 — 110V outlets (NEMA 5-15, 5-20). Emergency-level: 1.4–1.9 kW. Viable overnight; impractical for a 45-minute stop.

The tier score (1.0 / 0.7 / 0.1) is one of five weighted inputs in stop selection. The others are progress along route, detour distance, time impact, and observed reliability. A Tier 1 J1772 with a 0.5 km detour ranks higher than a Tier 2 RV hookup directly on the road — but either beats an out-of-reach Tier 1 two stops later.

Multi-source aggregation and deduplication

Charger databases have different coverage and different biases. NREL/NLR has good US coverage and reliable operational-status data. Open Charge Map has richer connector detail in some regions. OSM Overpass picks up rural and community-contributed chargers that commercial databases miss. Community chargers entered directly in Flow cover spots that none of the above include — a friendly home outlet, an off-grid RV park.

For every route segment, Flow queries external sources in parallel: NREL/NLR, Open Charge Map, and OSM Overpass, plus RIDB for campground hookups where they are relevant. Community chargers added directly by riders appear on the map and nearby-charger lookup but are not queried per route segment. The raw results overlap — the same physical J1772 pedestal often appears in two or three databases with slightly different names and coordinates.

The deduplicator resolves this with two rules:

• Within 25 meters of an existing record: always merge, regardless of name.
• Within 50 meters: merge if name similarity exceeds a threshold (normalized Levenshtein, after stripping words like "station," "charging," "EV," and level designations that differ between APIs).

When merging, connectors from both records combine — so if NREL has the J1772 and OCM has the DC fast unit at the same site, the merged record holds both. The chargingLevel field is recomputed from the merged connector list to reflect the highest level present. Source priority for metadata (name, address, network) follows NREL > OCM > OSM > RIDB > Community.

Search geometry

The router doesn't do a single bbox query around origin and destination. It splits the route polyline into ~50 km segments with 10 km overlap, queries each segment independently, and deduplicates across segments. This prevents API result caps (OpenChargeMap: 500, NREL: 200) from being consumed by urban clusters at the start of a route — ensuring rural chargers along the corridor are actually discovered.

The search buffer is 8 km either side of the polyline. That's wide enough to catch coastal town chargers where Mapbox's driving-profile polyline can drift several km inland from the actual coast road — but narrow enough that obvious off-route stops don't make it into the candidate pool. After the buffer query, each charger is projected onto the nearest polyline segment to compute its exact detour distance and route-position, so scoring reflects actual detour cost rather than raw perpendicular distance.

See also: How Flow avoids freeways and finds good roads →