BRouterEnricher via IBRouterService AIDL

## Goal The preferred enrichment backend: fully offline, free, no account, no terms-of-service concerns. ## Acceptance criteria - [ ] `IBRouterService` bound, with a clear message when BRouter is not installed and a link to install it - [ ] Route points passed as `lonlats` with a bike profile and `turnInstructionMode` set - [ ] Returned GPX parsed and voice hints (`TL`, `TR`, `KR`, ...) mapped to the canonical enum - [ ] Street names extracted where present - [ ] Missing segment files detected and reported as a recoverable failure so the chain falls through - [ ] Enrichment runs off the main thread with progress shown ## Files - `companion/.../route/enrich/BRouterEnricher.kt` ## Notes Depends on Spike C. OsmAnd's BRouter adapter is a working reference. ## Update — 2026-09-01: the divergence check is mandatory BRouter is a router, not a map-matcher: given via points it computes its own route under its own profile, which can differ from the imported line, and a cue sheet describing a route the rider is not on is the worst failure this app has (D25). - [ ] **Geometric divergence between BRouter's returned line and the source GPX computed on every enrichment**, and the result **rejected above 20 m** — falling through to the next tier rather than being shown - [ ] Divergence value stored with the cue sheet, and shown in the cue-sheet review (#55) - [ ] Via-point decimation follows the curve measured in #6, not a guess - [ ] `maxRunningTime` / timeout handled as a real outcome: fall through to the next tier, never a partial cue sheet - [ ] If #6 finds BRouter's reference-track mode reachable over AIDL, **use it instead** — this issue changes shape and the via-point path becomes the fallback - [ ] Availability and segment coverage handled by #64, not re-implemented here ## Update — 2026-09-02: reject per leg, not per route - [ ] The 20 m divergence check runs **per leg**, between consecutive via points - [ ] Legs within threshold keep their cues; legs beyond it have their cues dropped and the stretch marked unenriched - [ ] Wholesale rejection only when the failing-leg fraction exceeds the bound established by #6 - [ ] `NAV_CUE_CONFIDENCE` already travels per cue, so no new plumbing is needed — only the check's granularity was wrong One bad 50 m stretch must not discard a 100 km cue sheet (D41, NFR-P8).

robert commented

2026-08-31 17:14:44 +02:00

(Migrated from git.butzei.de)

Goal

The preferred enrichment backend: fully offline, free, no account, no terms-of-service concerns.

Acceptance criteria

IBRouterService bound, with a clear message when BRouter is not installed and a link to install it
Route points passed as lonlats with a bike profile and turnInstructionMode set
Returned GPX parsed and voice hints (TL, TR, KR, ...) mapped to the canonical enum
Street names extracted where present
Missing segment files detected and reported as a recoverable failure so the chain falls through
Enrichment runs off the main thread with progress shown

Files

companion/.../route/enrich/BRouterEnricher.kt

Notes

Depends on Spike C. OsmAnd's BRouter adapter is a working reference.

Update — 2026-09-01: the divergence check is mandatory

BRouter is a router, not a map-matcher: given via points it computes its own route under its own
profile, which can differ from the imported line, and a cue sheet describing a route the rider is not
on is the worst failure this app has (D25).

Geometric divergence between BRouter's returned line and the source GPX computed on every
enrichment, and the result rejected above 20 m — falling through to the next tier rather
than being shown
Divergence value stored with the cue sheet, and shown in the cue-sheet review (#55)
Via-point decimation follows the curve measured in #6, not a guess
maxRunningTime / timeout handled as a real outcome: fall through to the next tier, never a
partial cue sheet
If #6 finds BRouter's reference-track mode reachable over AIDL, use it instead — this issue
changes shape and the via-point path becomes the fallback
Availability and segment coverage handled by #64, not re-implemented here

Update — 2026-09-02: reject per leg, not per route

The 20 m divergence check runs per leg, between consecutive via points
Legs within threshold keep their cues; legs beyond it have their cues dropped and the stretch
marked unenriched
Wholesale rejection only when the failing-leg fraction exceeds the bound established by #6
NAV_CUE_CONFIDENCE already travels per cue, so no new plumbing is needed — only the check's
granularity was wrong

One bad 50 m stretch must not discard a 100 km cue sheet (D41, NFR-P8).

Rows
Columns

BRouterEnricher via IBRouterService AIDL #27