Dokumen ini mencatat tahap implementasi pada versi yang disebutkan. Untuk status akhir seluruh fitur, baca audit implementasi.
#28 — multi-leg charging recommendations
Validated 2026-09-19 on physical TECNO KF8, Android build 0.7.0-charge-plan.
The journey service now routes and compares a bounded set of charging alternatives: up to six eligible points within a six-kilometre corridor (two per corridor third), up to 24 routed connections, and two nearby destination-exit candidates. Adjacent connections are obtained before optional shortcuts. Skipping a point uses its own direct road connection; it does not inherit a detour through that charger. This is a bounded candidate search, not a global optimum claim.
Plans preserve actual starting SOC, strict reserve >20% at each arrival and sufficient energy from the destination to the next charger. A destination at exactly the same coordinates as a catalogue charger needs no invented exit leg. Targets up to 90% are evaluated first; only if that search fails are targets up to 100% considered. Partial targets and no-charge transitions are allowed. Charging cannot reduce middle-scenario SOC when it already exceeds a target required by the conservative scenario.
Ranking includes road time, charging time, five minutes setup, assumed 90% charging efficiency, 65% power factor at SOC 80–90 and 35% above 90. Effective power is capped by charger, known supply and vehicle limit when available. Unknown supply is not promoted to 2 kW: the user must enter an explicit assumption before such a station can contribute an estimated charging session. Plans remain conditional on infrastructure and model assumptions. Avoided/unavailable/incompatible stations are excluded; an OMOWAY dealer without verified charging-facility evidence is not eligible.
The UI displays both SOC scenarios per leg, charging targets/power/time, the destination-exit charger, total duration including that exit, and search limits. Starting charging planning invalidates the prior standalone route/energy display; the map draws the chosen legs and a dashed exit route. The standalone energy button stays disabled for a charging itinerary.
Verification:
- 35 Node tests pass, including charge-plan cases for actual initial SOC, destination escape, partial charging, >90 fallback, slow versus fast charging, avoidance, unknown supply, unverified dealer exclusion, unavailable station, superseded result and middle-SOC continuity.
- Two native JVM tests, Android assemble and lint pass; existing inset deprecation warnings remain.
- TECNO controlled-adapter scenario: start65%, two30km legs, one800W session38%→51.7%, arrival24.7%, then5km to exit charger at20.2%. UI showed the assumption and conditional status; these are synthetic validation inputs, not measured riding data or real facility verification.
- Runtime regression: direct prediction → charging plan clears direct summary and old energy, disables standalone prediction, and clicking the disabled button does not add an archive entry.
- Test catalogue/profile/predictions and network emulation were restored. Detailed capture is ignored at
research/runtime/issue28-device.json. - Standards and spec review findings resolved; final runtime evidence covers the previously reported UI issue.
Per-plan reduced-reserve consent and persisted plan reopening belong to #29. Multi-destination/day scheduling belongs to #30.