Dokumen ini mencatat tahap implementasi pada versi yang disebutkan. Untuk status akhir seluruh fitur, baca audit implementasi.

#26 — controlled Valhalla highway-preference comparison

Checked 2026-09-19 against https://valhalla1.openstreetmap.de/route using exactly two serialized POST requests, separated by more than one second. Public approximate citypoints only; no home/account data. Successful raw response bodies are retained under tests/fixtures/route-highways-0.json and tests/fixtures/route-highways-1.json relative to OMOX.

The requests were identical except for costing_options.motorcycle.use_highways:

{
  "locations": [
    {"lat": -6.981, "lon": 109.14, "type": "break"},
    {"lat": -6.895, "lon": 109.379, "type": "break"}
  ],
  "costing": "motorcycle",
  "costing_options": {
    "motorcycle": {"use_highways": 0, "use_tolls": 0, "use_trails": 0}
  },
  "units": "kilometers",
  "shape_format": "polyline6"
}

Identifying headers were User-Agent: OMOX-private-research/0.3 and X-Client-Id: omox-private-research. No further routing requests were made for this comparison.

Evidence use_highways: 0 use_highways: 1
Provider distance 58.737 km 41.760 km
Provider time 6,664.673 s (~111.1 min) 3,988.727 s (~66.5 min)
Toll / highway / ferry flags All false All false
Warnings Absent Absent
Decoded geometry vertices 2,169 1,041
Southernmost route latitude -7.087161 -6.980983
Northernmost route latitude -6.895014 -6.860163
Notable returned street names Jalan Raya Slawi-Jatinegara; Jalan Pancasila; Jalan Dk Lumpang Sambeng; route 158 Jalan Banjaran Raya; Jalan Pala Raya; route 1 / Jalan Raya Tegal - Pemalang

Both shapes decode to identical snapped endpoints: [-6.980984, 109.140069] and [-6.895015, 109.378998] in latitude/longitude order. Thus the comparison is not explained by different endpoint snapping. The zero-preference shape goes substantially south; the one-preference route uses the Tegal–Pemalang/Pantura street name. These observations come from returned geometry and maneuver names, not a claim that every intermediate road was visually or physically inspected.

Changing this one preference shortened the returned route by 16.977 km (28.9%) and about 44.6 minutes. This reproduces the earlier detour and demonstrates that the selected costing preference materially caused a less direct route for these points. It does not prove the new route globally shortest, legally current or physically passable.

Regression expectation: accept a valid motorcycle route containing ordinary trunk/Pantura roads when the retained restriction evidence permits it; do not equate all major roads with motorway. Keep independent toll/motorway restriction validation, rather than assuming a high highway preference proves safety or a low preference prohibits such roads. Current upstream summary has_highway is set for motorway classification, not every trunk road; see the provider decision research and upstream trip-leg builder. Upstream source does not pin the hosted server's deployed commit.

Fixture integrity (SHA-256 of the encoded trip.legs[0].shape string):

Use these recorded responses in deterministic parser/regression tests instead of repeatedly querying the public service. Distances are observations of this snapshot, not permanent assertions about future provider data.