Stack (merge in this order): #1418 → #1426 → #1465 → #1470 → #1517 → #1520. Each PR's diff against
devholds the ones before it; this one's own changes: 696285c5, 11a75d5d and the merge 666a8550.
What
The solve joins a road's end to another road only within 1 m of its line (vsolve.JOIN_M). Silistra's track ends 1.6 m off the next track, so the solve kept it at its own profile, 3.4 m above the other track (drivability 5.8, 1.5 with the passes).
vsolve.joins already had a wider reach (half width + 3 m, like the passes' snap), off because it gave Tokyo 4 seams over 30 cm. This PR turns it on for the ends of tracks only (BNE_VSOLVE_REACH_CLASSES, default track; BNE_VSOLVE_REACH=off turns it off).
Numbers
| drivability total | Silistra | other sites | grade_metrics, 8 places | |
|---|---|---|---|---|
| without the reach | 1281.5 | 5.8 | ||
| reach for tracks (this PR) | 1277.0 | 1.4 | unchanged | identical |
| reach for every road | 1296.7 | 1.4 | Rautenweg +7.3, Südtiroler Platz +5.1, Bowker +2.7, Riva +2.6… | Lauterbrunnen seam 0.56 m (0), Amsterdam 0.17 m (0.015), corridor steps Lagos 248 → 277, Tokyo 819 → 901 |
With the next cells' roads in each solve, Tokyo's 4 seams are gone with either setting.
Details: docs/research/2026-11-21-vsolve-reach.md. New test: JoinTest.test_the_reach_is_a_track_s_alone.
test_drivability's baseline: Silistra locked in at 1.4 (11a75d5d), as the test asks when a site improves. CI failed on that before.