PROBLEM 1 (latency): short in-town Auto routes (Twin Falls->Filer ~7 mi) took 60+
seconds because urban OSM-parking density exploded hybrid candidate evaluation.
Fixes in router.py:
- MIN_HYBRID_DISTANCE_KM 8.0 -> 24.0 (~15 mi): in-town trips never enter hybrid eval
at all -- this alone eliminates the reported latency on Matt's routes.
- HYBRID_MAX_TRAILHEADS 20 -> 8: fewer candidates even on long trips.
- HYBRID_OVERALL_TIMEOUT_S = 6.0: a wall-clock check inside the candidate loop bails
hybrid eval past 6 s (logger.warning) and keeps the single-mode / best-so-far winner.
- HYBRID_EARLY_ABORT_MIN = 30.0: once a candidate beats the single-mode winner by 30+
min, stop probing the rest and ship it.
- Per-stage timing logs (logger.info) in _route_auto / _try_hybrid_auto:
"single-mode probing took Xs", "hybrid candidate gathering: N candidates in Xs",
"hybrid probing took Xs across N tested candidates".
PROBLEM 2 (hybrid render): investigated the missing network polyline. The stated
hypothesis (a hybrid drive leg using a non-"network" segment_type) is DISPROVEN --
_build_hybrid_response emits segment_type=="network" for BOTH the drive and offroad
legs (verified against the live response), the OFFROUTE_NETWORK_LAYER filter matches
it, MODE_COLORS is fully defined, and the store passes data.route correctly. The one
real fragility is the MapLibre color match: if network_mode is ever null the whole
layer can fail to paint (wilderness still draws via its static color -- matching the
exact symptom). Hardened it with ["to-string", ["get","network_mode"]] so a
missing/unknown mode falls through to the blue fallback and the layer always paints.
I could not reproduce the exact missing-leg render headlessly (all backend shapes +
frontend filters are correct), so a Chrome MCP repro is recommended to confirm #2 is
resolved; if a render issue remains it should be diagnosed in-browser.
Tests: hybrid synthetic-trip distances bumped 20 -> 30 km (past the new 24 km gate);
new test_hybrid_early_abort_stops_probing; surface-change integration savings lowered
into the 15-30 min band so both candidate sources are still probed (not early-aborted).
Full offroute suite: 84 passed. npm run build: clean.
PROBLEM 3 (off-road wilderness timeout) is out of scope -- separate follow-up; the
wilderness pathfinder is untouched here.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* MVUM Layer 3b: OSM parking as multi-modal Auto transition candidates
Adds OSM parking lots as a third multi-modal-Auto transition source alongside
MVUM trailheads (3a) and surface-change points (3c), so Auto can suggest
"drive to a parking lot, switch to foot/2w/4w" trips where no MVUM trailhead
exists -- BLM/state land, urban edges, anywhere OSM has parking but the USFS
trailhead layer does not. Backend-only; consumes the already-ingested
/mnt/nav/osm-parking.db read-only (no data-pipeline change).
- mvum_parking.py: OSMParkingIndex (process-wide singleton via load_parking_index)
over a shapely STRtree of parking points, mirroring MVUMSpatialIndex /
TrailheadIndex. Read-only SQLite. Drops access in (private,no,permit) at load.
query_parking_near_line(coords, buffer_m=2000) with the same coarse-bbox +
precise-distance filter as TrailheadIndex. Records carry
{lat, lon, name, road_class="parking", parking_type, access}.
Perf note: the ingest already stored representative_point() in lat/lon, so the
STRtree is built straight from those columns -- parsing the 1.6M WKB blobs at
boot would add minutes for an identical point.
- router.py: _try_hybrid_auto generalized to gather candidates from each AVAILABLE
source (trailhead index if present + surface-change always + parking index if
present) instead of hard-returning when trailhead_index is None, so parking-only
candidates still work. Combined list keeps the existing closest-first sort +
HYBRID_MAX_TRAILHEADS cap. Signature unchanged; record shape already compatible.
- app.py / offroute_route.py: load + inject the OSM parking singleton, mirroring
MVUM_SPATIAL_INDEX / MVUM_TRAILHEAD_INDEX. Failure logs a warning, degrades None.
- admin.py: GET /api/admin/osm-parking/info -> {count, build_time_seconds,
memory_estimate_mb}, mirroring /api/admin/mvum-spatial/info.
- backend/scripts/ingest_parking.py + README-osm-parking-ingest.md: the
data-pipeline ingest lifted to the repo with argparse (--geojsonseq/--db, no
/tmp) + the download/filter/export/ingest/restart refresh recipe.
Tests: test_mvum_parking.py (loads, near-line close-only, private/no/permit
filtered, null-access kept) + test_offroute.py::test_hybrid_consumes_parking_
candidates (parking-only source probed as a leg-1 destination). Full offroute
suite: 82 passed.
Real-DB sanity (not deployed): index loads 1,489,054 usable parking objects
(182,945 access-blocked dropped) in ~12 s using ~950 MB RSS per worker; a Redfish
Lake/Sawtooth corridor query returns 8 lots. The ~950 MB/worker memory cost is
notable -- flagging for review.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* Numpy-pack OSMParkingIndex coords + lazy records to cut RSS (~950->~570 MB/worker)
Store parking coords as packed float64 numpy arrays (_lats/_lons) and the
attribute columns as interned lists (_names/_parking_types/_accesses), and build
candidate record dicts lazily in query_parking_near_line instead of materializing
1.5M dicts + 1.5M shapely Point objects up front. road_class is the constant
"parking" so it is not stored per row.
Measured on the real /mnt/nav/osm-parking.db (1,489,054 usable rows):
RSS/worker ~950 MB -> ~570 MB (~40%), build ~11 s. Across 2 gunicorn workers that
is ~1.9 GB -> ~1.14 GB.
NOTE: this does NOT reach the ~250 MB originally targeted. The remaining cost is
the shapely STRtree itself: it permanently retains the input geometries
(tree.geometries len == row count), so the transient `del points` does not free
them. Attribution on the real DB: columns-only 137 MB, retained Point objects
+230 MB, STRtree index +110 MB. Reaching ~250 MB would require dropping the
shapely STRtree for a coordinate-only structure (e.g. scipy cKDTree over the
lon/lat arrays), which changes the line-buffer query into a per-vertex radius
query -- a behavior change beyond this fix-up's scope. Flagged for a follow-up.
Tests unchanged except one assertion (`len(idx.records) == idx.count`); full
offroute suite 82 passed.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Matt <mj@k7zvx.com>
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* MVUM Layer 1: per-edge annotation + accurate closure counts
Annotate each network-leg segment with its MVUM access status for the selected mode,
using the Layer-0 MVUMSpatialIndex. No routing-decision or Valhalla changes.
- mvum_annotate.py (new): annotate_network_edges(coords,(lat,lon)), mode, spatial_index,
on_date) -> [EdgeAnnotation{coord_pair_index, matched_features, mvum_status}]. Walks
consecutive pairs, queries the index (10m buffer), applies a parallelism filter (acute
angle to the edge <=45deg, both directions), resolves per-mode access via the existing
check_access/get_mode_field/symbol_to_access (worst/most-restrictive across matches).
Mode map: foot->open (skip), 2w->e_bike_class1, 4w->atv, vehicle->highclearancevehicle;
auto is already resolved to a concrete candidate upstream.
- router.py: _route_D_network_only and _build_response annotate the network leg using the
injected self.spatial_index (getattr-guarded; skipped + debug-logged if None), attach
edge_mvum to the network feature, and add summary.mvum_closed_crossings +
summary.mvum_segments_annotated.
- offroute_route.py: inject app.config[MVUM_SPATIAL_INDEX] onto the router per request.
- DirectionsPanel.jsx: warning row when mvum_closed_crossings>0.
- tests/test_mvum_annotate.py: parallel match, perpendicular reject, seasonal closure,
symbol fallback, summary count.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* Fix seasonal date default + hoist annotation, single annotate for Auto
- Default mvum_on_date to datetime.now() at annotation time so seasonal MVUM
openings/closings actually fire in production (was effectively None -> no seasonal).
- Hoist per-edge annotation out of _route_D_network_only and _build_response into a new
central OffrouteRouter._annotate_network_segments(result, mode), invoked once at the end
of route() (annotate_mvum=True). _route_auto probes with annotate_mvum=False and
annotates only the winning candidate -> Auto runs annotation once instead of up to 4x.
Removed the inline annotation/edge_mvum/summary blocks from both scenario handlers.
Note: the central pass filters network features on properties.segment_type == "network"
(the actual tag) rather than the spec-suggested "kind", which is not a field here.
1 new test: test_route_auto_annotates_only_winner.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Matt <mj@k7zvx.com>
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
_route_auto now probes every eligible candidate (eligibility filter unchanged) and
picks the one with the smallest summary.total_effort_minutes, instead of returning the
first that succeeds. Fixes the case where a wilderness start + on-road end fell through
to foot and computed the entire network leg at foot pace. selected_mode = winning
candidate; ties keep AUTO_MODE_PRIORITY order; all-fail still returns the last error.
Wilderness leg still always foot (unchanged).
Per-leg breakdown: summary now carries wilderness_minutes + network_minutes (mirrors the
existing wilderness_effort_minutes/network_duration_minutes) in _build_response (A/B/C)
and _route_D. Frontend DirectionsPanel shows a transition badge "Auto: Foot Xmin +
<mode> Ymin" when wilderness_minutes>0 and selected_mode!=foot, else the existing
"Auto chose <mode>".
Tests: add min-time pick + per-leg breakdown; update probe-all assertions (no early
return) and make the selected-mode test time-based.
Co-authored-by: Matt <mj@k7zvx.com>
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* Rename mode constants: mtb->2w, atv->4w
Rename the OFFROUTE travel-mode identifiers mtb->2w and atv->4w across backend and
frontend. MVUM vehicle-access classes/columns (atv, motorcycle) are a separate
vocabulary and are left untouched. UI labels (MTB/ATV) and the cost.py __main__ demo
local variables (cannot be digit-initial identifiers) are unchanged.
Backend: MODE_PROFILES, MODE_TO_COSTING, MODE_TO_VALID_HIGHWAYS, AUTO_MODE_PRIORITY,
_MODES_* sets, CATEGORY_ELIGIBLE_MODES, route()/compute_cost*/_pathfind_wilderness
Literals, VALID_MODES, and tests. offroute_route.py adds a backward-compat shim mapping
legacy mode=mtb->2w and mode=atv->4w before validation so bookmarked URLs still work.
Frontend: store.js routeMode doc, DirectionsPanel TRAVEL_MODES ids + SELECTED_MODE_LABEL
keys, Panel TRAVEL_MODES ids, ManeuverList network_mode->verb map keys, api.js jsdoc.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* Fix user-visible labels: MTB->2W, ATV->4W + grep cleanup
Update the user-facing travel-mode labels to match the renamed ids:
- DirectionsPanel TRAVEL_MODES (2w -> "2W", 4w -> "4W") + SELECTED_MODE_LABEL values.
- Panel TRAVEL_MODES labels.
- Stale test comment: mode=mtb -> mode=2w.
Grep pass over frontend/src + backend/services/navi_offroute found no remaining
mode-identifier string literals to rename. Residual hits are all out-of-scope: MVUM
vehicle-access vocabulary (mvum.py, /api/mvum output, single-quoted test fixtures), the
cost.py __main__ demo locals, the offroute_route.py back-compat shim, and comments
referencing the historical names.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Matt <mj@k7zvx.com>
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
A single noisy DEM cell could fabricate a huge fake slope and make an edge
unconditionally impassable, forcing the pathfinder to route around passable terrain.
Replace the hard cliff (|grade|>max_grade -> skip) with a smooth exponential penalty:
no penalty up to max_grade, then base_time *= exp(overshoot * SLOPE_PENALTY_SCALE);
only grades whose penalty exceeds SLOPE_PENALTY_CAP (true bad data / vertical) are
dropped. Routing can now see through noisy cells while still strongly avoiding real
cliffs. Penalty only raises edge cost, so the heuristic stays admissible.
2 tests: smooth penalty traverses a >max_grade gap (finite, raised cost) yet routes
around / drops a past-cap grade; exactly-at-threshold grade incurs no penalty.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Replace MCP_Geometric in _pathfind_wilderness with a numba-jit anisotropic A* (new
astar.py): signed-slope speed (climbing != descending; tobler peaks at -0.05), hard
cliff, per-edge avg context multiplier, trail-takes-both via 256-entry lookup, per-edge
barriers (strict/pragmatic/emergency), multi-goal A* (first popped wins) with admissible
distance/base-speed heuristic. New compute_cost_multiplier_grid (slope-free context
multiplier) + exponential inflation (sigma=1.8; inf->HARD=50*p95 for blur, inf re-imposed).
numba>=0.59 added (numba 0.65.1).
fix: wilderness leg is always foot effort; mode parameter reserved for future flexibility.
_pathfind_wilderness keeps the mode param (threaded from _route_A/B/C) but hardcodes
cost_mode=foot for the cost grid, trail friction, speed function, base speed, and max
grade. Off-trail math for MTB/ATV/vehicle is not well-grounded and real-world wilderness
traversal is foot regardless (push the bike, walk past where the vehicle stops). User mode
still drives entry-point eligibility (query_radius highway filter) and Valhalla network
costing. Matches the original pre-#17 design: wilderness ALWAYS uses foot.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The old ST_DWithin(50km) scan returned ~226k candidate points near dense areas before
sorting (~3-10s/call). Replace with PostGIS k-NN ordering: ORDER BY geom::geography <->
point LIMIT k, which the existing GiST index on (geom::geography) walks nearest-first and
stops after K rows. SELECT still computes ST_Distance AS distance_m so callers see real
meters. radius_km is kept as a Python soft cap applied after fetch (drops rows beyond it),
preserving the caller expanded-radius fallback (now effectively a no-op).
2 tests: SQL contains <-> + LIMIT and no ST_DWithin; radius_km soft cap filters beyond-cap rows.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The three wilderness-scenario guards used table_exists() OR get_entry_point_count()==0
to check the index is non-empty — a full SELECT COUNT(*) that scanned the entire
2.94M-row table (~9-73s depending on load/cache). Add EntryPointIndex.has_entry_points()
using SELECT EXISTS (SELECT 1 ... LIMIT 1), which short-circuits at the first row, and
swap it into _route_A/_route_B/_route_C. get_entry_point_count() kept for admin-info/tests.
3 new tests: table missing -> False, empty -> False, rows -> True.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Align spatial eligibility with Valhalla's actual /locate vocabulary:
- road grade lives in classification.classification (8-value enum incl service_other,
which was missing from PAVED_HIGHWAY_CLASSES); track/path/footway are NOT grades,
they live in classification.use.
- _locate_on_network now also returns use (defensive .get chain; None in fallback).
- Renamed TRACK/PATH_HIGHWAY_CLASSES -> TRACK_USE_VALUES/PATH_USE_VALUES; atv/mtb now
match on use, vehicle still on the paved grade.
- 4 tests: service_other->vehicle, use=track->atv/mtb/foot, use=footway->mtb/foot,
none->foot.
Live-verified: the Boise end point (43.626,-116.215, service_other) now yields
[atv,mtb,vehicle,foot] instead of [foot]; Auto intersection picks vehicle.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Replaces the 3-tier vehicle-only discriminator with a per-endpoint eligible-mode-set
intersection:
- CATEGORY_ELIGIBLE_MODES maps OSM key:value (and key:* wildcards) to eligible modes;
_eligible_modes_from_category resolves a hint, or None when untyped.
- Untyped endpoints fall back to _spatial_eligible_modes: parallel Valhalla /locate
(auto/pedestrian/bicycle) + 3-tier snap/road-class rules (vehicle needs paved, and
paved+flat in the 5-100m grace zone; atv paved/track; mtb paved/track/path; foot always).
- _route_auto intersects both endpoints eligible sets, probes AUTO_MODE_PRIORITY within
it, and adds selected_mode + selected_mode_set. Both untyped endpoints resolve in parallel.
- _locate_on_network now returns road_class.
- api_offroute accepts optional start_category/end_category and forwards them (auto only).
Frontend: requestOffroute takes startCategory/endCategory; computeRoute passes
routeStart/routeEnd.category; startDirections preserves place.category.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
When mode="auto", OffrouteRouter._route_auto() probes AUTO_MODE_PRIORITY
(vehicle -> atv -> mtb -> foot) and returns the first mode whose network can
serve the route, tagging the result with selected_mode. route() already errors
when a mode cannot reach an endpoint, so the first status==ok is the most
road-capable feasible mode. Adds 4 isolation tests (route monkeypatched).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* Add navi-offroute service (extraction #8 — the last one)
Faithful port of recon's /api/offroute (POST) + /api/mvum (GET) and the
runtime offroute modules into a new :8428 service. Closes the loop: after
this, navi-frontend talks only to navi-backend.
Ported: router.py (OffrouteRouter, EntryPointIndex, 4 route strategies,
in-Python MCP_Geometric least-cost path, Valhalla integration, per-request
osmium extract), mvum.py (MVUMReader over navi.db), cost.py, friction.py,
trails.py, and barriers.py (runtime BarrierReader/WildernessReader only).
NOT ported (per Phase A §3/§15): prototype.py (dead at runtime), barriers.py
build_*_raster (offline GDB→raster prep). DEM imported from shared/dem.py
(PR #9), not duplicated.
Behaviour-faithful changes: hardcoded paths/URLs → env vars; the
profile.offroute.* config (osm_pbf_path/postgis_dsn/densify_interval_m) →
dedicated env vars (router drops deployment_config). Both routes public (no
auth, matching recon). PADUS via libpq peer-auth DSN (dbname=padus) — NO
secret. Owns no DB.
15 hermetic tests (offroute validation + mocked-router shape + close-always;
fixture-SQLite MVUM roads/trails/fallback/null; admin auth + no-secrets +
probe shape). Full suite 119 passed / 1 skipped. Adds scikit-image + rasterio.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* navi-offroute: PR #10 review cleanups (4 faithful-port deviations)
1. trails.py — drop recon-era "Run the Phase B rasterization script"
reference from the not-found error (confusing in navi-offroute context).
2. friction.py — add FileNotFoundError-before-rasterio-open check to
match barriers/trails consistency.
3. mvum.py — remove dead try/except shapely import + warnings.warn at
2 sites (shapely is a hard pyproject dep; the fallback was unreachable).
4. router.py — declare psutil in pyproject, drop the silent fallback;
the MEMORY_LIMIT_GB safety check was silently disabled in prod.
Adds test_friction_reader_raises_file_not_found_when_missing (16 navi-offroute
tests; full suite 120 passed / 1 skipped).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: zvx-echo6 <mj@k7zvx.com>
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
