California tolls: how to maintain

Files: /srv/toll/draft/CA, last built 2026-10-10 01:57 ET · the CA toll page · map test · facts below are read from the files each time this page is built (python3 /home/geo1/system/handbuilt/maintain_pages.py CA)

Procedure for this state

California, built in parts (ca_state.py runs ca_bridges.py, ca_tca.py, ca_sr125.py, ca_express.py): the seven BATA bridges and the Golden Gate (one direction each), 17-Mile Drive, The Toll Roads of Orange County (rate card by point, day, direction, time; Pacific times written in Eastern), South Bay Expressway (SR 125, priced by trip), and the express lanes (told, never charged).

When to check

• January 1: BATA's 2026-2030 schedule (bata_2026_toll_increase) steps up every year; the license-plate and invoice premiums start 2027.
• July 1: Golden Gate (ggb_toll_increase_2026) and the TCA rate card (tca_ratecard_2axle, three pages: 2, 3-4, 5+ axles).
• SR 125: SANDAG's schedule PDF (sbx_toll_schedule_2020, rates 'Effective 06/30/12').
• FasTrak pages refuse rapid scripted fetches (a bot page): re-fetch on b100 with cfpage.mjs.

Rebuild

bash chain.sh CA (long: the California highways file has 3.3 million ways). TCA ramp positions: tca_ramps.py; SR 261 ramps are assigned by hand in ca_tca.py (POS) with the reason.

Watch out

• TCA mainline plazas have no published position: each is the only OSM mainline gantry pair on its road segment (recorded per point).
• SR 125 Pay-by-Plate: SANDAG states its schedule is $2 higher than the machines (sbx_511sd_pay_by_plate; asserted). LA Metro's site refuses all scripted fetches: its lanes are told at all hours.
• Holidays and the BATA multi-bridge and carpool discounts are not modelled.

The chain every state shares

Hosts: geo1 (192.168.50.220) holds the sources, builders, drafts, map and API; b100 (192.168.50.14) is the test client. All times are Eastern.

1. Refresh a source

1. Fetch into a new dated folder: python3 /home/geo1/system/handbuilt/fetch_list.py CA [name ...] (no names = every line of urls.txt). A new source: add a line to urls.txt first, then fetch it by name.
2. A site that refuses geo1 (HTTP 403) or scripts: fetch it on b100 in headless Chrome with node ~/apitest/fetch_b100.mjs <out> <name>=<url> ..., then rsync -a <out>/ geo1@192.168.50.220:/home/geo1/maps/toll/CA/sources/. Its meta.json records that it came from b100.
3. Compare the new file with the previous dated one (diff, pdftotext -layout for PDFs). Same content: nothing to do. Changed amounts or a new effective date: rebuild (step 2) and update the expected amounts of the affected regression trips from the operator's own figures.

1b. The corpus (NJ, PA, VA start from it; DE is published from it)

python3 /home/geo1/maps/toll/bin/tollup.py CA update fetches the adapter's sources (bin/adapters/ca.py), stores a document only when its content changes, writes one rate file per version under /home/geo1/maps/toll/CA/rates/ and appends every price change to history.csv (every run to runs.log). TOLL_WEB=/home/geo1/maps/toll/corpus python3 /home/geo1/maps/toll/bin/publish.py CA writes the corpus /home/geo1/maps/toll/corpus/CA from it (DE: plain publish.py DE writes /srv/toll/DE). Never hand-edit either. Then rebuild the draft (step 2) so it picks up the new rates.

2. Rebuild

• Amounts or schedules changed, toll points unchanged: cd /home/geo1/system/handbuilt && python3 ca_state.py
• The full chain in one command: bash /home/geo1/system/handbuilt/chain.sh CA (logs in /tmp/ca1.log ... ca6.log; it prints the points, any point on no road and the untied count). The builders written since October 2026 share state_kit.py (sources, booths, terms, rows, write) and own_state.py (keeps a state's untied list to toll nodes inside the state, by the local geocoder; cache osm/node_state.json).
• Toll points added, moved or renamed, or a new OSM extract: the full chain (about 2-4 minutes):
  cd /home/geo1/system/handbuilt; H=$(ls osm/ca_highways_*.geojsonseq); J=$(ls osm/ca_junctions_*.geojsonseq)
python3 ca_state.py && python3 osm_streets.py CA $H $J && python3 ca_state.py && python3 osm_snap.py CA && python3 toll_ways.py CA && python3 ca_state.py
• Read the end of toll_ways.py's output: nodes claimed by two points: {} and points on no road: [] must both be empty.
• A point whose booth is an OSM way (OSM way <id>) needs that way in osm/ca_lane_ways.json (the builder writes it). A point on a road priced entry to exit needs gantry in its location name, or toll_ways.py treats it as a position-only exit marker.
• Rebuilding NY changes osm/ny_not_toll_nodes.json (congestion-zone gantries that are not tolls); every state's untied list leaves those out, so rebuild NJ and PA after NY.

3. Load it into the API and test from b100

1. The API reads the toll pages and tolls.js when it starts. Restart it: cd /home/geo1/maps/api && kill $(cat logs/test.pid); (nohup /home/geo1/maps/node/bin/node server.mjs > logs/server.log 2>&1 & echo $! > logs/test.pid) (it is started by hand, not a service).
2. From b100 only: python3 ~/apitest/run.py ~/apitest/site/cases.json. Every trip must be OK. Add a trip for every rate or rule you change, with the amount taken from the operator's own source.
3. Engine change (tolls.js): edit /home/geo1/maps/web/tolls.js, check with /home/geo1/maps/node/bin/node --check, copy to /srv/maps/tolls.js with sudo install -o www-data -g www-data -m 644, restart the API, run the trips.

4. OpenStreetMap refresh (all states at once)

The builders match toll points to the OSM way ids the router drives, so the extracts here must come from the same us-latest.osm.pbf the router (Valhalla, /home/geo1/maps/valhalla_data) was built from (now 2026-09-27). New OSM data means: rebuild the router tiles, then for each state cut the box, rebuild its OSM inputs and run the full chain, then compare untied_toll_nodes.csv and the estimated positions below. The box for CA and the cutting commands are in the state's procedure above; the toll-node and way index comes from python3 tollnode_ways.py CA.

5. Publish

A rebuild writes the draft (/srv/toll/draft/CA); the published page (/srv/toll/CA) does not change. python3 /home/geo1/maps/toll/bin/promote.py review CA lists the draft for review: the map and the API then read it instead of the published page. Run the regression trips from b100. After approval, python3 /home/geo1/maps/toll/bin/promote.py publish CA copies the draft to /srv/toll/CA (the previous page is kept under /home/geo1/maps/toll/CA/published/<time>/) and takes it off the review list. After a change, update the coverage line for CA in the API page (/api.html, generator /home/geo1/maps/api/doc/api_page.py) and rebuild this page: python3 /home/geo1/system/handbuilt/maintain_pages.py CA.

Sources (41)

Each fetch is kept in its own dated folder under /home/geo1/maps/toll/CA/sources/<name>/; the builder always reads the newest one, so a new fetch takes effect at the next rebuild and the old one stays for comparison.

Facilities (28)

Positions placed by estimate (16): re-check when OpenStreetMap or the operator changes

Decisions by rule

R2: 31, R2a: 11, R2b: 9, R3: 16, direction: 9, not: 3, optional,: 14, plan: 4 (all in decisions.csv).

Open gaps (0)

None.

Untied OSM toll nodes: 0

Toll booths / gantries mapped in OpenStreetMap on or near this state's roads that no toll point claims; a route through one lists it under unpriced. After an OSM refresh, look at new ones here (untied_toll_nodes.csv): a new gantry on a priced road means a point is missing.

Regression trips for CA (11)

Run from b100 (never on geo1): python3 ~/apitest/run.py ~/apitest/site/cases.json. Every trip must come back OK; a changed amount is either a rate change (update the expected amount from the operator's own source) or a fault.