Building a trams-only city (the ten non-French, non-Czech T1 cities)

Written 2026-09-30 by the tram kit session (docs/handoff_tram_kit_2026-09-30.md),
from the built trams-only and tram-heavy cities (Riga, Dublin, Aarhus and the
French five) and the owner's calls of 2026-09-27 to 09-30. France has
france-tram-city and Czechia its own kit. These ten cities are spread over
seven countries, one or two per country, so this skill carries only what they
share. Anything about one country is in that city's brief
(docs/build_briefs/<slug>.md, which staging writes) and in the built city the
brief names as its template. japan-city, taiwan-city and brazil-city stay
the model for a country skill. Write one only if a country here gets a third
city.

Builds are HELD (owner, 2026-09-30) until the owner gives the go. The
kit's 24 calls were approved on 2026-09-30, but that is not the go. Build on a branch, never on master: app/ lands
at review time only (docs/review_time.md).

Order of work for one city

1. python scripts/brief_check.py <slug>. A failing claim means the brief
   needs correcting. Exception: "MIRRORS DISAGREE" is an Overpass host
   problem, so re-run it (Liepāja, 2026-09-30).
2. Read the brief's "For the owner, with the build" table: mode,
   coverage, scope, lines, rings and each call with its recommendation. If
   the table is missing, ask staging for it before writing code.
3. Zurich only: add-country first (Switzerland is new; see its sheet).
4. add-city Step 0 is already in the brief. Go to scaffold-city:
   scripts/scaffold_city.py ... --mode <mode> --dry-run, then the real run.
   There is no batch scaffold (DECISIONS "The tram kit: no batch scaffold").
   Then copy the template city's config for everything scaffold_city.py
   leaves generic. The sheet below names each city's template.
5. Set the rings (section 3) by hand, and the CRS where section 3 says to.
   scaffold_city.py writes the standard rings and a WGS84 UTM zone.
6. Find the operator's own per-line stop counts (section 2, "Gate 3")
   before writing step 1, and put them in config, or record why none exist.
   The brief's station count is not one: it is usually OSM's.
7. Write fetch_sources.py, which always downloads the feed (section 4),
   then step 1, step 2 and the render.
8. Fill the page scaffold_city.py wrote (it is already in the city-page
   format) from section 6, and write the city's sections of
   docs/excluded_categories.md and docs/data_sources/<country>.md. The
   format for all three is docs/city_page_format.md; never start a page by
   copying an older one. Then run "Before publishing".

The owner's standing calls - do not re-ask

• Trams-only maps are approved (DECISIONS "Yes to trams-only maps",
  2026-09-29). With no metro, the trams are the rapid transit (Riga's
  precedent), and every tram stop gets rings.

• No stop thinning on the tram list (DECISIONS "Tram audit Phases 3-4",
  owner). Riga's 0.5 mi filter does not carry over, even to the two Latvian
  cities built on Riga's modules. New Orleans keeps every stop too
  (owner, 2026-09-30, call 13), so no city on this list is thinned.

• Rings follow the spacing rule (DECISIONS "Tram cities use the existing
  spacing rule", owner, 2026-09-29), section 3.

• The light-rail test (docs/tram_city_list.md, owner, 2026-09-29) decides
  whether a system is a tram at all, and it decides mode (section 1).

• Feeds are rolling: fetch at build, never cache one across weeks.

• The currency rule (the master list's "Five rules", owner, 2026-09-29):
  the source drops closed businesses, its newest row is under five years old,
  and the data date goes on the page. Kansas City (frozen 2026-01-15) and
  Göteborg (undated rows) carry it in the page text (section 6).

• Denmark's placement is OSM's DAR address points (owner, 2026-09-27),
  keyless and under ODbL. The Datafordeler account is reopened only if a build
  proves it needs one.

• Rejseplanen's GTFS is never used (owner, 2026-09-29, for Aarhus).

• Zurich's partial retail is the alcohol-licensed shops, on the tobacco-
  retail precedent of Seoul and Gyeonggi, with the gap disclosed (owner,
  2026-09-30).

• Den Haag's horeca layer is used on Amsterdam's precedent (owner,
  2026-09-30): credit "Gemeente Den Haag", and never call the layer current or
  complete.

• A GTFS feed that declares no licence is not used: take OSM (Olomouc,
  owner, 2026-09-30). Daugavpils and Liepāja follow it.

• RideKC's GTFS is not used (owner, 2026-09-30): Kansas City's rail is
  OSM.

• The kit's 24 calls, approved as recommended (owner, 2026-09-30: "Agree
  with recommended options for all"; the numbers are the handoff's):

  1. the page-text template (section 6);
  2. a route is drawn only if it runs at least every 20 minutes by day, and
     a stop served only by slower routes is listed as infrequent (section 1);
  3. the project's own colours where the source has none (Odense, Daugavpils,
     Liepāja, Kansas City, Tucson);
  4. Odense's SDU Syd/Hospital Nord added by node, and Hospital Syd out until
     it opens;
  5. Odense's mode is tram;
  6. Daugavpils's routes under the call 2 floor (superseded by call 28:
     every route is drawn);
  7. Daugavpils and Liepāja are narrowed, "Merged";
  8. notice 42 extended to name VZD's address register, in VZD's wording
     with the year;
  9. Liepāja's Brīvības iela and Klaipēdas iela added by node;
  10. Kansas City keeps licences valid for 2025 and 2026 only;
  11. Kansas City's dba_name withheld where it reads as a person (Houston's
      rule);
  12. Kansas City's fee-code types dropped, with the count on the page;
  13. New Orleans keeps all 110 stops;
  14. New Orleans drops "Special Events-Other (Vendor)" and "Home
      Based-Office Use Only";
  15. Tucson's ACC_NAME withheld for Sole Proprietorship, Individual and
      Married;
  16. Florence keeps its 639 exempt food rows, filtering what the type code
      names as non-public, on Milan's fuori piano precedent;
  17. Florence is commune-only, with T1's 4 Scandicci stops listed as
      outside;
  18. Zurich's Forchbahn is left out and named on the page;
  19. Zurich's Glattalbahn lines each take the stub test, and a stub is
      dropped;
  20. Göteborg's blank-typ rows are classified by name where the name
      shows a counter, and the rest dropped;
  21. Göteborg's lines 4 and 12 are drawn to their ends, with the Mölndal
      stops listed as outside, unless the stub test fails;
  22. Den Haag's RandstadRail E is left out as a stub;
  23. Den Haag is narrowed, "Merged";
  24. Den Haag's 158 pending horeca permits are left out.

  Calls 18-24 were approved before their briefs landed. If a brief's
  measurement changes the facts behind one (a Glattalbahn line that is not
  a stub, a Mölndal line that is), the call goes back to the owner.

• The four calls from the last briefs (owner, 2026-09-30):

  25. Zurich's tram 20 (Limmattalbahn, 4 of 26 stops inside) is left out as
      a stub;
  26. Den Haag's tram 1 (19 of 37 inside, 51%) is drawn to its end, with the
      stops outside listed;
  27. Den Haag's page carries the permit layer's date (section 6);
  28. Daugavpils draws its whole network, routes 1-5, and the page states
      the waits. In the owner's words: "like with buffalo 50% of network
      shouldn't fall if the norm is longer waits." Call 2's floor would have
      kept only route 1, 19 of the 38 stops.

1. Is it a tram, and what is its mode?

The light-rail test (docs/tram_city_list.md) has three parts, read
against San Diego, Calgary and Edmonton:

• Track: tunnel and bridge share, and OSM's light_rail/tram split.
• Frequency: every 15 minutes or better by day. This is a gate only on
  converted railway (Aarhus's L1). On purpose-built track a slower timetable
  is disclosed on the page (Buffalo's flat 20).
• Spacing: about 550 m or more. This supports the verdict but is not the
  gate.

A city that passes left the tram list on 2026-09-29. All ten here stayed,
so they are trams, and mode is "tram" unless step 1 keeps something of a
higher order. mode is the highest-order mode the map DRAWS, decided from
what step 1 keeps (GTFS route_type, OSM route=), never from a marketing
name. France's feeds showed that mode flags lie (Reims's tram is typed metro).
Here that means:

City	mode	Why
Den Haag (EDGE)	tram (the brief, 2026-09-30: every drawn line, RandstadRail 3 and 4 included, is OSM route=tram)	Line E is left out as a stub (owner, call 22): 4 of its 23 stops are in the city, as Ostrava's line 5 and Madrid's ML2 and ML3 were. So metro is never drawn
Zurich	tram	The Forchbahn (S18, OSM route=light_rail) is left out (owner, call 18) and named on the page with the S-Bahn. Inside the Stadt it runs on tram 11's track and stops, so drawing it would add no rings
Odense	tram (owner, call 5)	Aarhus, its template, is light_rail (its new tramway passed the test). Odense's Letbane is OSM route=tram and street-running, and it stayed on the list
the other seven	tram	none

Frequency: disclose the network's norm, drop only an outlier (owner,
calls 2 and 28). Street trams have no frequency gate. Read each route's
daytime headway from the operator's own timetable.

• Where longer waits are the network's norm, every route is drawn and the
  page states the waits (call 28, Buffalo's precedent). Daugavpils is the
  case: route 1 every 10-15 minutes, the Stropu loop (3 and 5) every 20-30
  minutes each way, routes 2 and 4 about hourly. All five are drawn, because
  dropping the slower routes would take half the network (19 of 38 stops).
• Call 2's 20-minute floor (Buffalo, the slowest drawn) is for an outlier
  route: a slow route on a network that otherwise meets it, such as
  Zurich's 50 and 51 if they turn out not to run by day. Its stops, if no
  other route serves them, go to excluded_stations.csv as infrequent,
  Aarhus's class (app/station_scope.py), and the page says so.
• The line between the two is the owner's. When a slow route is more
  than an outlier, bring the share of stops it alone serves.

2. Stations - every stop

• Every stop in scope gets a ring. Collapse platforms to stations and
  thin nothing. pipeline/stations.py's gates still run: spacing, per-line
  counts, and the boardable filter.
• GTFS feeds: take the feed's own parent_station row first. Without
  one, collapse by unchanged name as the template city does. On an ODbL
  feed, France's pure-extract rule applies (france-tram-city section 2):
  the first platform in stops.txt order, never a mean, rename nothing. The
  rule came from the French portal's ODbL conditions, so it does not reach
  a CC0, CC BY or public-domain feed. A rename or a merge of two differently
  named stops goes to the owner.
• OSM rail: route-relation membership, never a tag search (osm-rail).
  One Overpass query per city. Bound every name search with a bbox. Two
  relations per line, so draw the one with the most way geometry
  (map_common.load_osm_line_shapes). The collapse is Aarhus's
  (pipeline/aarhus/step1_stations.py): stop members collapsed by name at
  their mean within 200 m, exiting if a name spreads wider.
  • A tagged stop missing from every route relation is added by its node,
    named in config (STATION_ADD). Odense's SDU Syd/Hospital Nord
    (opened 2023-08-25) and Liepāja's Brīvības iela (the named terminus) and
    Klaipēdas iela are the cases. Check each against the operator's own stop
    list.
  • A stop that is not yet open stays out and becomes a watch item
    (Odense's Hospital Syd, 2027).
• New Orleans keeps every stop (owner, 2026-09-30, call 13). Its
  streetcar stops stand a block or two apart (164 m median), so the rings
  merge into a band along each line. The kit's handoff had named it the one
  exception for the street-stop filter, but the filter's own shape test
  (docs/sub_transit_line_filters.md, "When this applies") fails. That
  filter is for a central corridor with sparse stations plus dense surface
  branches (Muni Metro, Boston's Green Line). New Orleans is uniformly
  dense, with no corridor, the case the doc says not to thin. Thinning
  would also leave a business a block from a stop outside the rings. The
  band is a true reading of distance from the line, and the page says so
  (section 6).
• A line mostly outside the scope is a stub question, as for the Czech
  and French cities. Decided (owner, 2026-09-30):
  • Florence's T1 is drawn to its end, with its 4 Scandicci stops listed as
    outside (call 17);
  • Den Haag's E is left out as a stub (call 22);
  • Zurich's Glattalbahn lines take the stub test one by one, and a stub is
    dropped (call 19);
  • Göteborg's 4 and 12 are drawn to their ends, with the Mölndal stops
    listed as outside, unless the stub test fails (call 21).
    Where the business source covers the city only (Zurich, Göteborg, Den
    Haag's horeca layer, Florence), the scope cannot go regional, so
    France's scope rule does not apply.

Gate 3 - the operator's own stop count, every city

Required, not optional. pipeline/stations.py's gates 1 and 2 (spacing,
boardability) run by themselves. Gate 3 runs only when step 1 passes the
operator's figures in, and in the first ten builds it ran once (Odense,
25 of 25). Four drafts said "No gate 3" and five said nothing (the tram kit
retrospective, docs/tram_kit_retrospective.md). It is the one check
outside the data: Toronto's 118 and 77 agreed with every internal check and
with nothing the TTC published. On OSM rail it matters more, not less.
OSM is a volunteer map, and the brief's station count is usually OSM's own.

• Config carries it, as Odense's does (pipeline/odense/config.py):

  • OPERATOR_STATION_COUNTS = {ref: n}, one entry per drawn line;
  • OPERATOR_COUNTS_SOURCE, which names the page or document, its
    publisher and the date read.
    Step 1 passes it to verify_stations(expected_per_line=...), with the
    per-line counts taken from the collapsed stations
    (odense/step1_stations.py).

• Count the whole line, before the scope split. An operator lists every
  stop on a line, including those past the city boundary (Florence's T1 into
  Scandicci, Göteborg's 4 and 12 into Mölndal). Compare per line, an
  interchange counting once on each line it serves.

• Reconcile in config, never by loosening. Where the operator's figure
  and the build differ for a known reason, write the reason beside the
  figure. Odense: "26 stops less Hospital Syd, not yet open". Other known
  reasons:

  • a STATION_ADD stop the operator lists;
  • a stop the operator counts per direction;
  • a temporary line's stops for the current timetable period (Zurich's 50
    and 51 run to 2026-12-12).

• verify_stations prints a mismatch; a tram step 1 stops on one. Wrap
  it like this:

  res = station_gates.verify_stations(...)
  if res["per_line_mismatches"]:
      sys.exit(...)  # name each line with both figures

  Do not change stations.py to raise: other cities call it.

• The source must be independent of OSM. In order of preference:

  • the operator's own line pages, stop list, network map or timetable
    PDF, read for the count only, never republished;
  • a public authority's page for the line;
  • a dated secondary source (English Wikipedia's table of lines, as
    Stockholm's gate 3), named as secondary in OPERATOR_COUNTS_SOURCE.

  These are never the source:

  • OSM, or a brief count derived from OSM. That is circular, as Kansas
    City's station check was: it compared OSM against the brief's 19, which
    came from OSM.
  • A feed or site the owner barred (RideKC's GTFS and its schedules). Look
    instead for another publisher of the same line (for the KC Streetcar,
    the KC Streetcar Authority's own site, if it lists the stops).

• Where nothing usable is published, say so in config:
  OPERATOR_STATION_COUNTS = None, plus OPERATOR_COUNTS_GAP, which records
  what was looked for, where, and the date. The draft entry repeats it, and
  the drafts file and review time list it as a gap. A silent skip is the
  failure this section exists to stop.

• Back-filled 2026-10-01 in the nine built without it, with 16 earlier
  cities (DECISIONS, "Gate 3 back-filled in 25 cities"). It found real gaps
  in New Orleans, Daugavpils and Den Haag, each OSM missing a stop the
  operator serves: those step 1s print the mismatch with the exit commented
  out until the owner's station fix (PLAN.md, "Gate 3's station calls").
  A new tram build runs gate 3 before it is committed. A mismatch is a
  station fix with a drift check, not a page edit.

3. Rings - the spacing rule

Measure the median nearest-neighbour gap among the stations in scope,
after collapse, with every stop kept.

• About 550 m or less: RING_EDGES_MILES = [0.0, 0.05, 0.1, 0.2, 0.3],
  RING_LABELS = ["0-0.05 mi", "0.05-0.1 mi", "0.1-0.2 mi", "0.2-0.3 mi"]
  (Aarhus's config). The config comment records the figure.
• Over about 550 m: the standard [0.0, 0.1, 0.2, 0.3, 0.6]. Drop the
  page's half-size sentence. A build figure between 540 and 570 m goes to the
  owner with both in-ring shares. Kansas City read about 560 m as a mean at the
  screen, but 413 m as a median over OSM's 19 stops in its brief, so it
  takes halved rings.
• The briefs' gaps (2026-09-30): Odense 441 m, Daugavpils 298, Liepāja 329,
  Kansas City 413, Florence 322, Tucson 265, New Orleans 164, Zurich 283,
  Göteborg 378, Den Haag 335. All ten take halved rings; recompute on the
  stations actually drawn.
• Riga is on the standard rings with thinning. Its Latvian followers are
  NOT: no thinning and halved rings, as above.
• Distance in a metre CRS, never EPSG:4326, and the CRS is per city:
  Odense EPSG:25832 (Aarhus's, DAR's own); Daugavpils EPSG:32635;
  Liepāja EPSG:32634 (21.0° E, not Riga's 35N); Zurich EPSG:2056
  (LV95; the register's ekoord/nkoord are already in it). The US,
  Italian, Swedish and Dutch cities use their own WGS84 UTM zone, which
  scaffold_city.py derives from the longitude: Göteborg is 32N, not
  Stockholm's 34N. Override the scaffold only for Odense and Zurich.
• After the city lands: python scripts/ring_rules_table.py --write.

4. Feeds are rolling, and OSM is the fallback

• fetch_sources.py always downloads the feed, even when
  data/<slug>/raw/ holds one. Record the fetch time and the feed's own
  feed_info.txt window (or the publisher's metadata) in
  outputs/<slug>/provenance.json, which the page's caption reads. A step
  never fetches (check_no_fetch_in_steps.py).

• Read the licence before a feed is used, with one licence-read per
  source (read-licence). RTA's and GEST's GTFS are unread, and neither
  brief uses them. RideKC's GTFS is barred (owner, 2026-09-30): ridekc.org's
  site terms restrict schedules and require written consent, Philadelphia's
  shape. So Kansas City's frequency is stated as a fact, without citing the
  schedule, and the KC Streetcar and RideKC logos are never used.

• OSM through osm-rail is the rail source for every brief so far. It
  is used where a feed is unlicensed, barred, or simply not needed for
  geometry: Odense (Rejseplanen barred), Daugavpils and Liepāja (no declared
  licence), Kansas City (barred), New Orleans, Tucson and Florence. Zurich,
  Göteborg and Den Haag were measured on OSM at their screens. OSM is covered
  by the site's OpenStreetMap notice. Say on the page that the lines and
  stops come from OpenStreetMap.

• Relations with no stop members are placed in NOT_DRAWN with a reason,
  never dropped silently: New Orleans's 46 and 49 (read whether they run),
  and Florence's T3.2.1, T3.2.2, T2.2 and T4 (under construction; re-check
  when T3 opens, due end of 2026).

• Colours: the feed's route_color, else OSM's colour. A CSS keyword is
  resolved through the named-colour table: New Orleans's green, red and
  blue, Göteborg's line 11 "black". Otherwise use the project's own
  palette, as Le Havre and Riga have: Odense, Daugavpils, Liepāja, Kansas
  City and Tucson. Run pipeline/linecolour.py's checks and
  check_map_markup.py. Two lines with one colour are refused (Dijon).

• Every drawn line gets a permanent on-map label (its real public name) and
  a legend entry.

• Shared code where it pays: pipeline/osm_tram.py, one module for both
  kits (agreed with the Czech kit, 2026-09-30). Seven of these ten briefs
  take OSM rail, the other three were measured on it, and five of the six
  Czech cities do too. Use it. If it does not exist yet,
  the first OSM tram build writes it, from Aarhus's stop_rows()
  generalised:

  • relations kept on ref and route, and on operator only where the
    config names one;
  • STATION_ADD by node id for a tagged stop on no route relation;
  • stop members tagged public_transport=stop_position or
    railway=tram_stop, exiting on an untagged or unnamed member and naming
    it (Daugavpils's Stropu ciemats is on the routes with no stop tag);
  • every matching relation is kept or listed in NOT_DRAWN by relation
    id with a reason, and the step exits on one that is neither (the Czech
    screens found zero-stop refs, unref'd variants and depot runs);
  • several relations per ref: a line's stops are the union of its kept
    relations, and load_osm_line_shapes draws the one with the most
    geometry;
  • scope over a union of polygons (Liberec + Jablonec), with stations
    outside named with the place they lie in, as Rennes names communes;
  • a lines-only mode that selects and validates relations for the line
    geometry without producing stations, for a feed with stations but no
    shapes.txt (Brno);
  • Aarhus's name collapse, and pipeline/stations.py's gates.

  Its control: run on Aarhus's inputs, it reproduces Aarhus's
  stations.csv. Aarhus itself is not rewired, and the later OSM cities are
  wrappers. The same goes for Latvia's step 2: Daugavpils's build lifts Riga's step 2
  (226 lines) into pipeline/countries/latvia_register.py, and Riga's
  drift_check.py is the control (Riga's outputs must not move). Shared
  modules are the app/chrome role's paths (docs/session_roles.md), so tell
  that role before writing one.

5. Coverage - full, narrowed or one bucket

The macro map's fill (coverage) follows the categories value (owner,
2026-09-30; the add-city skill on macro-legend): "All three" is full;
"Two", "Merged" and either "thin" are narrowed; "Food premises only" is
one_bucket. Food shops are food. A token third layer does not lift a
city, and that is the owner's call per city (ONE_BUCKET_BY_OWNER).

City	categories	coverage	Why
Odense	All three	full	CVR, as Aarhus
Daugavpils, Liepāja	Merged	narrowed	Riga's two layers (shops and services merged; food thin by construction)
Tucson	All three	full	City licence register with all three
Kansas City, New Orleans	All three, one thin	narrowed	City licence registers; Kansas City's food service and New Orleans's personal services are thin (categories "Food service thin", "Personal services thin"; DECISIONS 2026-09-30, A2)
Florence	All three	full	The Comune's four layers
Zurich	Two	narrowed	Food plus partial retail (alcohol-licensed shops, kiosks, petrol stations); no personal services. Boston's precedent: retail that is not only food shops makes two
Göteborg	Food premises only	one_bucket	Food service plus food shops, which count as food
Den Haag	Merged	narrowed	Rotterdam's shape (BAG winkelfunctie plus the horeca layer), which is narrowed, "Merged". Approved narrowed over the tram list's "full" (owner, call 23)

A narrowed or one-bucket page says what is missing in its own bullets
(section 6), and check_scope_disclosure.py holds it to
docs/excluded_categories.md, where the gap goes in the city's own section,
never the shared "What is missing rather than excluded" one
(docs/city_page_format.md, section 2). The layer control and
render_map_help name only the categories the map has (Riga's and Liepāja's
"two business categories (Shops and services, and Food service)").

6. Page text - approved by the owner, word for word (2026-09-30)

The page scaffold_city.py writes is already in the city-page format
(docs/city_page_format.md, section 1): render_city_title, the map, the
captions, the bullets below under bold headings, then render_map_help,
render_excluded_stations, render_country_links and render_site_notices
last. The wording is the template as the ten built pages carry it after the
2026-10-01 prose pass (app/pages/130_Odense_Heatmap.py to
139_Zurich_Heatmap.py). Braces are per-city facts, filled from this city's
own measurements (step 1, step 2 and the rendered map), never from the
brief's screen figures. A bullet in braces appears only where it applies. A
sentence from this template needs no read-back; one it does not cover is a
proposal, flagged in the session's drafts file
(docs/city_page_format.md, section 6).

The captions, directly under the map, from provenance.json: the
business source with its own date, then the rail's fetch date, for example
(Florence's and Göteborg's shape; a second source adds its own clause, as
Den Haag's BAG units do):

{Premises data} from {publisher}{ (license)}, {last updated / fetched}
{date}; the tram lines and their stops from OpenStreetMap, fetched
{date}.

The prose, as bullets under three bold headings. The first heading is
The trams, The tram for one line, The streetcar(s) for a US
streetcar.

The trams

• {N} {operator} tram lines are drawn, {lines}, each labeled on the map
  and in the legend, redrawn from OpenStreetMap's route geometry{, in colors
  this project chose, since the source records none / , in OpenStreetMap's
  own colors}.
• {Trams run about every {n} minutes by day{; less often in the
  evenings}.}
• {Buses{ and suburban trains} are not drawn{: why}.}
• {City} has no metro, so its trams are its rapid transit, as in Riga.
  Every tram stop here gets rings.
• Tram stops sit closer together than metro stations, a median of
  {spacing} m here, so the rings are drawn at half the usual size (0.05 to
  0.3 mi).
• The map covers the {city unit}. {Where a line runs past it: "{Line}
  runs on into {place}, so its {k} stops there are left out."}
• {Where a line runs past it: "The line is still drawn to its end, but those
  stops get no ring and their businesses are not counted. They are listed
  below."}

The businesses

• {The business source, in the template city's wording: which register, how
  the dots are placed, and whether a dot shows a name, a type or an address,
  one claim to a bullet.}
• {Narrowed: "This map has two categories, not three. {What the source
  holds, and what is missing.}" One bucket (Göteborg): "This map shows
  food only, not three categories. {What the source holds.}"}
• {Kansas City: "The license data dates from 15 January 2026, and holds
  licenses valid for 2025 and 2026. Businesses that opened since then are
  missing, and any that closed may still be shown."}
• About {two storefronts in five} sit within a ring.

Reading the density

• {Göteborg: "The register carries no dates. It lists the food
  businesses active on the day it was fetched ({date}), and says nothing
  about when each opened." Den Haag (owner, call 27): "The permit data
  runs to 2025. The city last edited its permit layer on 23 May 2025, so
  premises that opened or closed since then may be missing or still
  shown."}
• Read the density as a register, not a street survey. {The source's
  own caveat.}
• {New Orleans: "Streetcar stops here stand a block or two apart, so the
  rings join into a band along each line. Read them as distance from the
  line."}

Notes on the template:

• The in-ring share is said in words, as on all ten pages: "About two
  storefronts in five sit within a ring", "About one storefront in eight
  sits within a ring". All ten put it last under The businesses, and
  all ten name the third heading Reading the density (the family wording
  batch, 2026-10-03).
• A US streetcar says it the way Kansas City's, Tucson's and New Orleans's
  pages do: "{City} has no metro: its streetcar is its rapid transit, as
  Riga's trams are, so every stop gets rings", and "Streetcar stops sit
  closer together than metro stations ...". Their scope bullet adds ", as
  the US Census Bureau draws its limits. Every streetcar stop is inside it."
• The stops left out are "listed below": render_excluded_stations lists
  them on the page. Never a repository path, script or check name in page
  text.
• No ratio to OpenStreetMap: that comparison is the French pages', where
  SIRENE counts registered establishments; no tram-kit page carries one
  (owner, call C2, 2026-10-01).
• No controls paragraph or heat caveat of the city's own:
  render_map_help("{three business categories (Retail, Food service and Personal services) / two business categories (Shops and services, and Food service) / ...}") renders them, with the categories this map has.

A city whose median stop gap is over about 550 m keeps the standard rings
and drops the ring bullet. Where the rail comes from a feed rather than
OpenStreetMap, the first bullet names the feed ("from {operator}'s own
published timetable feed"), as France's template does.

Departures are owner calls, never edits. Anything that does not fit the
template goes to the owner as a proposed sentence.

7. The macro map and the app entry

• Two keys (owner, 2026-09-30, landed with the macro-legend branch):
  colour = mode, fill = coverage, as sections 1 and 5.
  scaffold_city.py --mode is required and writes mode into the
  cities.py entry; cities.py raises at import without it.
• Region: "Europe" for the seven European cities. "United States West"
  for Tucson; "United States East" for Kansas City and New Orleans, since
  Houston (−95.4°) is East. --country is spelled as app/cities.py spells
  it, and "Switzerland" is new.
• The label width is measured in a real browser for check_macro_labels.py.
  Landing is in groups at review time with a city-added deploy-verify per
  group. app/ changes reboot the app.

Before publishing (each city)

python scripts/check_personal_exposure.py <slug> (verdict in the session's
drafts file and docs/privacy_verdicts.md;
the suspects are Kansas City's dba_name, New Orleans's ownername,
Tucson's ACC_NAME on personal ownership types, Florence's beauty and
laundry points (no names, but sole traders' premises) and Den Haag's
AANVRAGER);
gate 3 ran against the operator's per-line counts with no unexplained
mismatch, or OPERATOR_COUNTS_GAP records why it could not (section 2);
check_provenance.py names the city OK; check_scope_disclosure.py passes;
a row in docs/data_sources/<country>.md for every new source, the join and
address layers included (VZD's aw_eka.csv); the page and the city's
reference-doc sections read as docs/city_page_format.md sets out (headings
that name the city, its gaps in its own section, process notes out or inside
<!-- internal --> markers, no repository path on the page);
drift_check.py <slug>; no TODO left in pipeline/<slug> or app/; a
decisions entry in the session's drafts file
(docs/decisions_drafts/<session-or-branch>.md, decisions-entry). Then
publish-city.

Heavy jobs (announce to every live session first, docs/session_roles.md):
Odense's step 2 reads the national CVR cache, and the Latvian step 2 reads
the national excise file and VZD's 141.7 MB address file. The rest are
city-sized.

The ten, one sheet each

Proposed values are the briefs' or, where a brief is not yet written, the
tram list's. The briefs are staging's. Where they differ, the brief wins, and
the call list in docs/handoff_tram_kit_2026-09-30.md collects the owner's
calls.

City	Template (built)	Business leg	Rail source	CRS	Traps
Odense	Aarhus (pipeline/aarhus/, countries/denmark*.py, taxonomies/denmark_db25.py)	CVR, kommune 461; placed on OSM's DAR points (98.4%); personally owned rows show the address, as Aarhus	OSM, 2 route=tram relations, ref L; + SDU Syd/Hospital Nord by node	25832	No colour in OSM; Hospital Syd opens 2027; build on the shared CVR cache, never refresh it from a branch
Daugavpils	Riga (pipeline/riga/, taxonomies/riga_source.py)	VID excise (food, placed on VZD's aw_eka.csv) + VZD cadastre use class 1230 (shops and services), ATVK 0002000	OSM, routes 1-5, all drawn (owner, call 28): 38 stop names; the page states the waits (route 1 every 10-15 min, the Stropu loop 20-30, routes 2 and 4 about hourly)	32635	aw_eka.csv's KOORD_X is the northing (EPSG:3059); use DD_N/DD_E; STATUSS = EKS; the VZD credit (notice 42 extended); hourly routes (section 1)
Liepāja	Riga	as Daugavpils, ATVK 0005000	OSM, ref 1; + Brīvības iela and Klaipēdas iela by node	32634	as Daugavpils
Kansas City	Houston (a US register, the sole-owner name rule); naics.py through a NAICS 2022 title map	Socrata kkhs-93m4, Public Domain, frozen 2026-01-15; valid_license_for 2025 and 2026 only; fee-code types ("Misc Rate 129") dropped and counted	OSM (RideKC's GTFS barred, owner), ref 601, 19 stops, all inside; 413 m	32615	The data date on the page; dba_name withheld where it is a person; no logos; frequency as a fact, not a cited schedule
New Orleans	Philadelphia (a US text taxonomy, premises-taxonomy)	Socrata iqay-p646, CC0, daily; drop "Special Events-Other (Vendor)" and "Home Based-Office Use Only"; ownername never shown	OSM, streetcars 12, 47, 48 and 2, 110 stops, all inside; 164 m	32615	Every stop kept (owner, call 13); 46 and 49 have no stops (NOT_DRAWN unless they run); CSS-keyword colours
Tucson	Houston (the sole-owner name rule); naics.py unchanged	BUSLIC layer 3 (never layer 1), active (strip LIC_STATUS) and HOME_OCCUPATION = F; licence SILENT, permissive reading (owner, 2026-09-30)	OSM, Sun Link, 21 stops, all inside; 265 m	32612	Notice "Business licence data: City of Tucson"; never call the pins complete; ACC_NAME withheld for Sole Proprietorship, Individual and Married; 10 min weekdays 07-18, 20 otherwise, disclosed
Florence	Milan and Rome (the Comune's layers; Milan's fuori piano)	Four Comune GeoJSON layers (datigis.comune.fi.it/json/), EPSG:3003, no names or addresses (the dots show the type); licence read: CC BY 4.0, credit the Comune di Firenze and state the changes	OSM, T1 and T2 in OSM colours, 39 stops in the comune (T1 20 of 24); 322 m	32632	639 exempt food rows kept, non-public ones filtered (owner, call 16); Scandicci's 4 stops outside; 3,424 rows share a point; T3 under construction (NOT_DRAWN); no giglio or logo
Zurich	Stockholm (one register, city only), Seoul/Gyeonggi (partial retail)	Gastwirtschaftsbetriebe via the WFS (the CKAN downloads are an Angular shell), layer gastwirtschaftsbetriebe, CC0, 3,487 rows all Offen and 2026: food 2,325 + partial retail 1,028 (Kleinverkaufsstelle, Kiosk, Tankstelle)	OSM: trams 2-11, 13-15, 17, 182 stops; 283 m. Out as stubs: Forchbahn S18 (20%), Glattalbahn 12 (11%), Limmattalbahn 20 (15%); 50 and 51 only if they meet the 20-minute floor	2056	add-country first; the S-Bahn and Forchbahn named as excluded
Göteborg	Stockholm (taxonomies/sweden_livsmedel.py)	Livsmedelsverksamheter, CC0, daily: take the CSV (utf-8-sig, ;), never the rowstore JSON; 5,066 rows: food service 2,167, food shops 906; 274 blank typ classified by name (about 74 kept) or dropped	OSM: trams 1-13, 127 stops; 378 m; 4 (75%) and 12 (72%) drawn to their ends with five Mölndal stops outside; Lisebergslinjen out	UTM 32N (11.97° E; not Stockholm's 34N)	Undated rows (the page says so); line 1's OSM colour is white (check contrast; Lille's darkening if it fails); take the hex where a relation has two colour values
Den Haag	Rotterdam (BAG, taxonomies/rotterdam_source.py); Amsterdam for the horeca precedent	BAG winkelfunctie units (6,560) + Horeca_nieuw layer 2 (2,543 granted or notified, about 2,352 food after exclusions; edited 2025-05-23)	OSM: 14 HTM lines, 161 stops; 335 m; RandstadRail E out; tram 1 keeps 19 of 37 (51%)	UTM 31N, as Rotterdam	Never fetch AANVRAGER, KVKNUMMER or RECHTSVORM; the trade name is OMSCHRIJVI, double-encoded UTF-8 to repair; 1 and 19 share #c01115 (one moves); 10 and 34 need colours; the layer's date on the page