Coding and AI classes in Trier

import json

LIMIT = 200

def page(offset):                         # one saved answer from the Trier service
    with open(f"stops_offset_{offset}.json", encoding="utf-8") as f:
        return json.load(f)

first = page(0)
print("The service says it matched:", first["numberMatched"])

stops, offset = [], 0
while True:
    features = page(offset)["features"]
    stops += features
    if len(features) < LIMIT:             # "a short page must be the last one"
        break
    offset += LIMIT
print("Stopping at the first short page:", len(stops), "stops")

The third page brought 198 stops instead of 200, so Version 1 decided it had reached the end. It had not. The service simply sent a slightly short page in the middle, and 62 more stops were waiting on the fourth page. A program that trusts "short means last" would hand an AI model, a map or a report about nine percent too few stops, and nobody would see an error message.

stops, offset = [], 0
while True:
    features = page(offset)["features"]
    if not features:                      # only an empty page means the end
        break
    print("  offset", offset, "gave", len(features))
    stops += features
    offset += LIMIT

ids = [s["properties"]["gid"] for s in stops]
print("Stopping at the first empty page:", len(stops), "stops,", len(set(ids)), "different ids")
twice = sorted({i for i in ids if ids.count(i) > 1})
print("Ids that came twice:", twice)

Data: Stadt Trier, Stadtbus - Haltestellen (SWT), via the geoportal.rlp.de feature service, Datenlizenz Deutschland Namensnennung 2.0 (dl-de/by2-0, www.trier.de); pages saved on 3 November 2026. Every count in the table, and the reading of it, is ours.

Version 2 keeps going until a page is empty, which is the only safe sign of the end. It then finds a second, smaller problem: one stop, id 35006, appears on two neighbouring pages, so 660 records hold 659 different stops. Paging can repeat or skip items at page edges, which is why a careful collector always de-duplicates by id. The final surprise is the service's own total. It says 648 on every page, yet it delivers 659 different stops, and a single large request agrees with 659. A number a system reports about itself is a claim, not a fact, until you count.

There are also more stop records than stop names. The file lists 347 different names for its 659 records, because a busy stop is listed more than once: Trier, Porta Nigra alone appears eight times, most likely once per platform or direction. Deciding whether "a stop" means a name or a platform comes before any analysis, and the right answer depends on the question.

1. Children, 9 to 12. Read a long book list aloud ten titles at a time. Make one "page" only nine titles long by mistake. Does the listener stop early?
2. Teenagers. After running the two versions, add a third that halts when the service gives no next link at all. Does the service ever stop giving one?
3. Adults. Think of an export at work that comes in pages or batches. How would you prove you received everything exactly once?
4. Going further. Group the 659 records by stop name and list the five names with the most platforms.