Kingston data projectIs the Polson Park rink open? An agent that has to say “as of when”
The City of Kingston publishes its public outdoor rinks as open data: seventeen rinks, each with hours, a status and a few yes or no fields. We built a tiny agent around it and a check a person wrote first. Code, output and counts are ours; the file was downloaded on 23 October 2026.
What the City's file holds (our tally of 17 rows, downloaded 23 October 2026)| Field | Values in the file |
|---|
| Status | Closed on all 17 rows |
| Type | City Operated 13, Community Rink Agreement 4 |
| Hours of Operation | 7:00 am - 9:00 pm on 16 rows, one rink with later hours |
| Washroom | Yes at 6 rinks |
| Staffed | Yes at 7 rinks |
| Capacity | Only the numbers 0, 5, 10 and 25, with no unit or definition anywhere in the dataset description |
| Layer last edited | 21 March 2026, according to the City's map service |
Every rink reads Closed. On a late October download that is no surprise, and the City's own description says that the existence of the rinks is weather dependent and that unstaffed rinks may be maintained by volunteers under a community agreement. But an agent asked “is Polson Park open tonight?” in January would still read Closed from a file last edited in March, and answer with total confidence.
So the tool returns the edit date with every row, and the check refuses answers that leave it out:# Kingston rink agent: every answer must say how old its data is (our classroom example)
import csv, json
from datetime import datetime, timezone
RINKS = list(csv.DictReader(open("kingston-rinks.csv", encoding="utf-8-sig")))
EDITED_MS = json.load(open("layer0.json", encoding="utf-8"))["editingInfo"]["dataLastEditDate"]
AS_OF = datetime.fromtimestamp(EDITED_MS / 1000, timezone.utc).strftime("%d %B %Y")
def rink_status(name):
"""Tool: one rink's row, plus the date the City last edited the layer."""
for r in RINKS:
if r["Name"].lower().startswith(name.lower()):
return {"rink": r["Name"], "status": r["Status"], "hours": r["Hours of Operation"],
"washroom": r["Washroom"], "staffed": r["Staffed"], "as_of": AS_OF}
return {"error": "no rink starting with " + name}
def check(answer, result):
"""Human-written check: a status claim needs the as-of date and must match the tool."""
problems = []
if result["status"].lower() not in answer.lower():
problems.append("status differs from the tool")
if result["as_of"] not in answer:
problems.append("no as-of date")
if "capacity" in answer.lower():
problems.append("uses Capacity, which the file never defines")
return "OK" if not problems else "REJECTED: " + "; ".join(problems)
result = rink_status("Polson Park")
print("tool ->", result)
# Two stand-in answers in place of a model, so the check can be tested offline.
honest = "The City's layer, last edited " + AS_OF + ", lists Polson Park Outdoor Rink as Closed."
hasty = "Polson Park rink is Closed tonight and holds 10 skaters at capacity."
for a in (honest, hasty):
print(check(a, result), "|", a)
What it printed on our machine:tool -> {'rink': 'Polson Park Outdoor Rink', 'status': 'Closed', 'hours': '7:00 am - 9:00 pm', 'washroom': 'Yes', 'staffed': 'Yes', 'as_of': '21 March 2026'}
OK | The City's layer, last edited 21 March 2026, lists Polson Park Outdoor Rink as Closed.
REJECTED: no as-of date; uses Capacity, which the file never defines | Polson Park rink is Closed tonight and holds 10 skaters at capacity.
The date travels with the fact
Every tool result carries the layer's last-edit date. An answer that drops it is rejected, because a status without a date is a guess dressed as a fact, especially for something as changeable as ice.
Undefined means untouchable
Capacity holds 0, 5, 10 and 25, and nothing in the City's description says what they count. The hasty answer turned 10 into skaters. Our check rejects any answer that uses the field until someone finds a definition.
Stand-in answers, on purpose
The two answers are written by us in place of a model, one honest and one hasty, so the check can be tested on any laptop with no AI service. In the adult course the stand-ins are replaced by real model calls and the check stays.
What a better agent would add
Two OBJECTIDs, 2 and 3, are missing from the sequence; the rows may have been retired, and the file does not say. A careful agent would also point users to the City's own rink page for live conditions, rather than treating a spring snapshot as tonight's news.
Used under the City of Kingston Open Data Licence 1.0, which the City adapted from version 1.0 of the Open Government Licence for Ontario; its grant covers commercial purposes as well (licence, dataset). Contains information licensed under the Open Data Licence - City of Kingston. The City does not endorse this page or our classes.
