Fire Department

What this source is

The city produced this data under public-records request 26-8034 as a printed report from the Fire Department’s records system — 18,259 pages of paginated rows, one row per incident, covering January 2015 to July 2026. It is not a data export: every field is recovered from the printed page.

The report’s final page prints its own totals: 674,976 incidents, of which 630,155 are marked Completed and 62 Reviewed. This project’s parse recovers 674,976 records, the printed figure exactly. The counts on this page are the parse the report’s own totals page verifies.

93.4% of records are marked Completed and 0.01% Reviewed. The report defines neither flag anywhere, so they are reproduced as printed and no meaning is claimed for them.

The rows are ordered by incident number rather than by date, and carry no station, unit, district or apparatus column, so there is no geography below citywide and no per-unit figure to extract. 220 distinct incident-type labels appear across the document.

The counts on this page are records as the department produced them, and the department responds outside the city: a small number of rows — measured at 0.12% from the geocoded address distribution — are mutual-aid incidents on Emeryville and Alameda streets. The log carries no city field, so they cannot be separated out.

The document itself is not published on this site: every row carries a full street address, apartment numbers included, and most rows are medical calls. Only the counts on this page are.

Coverage defect in the incident log

April 2022 carries 2,749 records across 30 of its 30 days; June 2022 carries 2,931 records across 30 of its 30 days; July 2022 carries 3,067 records across 31 of its 31 days; August 2022 carries 3,069 records across 31 of its 31 days; September 2022 carries 3,090 records across 30 of its 30 days; October 2022 carries 2,994 records across 31 of its 31 days; November 2022 carries 2,988 records across 30 of its 30 days; December 2022 carries 3,299 records across 31 of its 31 days; January 2023 carries 3,365 records across 31 of its 31 days; February 2023 carries 2,163 records across 22 of its 28 days. May 2022 carries 8,688 records across 31 of its 31 days, of which 393 are dated to the single day of May 18, 2022.

2022 records 47,536 records against 57,570 in 2020 and 58,537 in 2021. Within 2022, the nine affected periods carry 32,875 records, where the group's other 3 periods average 4,887 each; nine periods at that rate would be 43,983, 11,108 more than the dataset holds — about 2.3 periods of records.

2023 records 56,517 records against 58,537 in 2021 and 61,385 in 2024. Within 2023, the two affected periods carry 5,528 records, where the group's other 10 periods average 5,099 each; two periods at that rate would be 10,198, 4,670 more than the dataset holds — about 0.9 periods of records.

Classification moves in the same window, so the affected months are not simply short of records: EMS / rescue is 66.0% of records before the window and 61.1% inside it; Fire is 4.4% of records before the window and 8.8% inside it; Service call is 3.1% of records before the window and 7.2% inside it. Category counts either side of the window are therefore not comparable, over and above the missing records.

Consequences for the figures on this page: 2022, 2023 are affected, and the affected months are left out of the charts above rather than drawn as real counts. The document carries no field marking a record as re-dated or re-entered, and no city document in this site’s corpus has been found that accounts for the window, so the cause is not established here.

Duplicate rows, counted and kept

6,245 rows — 0.9% of the log — agree with at least one other row on every field the document carries: incident number, exposure, date, time to the second, type code, description and address. No field in the report distinguishes them, so they cannot be removed on any stated basis, and the report’s own printed total counts them. Every count on this page is therefore a count of records as printed. 2,707 of them are printed immediately below the row they repeat. Collapsing every such group to a single row would remove 3,650 rows, so every count on this page is high by at most 3,650 — 0.5% — against a count of distinct incidents.

The records system changes mid-year

The department’s incident numbering changes from a zero-padded sequence to a year-prefixed ten-digit id, and it does not change on a year boundary: 2024 carries 35,043 records numbered in the earlier scheme and 26,342 in the later one. The later scheme first appears in 2024. The two schemes bring two incident-type coding schemes with them — 220 distinct type labels appear across the document, some of them the earlier era’s three-character NFIRS codes with local letter suffixes and some the later era’s two-digit NFIRS subcategory rollups — so no code-level series is comparable across the changeover, and none is published here. The categories charted above are built from the NFIRS first digit, which both schemes share.

The department’s published incident count does not match this log

The Fire Department’s budget narrative states that its Field Operations Bureau “handles nearly 80,000 emergency incidents a year, which includes fire, rescue, medical aid and other calls for service” (budget attachment, p. 459). This log holds 60,569 records dated in 2025. An earlier budget narrative used the same sentence with “over 60,000” (budget attachment, p. 211), which the log’s figures do fit.

The two numbers count different things, but which things is not established: a plausible reading is that the department’s figure counts unit responses, several of which can be dispatched to one incident, while this log has one row per incident and no unit column at all to test that with. The discrepancy is stated rather than resolved, and the log’s figure is the one charted on this page because it is the one this site can show a document for.

What this data cannot answer: response times

Response times are NOT derivable from this document. The report carries exactly one timestamp per incident (a single Date/Time column, the alarm or call time) and no dispatch, en-route, arrival, or clear time. There is no second time field anywhere in the 18,259 pages, no unit/apparatus column, and no per-incident duration. Response time is by definition the interval between two of those missing timestamps, so no arithmetic on this log can produce it, and no aggregation can recover it. If response times are the goal, the native-format export must still be pursued: specifically the NFIRS/CAD incident table with alarm, dispatch, arrival and clear datetimes, and the apparatus-level response table.

How that was established, so it can be re-checked rather than taken on trust — full-document column scan: 0 words outside the 7 known columns; 900-page random sample: 33,272/33,272 data rows with exactly 1 time token, 0 rows with more than 1; 1 distinct header band document-wide.

What “adopted”, “proposed”, and “actual” mean

Proposed = the spending plan the Mayor submits to the City Council. Adopted = the appropriation the City Council approves when it passes the budget — an authorization to spend, set before the fiscal year begins, not a record of what was spent. Actual = spending as reported after the fact in audited financial statements (ACFRs). Oakland budgets in two-year cycles; fiscal years here are labeled by their end year (FY2022–23 → 2023). Amounts are department totals across all funds unless noted.