CA-21 · House

The district, in brief

California's 21st is recorded in the facts as redrawn for 2026, on a map first used this cycle. The text says nothing about the new lines. The 2024 presidential baseline on file, a two-party margin of 3.9 for the Democrat, is on the previous lines, not the current ones; the 2024 House result on those lines was a two-party margin of 5.15 for the Democrat.

The incumbent is Democrat Jim Costa, re-elected with 52.6% of the vote in 2024. He advanced from the primary with 42.23% alongside Republican Kyle Kirkland, a casino owner who ran for the 20th district in 2024, who took 26.31%. Eliminated were Republican Lorenzo Rios, CEO of the Clovis Veterans Memorial District (14.93%), Democrats Lourin Hubbard, a labor organizer (10.28%), and Eric Garcia, a therapist (5.08%), and independent Lance Kruse (1.18%). As of May 13, 2026, Costa had raised $1,161,499 and held $818,142. Kirkland's figures differ between the two tables carrying that date: $311,664 raised with $52,982 on hand in one, $723,332 raised and spent with $0 on hand in the other. An endorsement list headed by Senators Padilla and Schiff, Governor Newsom, Lieutenant Governor Kounalakis and Attorney General Bonta, with the California Democratic Party, appears without a named candidate.

Every rater listed calls the seat D: Decision Desk HQ (July 14, 2026), the Cook Political Report (November 5, 2025), Inside Elections (August 20, 2026), Sabato's Crystal Ball (November 5, 2025) and The Economist (May 6, 2026). No district polls appear in the text.

A paragraph quoting a model forecast from the day this brief was written is omitted; the current forecast is at the top of this page.

The facts, each beside its sentence

Not in the source: The facts mark the district as redrawn for 2026, but the text describes nothing about the new lines; the presidential baseline is on the old lines.; The two fundraising tables, both dated May 13, 2026, give different figures for Kirkland.; Endorsement lists appear without the name of the candidate they endorse; the post-primary endorsements section is empty.; No district polls in the text; the model's basis is prior only, with no district polls.; Ratings are given only as 'D', with no Lean/Likely/Safe gradation.; No presidential vote percentages for the district appear in the text.

Read from the district's section of Wikipedia's state page (CC BY-SA), archived as 5cec783f78c8; written by claude-fable-5-1 (batch reader) on 2026-09-30 and checked mechanically: every quote occurs verbatim in the archived text, every number in the brief occurs in the inputs. No probability here is the reader's.

Who holds the seat now

Jim Costa (Democrat), in this chamber since 2005, 11 terms. Committees: House Committee on Agriculture; House Committee on Foreign Affairs.

From the congress-legislators dataset (CC0), archived as ae8cb9477c48. For a Senate race this lists both of the state's senators; the one whose term ends in the election year holds the seat on the ballot. For a House district redrawn for 2026 it is the member elected under the old lines with this number, who may be running elsewhere; the dossier above names who is running here.

How it has voted for president

A shift of 11.1 points toward the Republicans from 2016 to 2024. county returns placed on the current lines by 2020 block population; MEDSL county file (2016), an MIT-licensed county file (2020, checked against certified state results) and MEDSL precinct returns (2024). For a district split across counties this is an estimate, and where the state published the exact 2024 figure for these lines, that exact figure is the one the model uses.

Who are the voters?

Of about 459,505 adults 25 and over. In 2020 the district voted 60.4% Democratic (two-party) for president. With the same preferences and a 2022-style midterm turnout, it would vote 60.1% Democratic: the midterm electorate alone moves the margin 0.6 points toward Republicans. Group shares are estimated from how results vary across the state's census tracts (ecological inference, anchored to Pew's 2020 validated-voter figures) and calibrated to this district's actual 2020 vote; turnout is the Census survey's state rate by race and national rate by education, whose state samples carry margins of several points.

Not answered here: how each group will vote in 2026. These are 2020 preferences; this profile is description, not yet an input to the forecast.

The money

FEC bulk candidate summaries through June 30, 2026; $4,961 of independent expenditures for or against these candidates this cycle (FEC Schedule E, types 24E and 24A), which is the parties' and outside groups' own revealed view of whether this seat is in play.

The district in numbers

ACS 5-year 2023, assembled from census tracts by the 2020 population of their blocks inside the current lines, because the Bureau has not tabulated these lines; medians are population-weighted means of tract medians. Unemployment is the BLS county series placed on the district by block population, not seasonally adjusted, so changes are same-month comparisons. Income, transfers and jobs are BEA county accounts placed the same way, compared with the average district. Every other figure is a named column of a named table in the Bureau's summary file, archived before it was read. This is what a reader knows about a district before looking at a poll; the model does not yet use it, and whether it should is a registered question.

Who is favoured, and by how much?

The 90% interval runs from +1.5 to +31.5 points, D-R. It is wide because this district has never been polled.

What is this built on?

A model trained on every contested House race 2012-2024 predicts this district's lean from its own record: presidential and House results on the current lines, incumbency, both candidates' prior offices, outside and party money before 1 October, demographics and public-assistance shares from the Census release before the election, local unemployment and income, and religious tradition. The national margin comes from presidential approval, the economy, the midterm penalty and the generic ballot corrected for its measured lean. Draws share national and state errors, so the seat total and this district come from the same simulation.

Not answered here: district polls, which are not yet folded in.

Why does the model say this?

Each figure is how far this district's predicted lean moves when that one feature is set to the typical value across 2026 districts, everything else held: PAC money balance: 6.2 pts toward the Democrat; 2024 presidential lean (D+7.7): 5.9 pts toward the Democrat; 2024 House lean: 4.6 pts toward the Democrat; moved from other state share (0.4%): 2.9 pts toward the Republican; experience gap (+4 levels, D minus R): 0.9 pts toward the Democrat.

How sure is it, and of what?

Total sd 9.08 points, made of the seat's own volatility 90%, correlated polling error 10%. The decomposition is stored with the forecast rather than reconstructed, so a probability whose dominant input is a placeholder is visibly that.

Not answered here: a nowcast. 'If ballots were cast today' needs something that observes today, and for a House district nothing does. Publishing the election-day figure under both labels would make one forecast look like two.

How was this number made?

Run 6f211e6f-2bdc-4dda-8f72-81fc628dab0b, house_district_prior/0.2.0, as of 2026-10-02. Promoted by a pre-registered gate: on 2016-2022 its Brier score was 0.036 against 0.049 for the previous method, and on a single 2024 run 0.023 against 0.032 (experiments/district_gate_result.md, with an independent review).