Audits
Were 4 in 10 unsheltered King County homeless in 2024 last housed elsewhere?
Version 5. FactHub's auditors are automated reviewers, separate from the process that writes articles. Each checks a fixed, published list of things and grades the results with its own published rule. They can be wrong; an audit is not a human expert review. What this means · All auditors
Across versions
| Version | First Look | Skeptic |
|---|---|---|
| Version 5 | 7/7 | 6/7 |
First Look
A: Passes the first look: nothing obviously wrong. Minor notes may be reported as feedback without lowering the grade.
- Title matches the evidence
- No obvious contradictions
- Plausible justification
- Sources are relevant
- Common questions answered
- Headline and verdict agree
- No Wikipedia or open wikis
7 checks
-
Pass Title matches the evidence
The title 'Were 4 in 10 unsheltered King County homeless in 2024 last housed elsewhere?' rounds the 39% (323 of 821) in ev-2024, and the title's scope (unsheltered, King County, 2024) matches that evidence. The conclusion and limitations disclose that only 56% answered and that the true share could be 22% to 66%, and the Mostly true rating reflects this. The title's number is the evidence's figure, not one the evidence lacks.
-
Pass No obvious contradictions
Key points 1 to 3 and 6 are established and agree with the conclusion. Key point 4 (22% to 66%) matches the conclusion's bounds. Key point 5 (16% to 23%, unclear growth) matches the conclusion. Internal arithmetic checks out: 163+150+10 = 323, 323/821 = 39.3%, 323/1466 = 22%, (323+645)/1466 = 66%, and the Table 13 categories sum to 821. The 2019 tenure shares (31%, 15%, 19% of 752) and the 30% to 41% San Francisco change also agree with the sources quoted.
-
Pass Plausible justification
The conclusion explains Mostly true: 39% of the 821 who answered were housed outside the county, 44% did not answer, answers are unweighted, and the claim covers all unsheltered people. A reasonable reader would accept that these gaps justify Mostly true rather than True.
-
Pass Sources are relevant
Every source is plausibly on topic. The KCRHA 2024 and 2026 reports, the Count Us In reports for 2017 to 2019, the San Francisco, Portland, Los Angeles, Alameda, Sacramento and San Diego homeless counts, the UCSF California study, the FactHub dataset records and the Census ACS table all match the claims they support.
-
Pass Common questions answered
There are 11 objections, each with a response that addresses its question: the regional comparison, the reasons for moving, the 2019 rise, local versus newcomer share, the census baseline, arrival while homeless, shelter inclusion, nearby counties, institutional releases, sampling, and newer data.
-
Pass Headline and verdict agree
Step a: Mostly true gives Mostly yes. Step b: 'Were 4 in 10 unsheltered King County homeless in 2024 last housed elsewhere? Mostly yes.' Step c: the reader is told that about 4 in 10 were last housed outside King County, and that this is mostly right. Step d: this matches the claim ('About 4 in 10 ... were last housed outside the county'), the Mostly true finding and the answer (39% of those who answered). The claim is positive with no negation, so there is no polarity problem.
-
Pass No Wikipedia or open wikis
None of the 27 cited sources is Wikipedia or another open wiki. Their publishers are government agencies, universities, the UCSF study, the U.S. Census Bureau and FactHub records. The Wikimedia Commons credit applies only to the article's image, which is not a cited source. The evidence institutions named (KCRHA, All Home, Applied Survey Research, Portland State University, Multnomah County, LAHSA, UCSF, Alameda County, Sacramento Steps Forward, the Regional Task Force on Homelessness) are not open wikis.
Coverage: I read the full article payload: title, claim, finding, answer, conclusion, summary, key points, context, evidence and limitations, the 11 objections with responses, method, editorial note, chart captions and the sources with their takeaway statements and quotes. I did not open any source, snapshot or URL. Source takeaways for KCRHA 2024, the Count Us In reports, San Francisco, Portland, LAHSA, UCSF, Alameda, Sacramento, San Diego, KCRHA 2026 and the ACS are shown as 'not yet checked'. The FactHub dataset takeaways carry no status. Among key points, only 1, 2, 3 and 6 are marked established; key point 4 is context and key point 5 is notShown. I did not treat any unchecked takeaway as established. Checks 1 to 5 rely on the article's own text and internal arithmetic: 323/821, the 22% and 66% bounds from 645 non-answers, the Table 13 and ACS category sums, the 2019 tenure and residence shares, the 2017 to 2019 trend, and the San Francisco 30% to 41% change. Not verified against sources: the 24% 2019 unsheltered bound (FactHub dataset), the 866 residence total behind the 2017 5.8% figure, the Los Angeles 64% figure against the source's 'two-thirds' wording, and the Multnomah 42% share. These are source-accuracy questions for the Source Checker. Check 7 is based on source names, publishers, URLs and the institutions named in evidence items.
Skeptic
A: The rating and confidence follow from the evidence, the article rates the real claim, and the strongest objections are answered. A single minor note for the author does not lower the grade.
- Lead questions hold up
- Rates the real claim
- Rating follows from the evidence
- Confidence matches the evidence
- Strongest objections answered
- Evidence is independent
- Rating states its terms
7 checks · 1 concern
-
Concern (minor) Lead questions hold up
Add the years to objection obj-why, e.g. "In 2017 and 2018, did at least 1 in 5 homeless people from outside King County say they came for homeless services or benefits?"
obj-west, "Is King County's share higher than Portland's, San Francisco's and Los Angeles's?" answered "No, about the same": - Surprising: plausibly so for readers who assume Seattle is a magnet. - Not the headline: neither the title nor the answer says it. - Settled as worded: the figures (36% to 42%) settle it. - Neutral: the response also says King County's share is double that of Alameda, Sacramento and San Diego. - Stands alone: "No, about the same" works on its own. It passes all five tests. obj-why, "Did at least 1 in 5 homeless newcomers say they came for homeless services or benefits?" answered "Yes, 20–25%": - Settled as worded: the threshold and the survey's answer wording are settled by the 2017 and 2018 tables. - Neutral: the response adds the 3% to 6% share of all respondents, and the lead set as a whole is balanced, since obj-west answers against the magnet reading. - Stands alone: "Yes." works on its own. The remaining issue is that the question names no year. It sits directly under a 2024 verdict, so a reader seeing it cold will take "homeless newcomers" to mean the 2024 group last housed elsewhere. The answer actually comes from 2017-2018 respondents who were living outside the county when they became homeless, a different group. The response clarifies this, but the question line itself invites the misreading. This is a careful-editor fix, not a reversal.
About objection:obj-why The finding
-
Pass Rates the real claim
The claim is "About 4 in 10 of King County's unsheltered homeless people were last housed outside the county." It circulates as a statistic about the county's unsheltered population. The article rates that population-level version and says so in method:1: "We rated the claim as people repeat it: the share of all King County's unsheltered homeless people." The claim's wording matches the survey question ("the last time you had stable housing"), so this is not a strawman. The article also does not stretch the claim into "came here homeless" or "came for services"; it treats those stronger readings as separate questions (obj-arrived, obj-why, editorial note). The conclusion stays within the 2024 unsheltered scope. Where it draws on 2017-2019 sheltered data or other cities, it labels them as context.
-
Pass Rating follows from the evidence
Before reading the finding, here is how I weighed the evidence. The best measure, ev-2024, gives 323 of 821 respondents (39.3%), which matches "about 4 in 10." Only 56% answered, the counts are unweighted, and the sample came from referral chains. Bounding the missing answers gives anywhere from 22% to 66%, with no stated sign of bias in either direction. That points to a claim whose best estimate matches but cannot be confirmed for the whole population. I would rate it Mostly true, with True as a defensible alternative. The article's "Mostly true" and its answer ("39% ... but only 56% answered, so the true share is uncertain") match that. The other lines of evidence do not push the label either way: the 2017-2019 trend, the West Coast comparisons and the reasons for moving are all context for the figure.
-
Pass Confidence matches the evidence
Moderate confidence fits a single unweighted survey with 44% missing answers. The title is posed as a question. The summary and key point 4 state the missing answers and the 22% to 66% range. limitation:ev-2024 covers the main limits: - the missing answers; - no weighting; - no date of last stable housing; - unsheltered people only; - possible tilt from hub-site sampling. The conclusion repeats them and adds returnees and people released from institutions. I found no unstated limitation that would change a reader's belief.
-
Pass Strongest objections answered
These are the strongest objections I can name, and where the article answers each: - Missing answers and no weighting: answered with explicit bounds in obj-rise and the conclusion. - Referral-chain sampling bias: obj-sampling argues both directions and admits it can't be checked. - "Last housed elsewhere" does not mean someone arrived homeless or came for services: obj-arrived and obj-why, with the 3% to 6% of all respondents. - Returnees and people released from institutions in other counties: obj-arrived and obj-institution. - Sheltered people are excluded: obj-shelter. - A comparable baseline for housed residents: obj-baseline, which uses census figures and says honestly that it can't tell. - Whether the share rose since 2019: obj-rise, with a bound on the change of surveyed group. Each is answered with evidence or bounded arithmetic rather than assertion. A secondary point I would add, but whose absence is not a finding, is whether people living in vehicles differ from those sleeping outside.
-
Pass Evidence is independent
The rating rests on one source, ev-2024 (the KCRHA 2024 survey), and the article says so plainly in method:1 and method:7. It does not present other items as independent confirmation of the 39%. There is some overlap: - ev-trend and ev-why both draw on the Count Us In reports. - ev-trend repeats the 2024 bounds derived from ev-2024. - The San Francisco figures appear in both ev-trend and ev-west. Each item, though, is used for a different question: trend, reasons for moving, and how King County compares with other cities. ev-west draws on separate agencies. No conclusion is inflated by counting the same dataset twice.
-
Pass Rating states its terms
(a) The answer states what holds the claim back: "only 56% answered, so the true share is uncertain." It gives the reason for Mostly true, not only the supporting figure. (b) The claim's key term, "last housed outside the county," matches the survey's own question, so the rating does not depend on choosing a definition. For readers who mean the older "where they became homeless" measure, key point 5 gives the 16% to 23% result. For readers who mean the whole surveyed group, key point 4 gives the 22% to 66% range.
Coverage: I examined the full article payload: the claim, finding, answer, confidence, summary, title, key points, context, conclusion, all four evidence items with their limitations, all 11 objections including lead objections obj-west and obj-why, the method steps, the editorial note, and the source and reliability notes. I did not read source snapshots. Every takeaway in the sources block is marked "not yet checked," none "established." I therefore treated all figures (39.3%, 821/1,466, the 22% and 66% bounds, the 2017-2019 shares, the reasons-for-moving shares, the West Coast and census figures) as stated by the article and did not verify them. Whether they match their sources is left to the Source Checker. Where I named objections, I did so from general knowledge only and phrased them as things the article could address.