Have overdose deaths outgrown King County's subsidized housing?

1. Pass Title matches the evidence

   The title asks whether overdose deaths outgrew subsidized housing. The evidence shows overdose deaths up 68% (114 to 192) against beds up 25%, and per-bed deaths up 25% even with rapid rehousing counted. The Mostly true rating matches, and the recording-change caveat is stated in the conclusion, the answer and the ev-recoding item. No numbers appear in the title.

2. Pass No obvious contradictions

   I recomputed the main figures and they agree with each other: 68% and 25% growth, 14.5 to 19.6 per 1,000 beds (35%), 12.6 to 15.8 with rapid rehousing, 21% to 29% share, 19% recoding-adjusted rise, 39% unknown-adjusted rise, 50 of 56 shifted deaths, 309 to 379 (23%), and 38% with the tiny-house and hotel units added. The peak at 281 in 2023 and the fall to 192 agree with the yearly series. Key point 3 states the recording question as an open point, which the conclusion also treats as unresolved.

3. Pass Plausible justification

   The conclusion explains why the rating is Mostly true and not True or Mixed. It cites the untestable 2021 to 2023 years, the 19% recoding-adjusted rise still being below the beds' 25% only if about 50 of 56 shifted deaths were recording changes, and that confidence is moderate. This reasoning follows from the evidence presented.

4. Pass Sources are relevant

   Cited sources are the FactHub dataset analyses on the same deaths and beds, the county's overdose review and methods report, the homeless authority's site database, the county's homeless-death report, the Medical Examiner's notes, the Denver Urban Institute evaluation, and the Seattle city auditor report on overdose concentrations. Each is on topic for the point it supports.

5. Pass Common questions answered

   There are eight common questions, each with a response that addresses it: the all-causes comparison, recording change, 2019 baseline, unknown living situation, rapid rehousing, housing category mix, the still-climbing question, and causation. Each response gives a figure or method point that answers its question.

6. Pass Headline and verdict agree

   Step b sentence: 'Have overdose deaths outgrown King County's subsidized housing? Mostly yes.' This tells readers that overdose deaths have mostly outgrown the housing. The claim says overdose deaths have risen faster than the housing since 2021, the finding is Mostly true, and the answer says the same. Neither the title nor the claim uses a negation, so a Yes to the title means the claim is true.

Coverage: I read the article text, key points, evidence items, common questions, method steps, conclusion, changelog and source takeaways, and I did not read the source snapshots or the web. Every cited takeaway in the sources block is marked 'not yet checked', and none is marked established, so I treated all cited figures as unchecked. My judgments rely on the article's internal consistency and on recomputing its arithmetic from the figures it prints. Key points are labelled established by the article itself, but the source takeaways behind them are not. This does not change any result here, but source accuracy is unverified. I did not evaluate the related nodes or the image beyond the caption. One wording point for the source checker, not counted as a finding: the article and the evidence item say the Denver trial showed 'no significant difference', while the quoted takeaway says deaths were 'about the same'.

1. Concern (minor) Strongest objections answered

   Address in obj-recoding or the conclusion whether visitors who die at housing sites being recorded as residents could account for part of the rise, not only as a limitation.

   The strongest objection is that the rise is an artifact of how living situation is recorded. obj-recoding faces it directly, with a worst-case bound and an honest 'can't be ruled out'. The article also answers the other main alternatives with figures: the drop in unknown living situations (obj-unknown), rapid rehousing beds (obj-rrh), shelter units mixed into the group (obj-category), the start year (obj-baseline), the 2023 peak (obj-peak), causation (obj-cause, Denver trial), and the mix of who moved in (conclusion, obj-baseline). One alternative is raised but never weighed. The limitation on ev-recoding says 'The county also doesn't say how often a visitor who dies at a housing site is recorded as a resident.' A guest who overdoses in a unit is a plausible source of growth, since overdoses among people outdoors and in shelters rose 66%. Yet neither obj-recoding nor the conclusion considers how much this could add or whether it could move the result. It is secondary to the recording question the article already treats as unresolved, so a reader's belief would not change much.

   About limitation:ev-recoding, objection:obj-recoding, conclusion The finding

2. Pass Rates the real claim

   The claim is 'Overdose deaths among people living in King County's subsidized housing have risen faster than the housing since 2021.' Critics often point to overdose deaths in supportive housing, so this narrower version plausibly circulates. The 'faster than the housing' framing is the fair test of such a claim, since raw counts would rise anyway as the housing grows. The context names the broader critics' version ('more people are dying after moving indoors'). The article tests that version as the lead common question ('More residents did die, but not faster than the housing grew ... rated False'). It also addresses the causal reading ('Living in subsidized housing is causing these deaths', 'Not shown'). The present-perfect 'have risen ... since 2021' could suggest the rise is still going on. The article covers that reading in key point 5 and obj-peak ('peaked at 281 in 2023 ... fell to 192 in 2025'). The conclusion stays within King County, 2021–2025 and the Medical Examiner's housing-service-site group.

3. Pass Rating follows from the evidence

   Taking the evidence as stated, the recorded counts clearly support the claim. Overdose deaths rose 68% against 25% bed growth, or +35% per bed. The result holds with rapid rehousing beds added (+25% per bed) and in the worst case on unknown living situations (+39%). The 2019 start shows a still larger rise. Against this, the only result that reverses the finding is the worst case on recording: if the housing group's whole rise in share (21% to 29%) came from recording, deaths rose 19%, below the beds. No evidence shows such a change, but 2021–2023 cannot be tested and the 2021 reports used a 'no fixed address' field. Before reading the finding I would have chosen Mostly true, with Mixed as the alternative. The deciding argument is that 'reversing the finding would need about 50 of the 56 shifted deaths to be recording changes.' That is reasonable, and the article calls it 'a close call.' The answer ('rose 68% and housing beds 25%, unless investigators changed how they record') is consistent with the finding.

4. Pass Confidence matches the evidence

   Moderate confidence fits the evidence. There is one source of death counts plus one independent bed count, and the recording question cannot be resolved for the key years. The article says so: 'Our checks can't test 2021 to 2023 ... they all use the same Medical Examiner counts; only the bed count is independent.' The title is a neutral question. The summary sets the recorded result (+35% per bed) beside the worst case (+19%, slower than beds' 25%), so it does not overstate. The limitations that matter are stated where they apply: beds are a one-night count, not residents; HUD leaves out income-restricted and recovery housing; turnover; how visitors are recorded; no published record of a method change; and the earlier 'no fixed address' field. Key point 6 says no true rate can be computed.

5. Pass Evidence is independent

   ev-od, ev-recoding and ev-all all rest on the same Medical Examiner records. ev-beds pairs those records with HUD's bed count. The 225 count of overdose deaths in permanent supportive housing from the county's overdose review also comes from Medical Examiner records. The article does not present these as independent. ev-od's limitation says 'Every death count in this article ... comes from the same Medical Examiner records ... only the HUD bed count is independent.' The changelog says agreement across comparisons was removed as a reason for the rating. The Denver trial (ev-denver) is independent and is used only for causation, as method step 5 states. Because the article is transparent about this, there is no problem of evidence presented as independent that is not.

6. Pass Lead questions hold up

   There is one lead question, obj-all-beds: 'Deaths the Medical Examiner investigated among King County subsidized-housing residents rose faster than the housing beds, 2021 to 2025', rated False. (1) Surprising: a reader who has seen only 'overdose deaths outgrew the housing: Mostly true' would likely guess deaths from all causes outgrew it too, so False is unexpected. (2) Not the headline: neither the title nor the answer mentions deaths from all causes. (3) Settled as worded: the claim fixes the measure (deaths the Medical Examiner investigated), the comparison (housing beds) and the period. The figures, 23% against 25% and 39.3 then 38.7 per 1,000 beds, settle it. Adding rapid rehousing beds (+34%) would only strengthen False. (4) Neutral: it corrects the critics' broader claim, but the headline gives the critics' overdose claim Mostly true, so neither side is singled out. (5) Stands alone: 'False: those deaths, from all causes, rose 23% from 2021 to 2025, while federally counted housing beds rose 25%.' 'Those deaths' refers to the question directly above, the words are everyday, and the sentence answers the question.

7. Pass Rating states its terms

   (a) The answer states what holds the claim back: '...unless investigators changed how they record where people lived.' Key point 3 gives the result in that case: 'If that explains the whole shift, overdose deaths rose 19%, slower than the beds.' (b) The rating depends on two choices of reading. The first is overdose deaths versus deaths from all causes; key point 4 gives the all-cause result (+23%, about as fast as the beds, flat per bed). The second is how 'the housing' is measured: key point 2 gives the result with rapid rehousing beds added, and key point 6 says no rate per resident exists. Readers who mean these other readings see their answer among the key points, close to the rating.

Coverage: I examined the full article payload: the claim, finding, answer, summary, context, key points, conclusion, five evidence items with their limitations, eight common questions (lead question: obj-all-beds), method steps, editorial note, changelog and the list of sources and takeaways. Following the division of labor, I took every evidence item and figure as stated. I did not check whether any source supports its takeaway. Every takeaway in the sources block, including all FactHub dataset analyses and the county, auditor, homelessness-authority and Urban Institute takeaways, is marked 'not yet checked'. None is marked established, so whether the figures (114→192 overdose deaths, 7,860→9,791 beds, 309→379 all-cause deaths, shares of 21%→29%, and others) match their sources is unverified and was outside my scope. The judgment that Mostly true is defensible rests on those figures being accurate. One assumption matters most: the worst case on recording (+19%, below the 25% bed growth) and the claim that reversing the finding needs about 50 of the 56 shifted deaths. If either figure were off, the close call between Mostly true and Mixed could change. I did not use the web or read source snapshots. Outside knowledge was used only to raise the visitor-recording alternative, phrased as something the article should address.