Seattle's homelessness spending tripled while homelessness kept rising.

1. Concern (minor) Context is complete

   State that King County's count method also changed in 2015 (larger area) and 2017 (Count Us In), not only 2022, and qualify the 2012 and 2015 count comparisons accordingly.

   Context says 'King County changed its counting method in 2022, so exact changes are uncertain; the rise holds under every method', and obj-method cites the pre-2022 rise '8,899 in 2012 to 11,751 in 2020' as method-independent. But the article's own datasets record method breaks in 2015 ('Count area expanded') and 2017 ('Count Us In canvasses all census tracts'); the promises dataset notes 'Homeless counts changed method in 2015, 2017 and 2022' and that 'The 2015 count covered a larger area than earlier counts', yet evidence:promises compares the 2015 count of 10,122 with the 2005 baseline without that caveat. The 2012-2014 counts are flat (8,899-9,106), and the step-ups coincide with the 2015 and 2017 changes. Sheltered counts, bed data and shelter stays do not depend on the street count, so the verdict stands, but the magnitude ('roughly doubled') is less certain than stated. Not raised as Material because the direction is supported by method-independent evidence. Confidence: High.

   About evidence:count, limitation:count, evidence:promises, conclusion The finding

2. Concern (minor) Uncertainty is stated

   Note that the 4th-most unsheltered rise is as counted, not method-adjusted, and qualify 'did worse than most of its peers': under the high adjustment the rise is below the peer median.

   The conclusion says flatly that King County's 'unsheltered homelessness rose fourth-most' and 'it did worse than most of its peers'. The dataset analysis king-change-ranks is explicitly 'As counted; the 2022 count-method change may account for 0 to about 3,600 of King County's 2024 count', and that change affects the unsheltered count specifically. For the total rate the article itself gives 45% as counted, 36% with the central adjustment, and 'about 14% above 2019' under the high adjustment, against a peer median of 26%. So under the high end of the article's own range King County did better than the peer median. evidence:other-regions correctly says 'under the central estimate', but the conclusion and the rents objection drop the qualifier. The verdict on the claim is unaffected, and the conclusion already lowers confidence for which changes would work. Confidence: High.

   About conclusion, evidence:rents, evidence:other-regions, limitation:count The finding

3. Not checked Central claims are accurate

   One-sentence summary: Seattle/King County homelessness spending rose about threefold while the homeless count roughly doubled, so the claim that the spending reduced homelessness is rated False. Claim ratings: 1. Seattle budget ~$71.3M spent 2017 -> $228.7M budgeted 2026 ('tripled'): Unverifiable here; both budget PDFs unreadable. Plausible in nominal terms; Cascade PBS (2018) reports a narrower $54M 2017 'set aside', a scope difference, not a conflict. Disclosed as nominal, actual vs budget. 2. Count 8,899 (2012) -> 16,868 (2024) -> 18,365 (2026); unsheltered 2,618 -> 11,829 (~4.5x): True as counted (dataset king-count-series). 2026 KCRHA PDF unreadable but dataset matches. 3. Sheltered 'about 5,700 to 7,100' in full counts: True (5,683-7,058). 4. Temporary beds 6,038-7,746; ES +83%, TH -51%, 0.64 -> 0.41 beds per person, ranks 12/67 and 53/60: True (king-capacity-change, king-capacity-ranks-2024). 5. Median shelter stay 24 -> 119 days, 57th of 67, peer median 30 -> 56, Houston 36: True (king-days-homeless-vs-peers, king-spm-ranks). 6. Rents: rent growth 60th of 67, unsheltered rise 4th, correlation ~0; total-rate residual intervals all include zero: True as counted (king-change-ranks, rent-growth-vs-change, king-total-vs-rent); see uncertainty-stated on method adjustment. 7. Promises missed (5,700 of 9,500 units; 565 HTH housed; 288 vs 2,065 Harrell; Pathways 738): Mostly True; matches promises dataset; the 'every promise' framing depends on the article's distinction between ending/reducing homelessness and activity targets, which it states. 8. KCRHA forensic findings ($44.7M negative cash etc.): Unverifiable (PDF unreadable). The Mayor's release does say KCRHA 'is unable to account for $13 million', as the article reports. 9. What worked: exits 49.2% (7th of 67), returns 15.3% below median 18.8%: True (king-spm-ranks). EHV leasing speed: Unverifiable from the dashboard text. 10. Houston: 3,325 in 2025 and unsheltered +15.8% confirmed by CFTH; 2011 baseline of 8,538 is not on that page, so 'about 61%' is Mostly True/unverified. 11. GAO 9% per $100 rent: True (GAO Fast Facts). 12. Finding False for 'spending has reduced homelessness': True under the plain reading; the counterfactual is explicitly scoped out. No central claim I could read is wrong. Result is NotChecked rather than Pass because the spending figures behind 'tripled' (a central claim in the title and answer) and the forensic findings could not be read. Confidence: Medium.

   About evidence:spending, source:seattle-budget-2019, source:seattle-budget-2026, source:clark-nuber-2026

4. Pass Claims are supported and labeled

   Every evidence item names its sources, and the key numbers trace to FactHub datasets with saved analyses or to official documents. Inferences are labeled as such: the conclusion says 'The data point to how the system operated', concedes 'What this check cannot show is the counterfactual', and separates confidence in the verdict from 'any claim about which specific changes would work'. No central claim rests on assertion. Confidence: High.

5. Pass Sourcing is strong and transparent

   Central claims rest on primary sources: HUD PIT, HIC and System Performance Measures (via documented, reproducible FactHub datasets with archived HUD workbooks), adopted City of Seattle budgets, King County program reports, the original plan and pledge documents (quoted with archived copies in the promises dataset), and an independent forensic review. Secondary sources (Seattle Times, Cascade PBS, PubliCola) are used only for context or clearly labeled 'independent reporting'. The article separates the Mayor's '$13 million' claim from the forensic report. Source-to-claim mapping is clear via sourceIds. Confidence: High.

6. Pass Framing is fair and neutral

   The article includes a full 'Some parts of the system worked well' item, concedes the counterfactual objection ('That may be true'), says King County's problems 'are not unique', and treats the rent explanation as 'a real driver'. The Mayor's $13M claim is corrected without partisan framing. The claim is stated in its natural form. Tone is neutral. Confidence: High.

Coverage: Read the full article payload and the four cited FactHub datasets (king-county-homeless-count-beds:2, homeless-system-performance:1, homelessness-rate-vs-rent:1 analyses, seattle-homelessness-promises:1) via get_node_for_audit. Read archived HTML snapshots of st-promises-2022, mayor-kcrha-2026, cfth-houston-2025, cascade-1b-2018, gao-20-433, colburn-aldern-2022 and hud-ehv-dashboard (the EHV dashboard itself is an embedded app; its per-authority figures were not in the text). Could NOT read (PDF not supported by the snapshot tool): seattle-budget-2019, seattle-budget-2026, kcrha-pit-2026, ten-year-plan, hth-recovery-plan, hth-2025-report, clark-nuber-2026. The FactHub dataset sources have no URL for get_source_snapshot but were read as nodes. The spending figures, the KCRHA forensic findings and the HTH 2025 figures therefore could not be checked against their sources.

1. Concern (minor) Confidence matches the evidence

   Say whether the 2012-2020 rise survives the 2015 count-area expansion and 2017 Count Us In change, or drop 'the rise holds under every method' from the context.

   High confidence that the count did not fall is reasonable: even the article's high method adjustment leaves 2024 'about 14% above 2019', and 2026 is higher again. But the context says 'the rise holds under every method', and obj-method uses '8,899 in 2012 to 11,751 in 2020' as the pre-2022 baseline under 'earlier street-count methods'. The article mentions only the 2022 change. It does not address whether King County's count also changed method within 2012-2020 (for example a larger count area in 2015 and the 2017 Count Us In canvass of all census tracts), which would mean the 'roughly doubled since 2012' figure in the answer and title spans several method breaks. The verdict holds; the magnitude is stated more firmly than the stated evidence supports. I would keep High for the verdict, with a caveat on the size of the rise.

   About conclusion, evidence:count, limitation:count The finding

2. Concern (minor) Strongest objections answered

   Address whether the 4th-most unsheltered rise survives the 2022 method adjustment, and whether shifting beds from transitional housing to shelter lengthened measured shelter stays.

   The strongest objections are (1) counterfactual, (2) housing costs, (3) count-method change, (4) not enough money. All four appear in the objections and are answered with evidence; obj-counterfactual is conceded fairly. Two gaps remain. (a) The rents answer rests on 'unsheltered homelessness rose 4th-most of 67', but evidence:rents applies the method adjustment only to the total rate (45% to 36%), while limitation:count says the 2022 change added about 1,100 (range 0 to 3,600) people, which falls on the unsheltered count. Taken together, at the high adjustment the total rise (about 14%) is below the 26% peer median, so 'it did worse than most of its peers' in the conclusion depends on the central estimate. (b) The article reads longer shelter stays as a system failure, but evidence:capacity shows emergency shelter beds up 83% while transitional housing fell 51%. The article does not address whether converting long-stay transitional programs and 24-hour enhanced shelter into the 'emergency shelter' category mechanically raises median days in shelter. Neither gap touches the verdict.

   About conclusion, evidence:rents, evidence:shelter-stays, evidence:capacity, limitation:count The finding

3. Concern (minor) Evidence is independent

   Note that the count, rents and other-regions items all rest on the same HUD Point-in-Time series; the bed and shelter-stay data show system strain, not the level of homelessness.

   The summary says 'HUD homeless counts, HUD bed inventory and system-performance data, and two decades of official targets all show homelessness growing'. Taken by method and sources: evidence:count (PIT, fh-count-beds), evidence:rents (PIT rates, fh-rate-rent), evidence:other-regions (fh-rate-rent, PIT counts) and the end-homelessness outcome in evidence:promises (the 2015 count) all rest on the same Point-in-Time series. evidence:capacity (HIC) and evidence:shelter-stays (SPM) are independent collections, but they measure capacity and stays, not whether homelessness rose. So the direct evidence for 'did not reduce homelessness' comes down to one measurement series, though that series is the standard measure and is corroborated by independent signs of strain. Independent support for the system-strain reading remains, so this is Minor.

   About conclusion, evidence:count, evidence:rents, evidence:other-regions The finding

4. Pass Rates the real claim

   Claim: 'Seattle's increased spending on homelessness has reduced homelessness.' A reader repeating it means homelessness went down because of the spending. The article rates that plain version and explicitly scopes out the stronger defensive version ('This check asks whether the spending reduced homelessness, not whether homelessness would have been worse without it'), which it also raises as obj-counterfactual. Using King County data for a Seattle claim is justified in context ('Seattle and King County share one homelessness system (HUD Continuum of Care WA-500)'). The conclusion stays within Seattle/King County, 2005-2026. I would rate the same claim.

5. Pass Rating follows from the evidence

   Before reading the verdict, from evidence:count ('8,899 ... in 2012, 16,868 in 2024 and 18,365 in 2026'), the method limitation (high adjustment still leaves an increase) and evidence:spending (budget roughly tripled), I would rate the plain claim False: homelessness did not go down. The article's 'False' and its answer ('count roughly doubled since 2012 while Seattle's homelessness budget tripled') agree. Mixed or Misleading would fit only the counterfactual version, which the article declines to rate and explains why.

Coverage: Article payload only (claim, finding, answer, confidence, summary, context, conclusion, all nine evidence items and limitations, five objections, six method steps, source list). Evidence taken as stated; per the Skeptic's division of labor, source snapshots were not used for this run's judgments. Outside knowledge used only to name objections the article should address.

1. Concern (minor) Consistent house style

   Use one rounding per figure ($228.7 million and $71.3 million everywhere), one form of '4th-most', spell out HUD at first use, and explain respondent-driven sampling and rapid rehousing.

   Rule: the same rounding for the same figure everywhere; abbreviations spelled out at first use; technical terms explained. (1) The 2026 budget is '$229 million' in the conclusion but '$228.7 million' in keyPoints, evidence:spending and obj-1b; the 2017 figure is '$71 million' in the conclusion and keyPoints but '$71.3 million' in evidence:spending. (2) The conclusion says 'fourth-most' while evidence:rents and obj-rents say '4th-most'. (3) HUD is used from the description ('using HUD data') and context ('HUD Continuum of Care WA-500') without being spelled out in reader-facing text. (4) 'respondent-driven sampling' (limitation:count), 'rapid rehousing' (evidence:what-worked) and '95% prediction interval' (limitation:rents) are not explained for a general reader. Each is a minor lapse; none changes a figure's value.

   About conclusion, evidence:spending, limitation:count, evidence:what-worked, limitation:rents The finding

2. Pass Title states the finding

   Title: 'Seattle's homelessness spending tripled while homelessness kept rising.' It names the subject, states the verified finding in positive form, does not repeat the claim, and a reader who sees only the title knows the spending did not bring homelessness down. seoTitle ('Seattle homelessness spending tripled since 2017 while the homeless count kept rising') states the same finding with search wording. No clickbait or loaded words. Rule (lead with what is true, resolve the claim) is met.

3. Pass Verdict fields agree and stand alone

   Read as: FactHub rates the claim "Seattle's increased spending on homelessness has reduced homelessness." False (High confidence): King County's homeless count roughly doubled since 2012 while Seattle's homelessness budget tripled since 2017. The claim is the assertion as it circulates, one declarative sentence, not its correction. The answer stands alone and says what is true. Title, summary ('all show homelessness growing ... while spending rose'), keyPoints and conclusion ('did not reduce homelessness') all express the same False conclusion; no stale wording from another label.

4. Pass Each section does its job

   summary is one sentence naming the lines of evidence (HUD counts, bed inventory, system-performance data, official targets). keyPoints state established findings. context is one paragraph of framing (shared system, count method, scope of the question). conclusion weighs the evidence, takes on the competing housing-cost explanation, notes what worked, and states what is unproven ('What this check cannot show is the counterfactual') and how confidence differs. editorialNote gives scope ('2005-2026'), caveats (not inflation-adjusted, promises from news excluded) and review status ('No external specialist review has been recorded'). No hand-written independence note.

5. Pass Limitations are specific

   Each limitation is specific to its item: spending (2017 actual vs 2026 budget, no inflation adjustment, COVID funds, double counting, the $1 billion figure); count (single nights, 2022 method change with 0-3,600 range, carried-forward years); capacity (beds exist vs staffed; permanent beds not shelter); shelter-stays (self-reported, 77-87% coverage, COVID); rents (within normal variation, peers also worse); promises (some met, closest measure, news reliance, archived plan); governance (evaluation not audit, no fraud found, $13M not in report); what-worked (partial measures); other-regions (Houston rents, its 2025 unsheltered rise, method changes). None is boilerplate.

6. Pass Tone is measured and respectful

   Language is measured: 'The evidence converges', 'The fair reading is', 'Confidence is high that ... it is lower for any claim about which specific changes would work'. Objections are answered respectfully and with evidence; obj-counterfactual begins 'That may be true'. 'Every official promise since 2005 to end or reduce homelessness was missed' is an absolute, but the article qualifies it in the same key point ('many activity targets were met') and the promises limitation, so it is justified. No ridicule or sarcasm.

7. Pass Readers can verify it

   All 19 sources carry role labels ('Official budget document', 'Independent forensic accounting review', 'FactHub dataset', 'News explainer'), full titles and descriptions of what each supports. Key figures have primary anchors: archived budget PDFs, archived plan and program documents, the KCRHA count FAQ, and FactHub datasets with saved analyses. Method steps cover claim definition (step 1), several independent HUD collections and the alternative explanation (steps 3-4), sources reviewed (steps 2, 3, 5) and recorded limits (step 6).

Coverage: Article payload only: title, seoTitle, socialTitle, description, claim, finding, answer, confidence, summary, keyPoints, context, conclusion, nine evidence items with limitations, five objections, six method steps, editorialNote, chart captions, and 19 sources with roles, titles and descriptions. No live web or source snapshots were used for this run.