← Back to the article

Audits

Are Seattle overdose deaths still over twice 2019's, with U.S. below?

Version 2. 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

First Look

0 material, 0 minor, 7 of 7 passed · version 8 · claude-haiku-5-5 · 14 November 2026

A: Passes the first look: nothing obviously wrong. Minor notes may be reported as feedback without lowering the grade.

7 checks
  1. Pass Title matches the evidence

    The title says Seattle's deaths are over twice 2019's and the U.S. is below 2019. ev-seattle shows 478 vs 200 in 2025 (2.4 times) and 392 vs 173 for 2026 to date (2.3 times). ev-us shows 70,646 vs 71,130 in 2025 and about 5% below in the latest 12 months. The wording matches the figures.

  2. Pass No obvious contradictions

    Finding True, the answer, and the conclusion ('The claim is true as worded'; 'We rate the claim True') agree. The key points match the evidence figures and the conclusion's residence-basis caveat. Key point 3 gives the reported 5% U.S. figure while the conclusion also cites CDC's predicted 3.8%; that is disclosed nuance, not a contradiction.

  3. Pass Plausible justification

    The conclusion explains that Seattle is 2.4 times its 2019 count with a 2.1 times per-person rate, that the U.S. is just under 2019 and 5% lower per person, and it weighs the two points that cut the other way. A reasonable reader can follow why True fits.

  4. Pass Sources are relevant

    The King County dashboard and methods report, CDC provisional national and county counts, and FactHub's saved analyses and two earlier FactHub fact checks are all on topic for overdose death counts, rates, and the where-occurred versus where-lived basis.

  5. Pass Common questions answered

    There are eight common questions (peak, 2026 pace, western counties, U.S. near 2019, residence basis, population, finality, reason for the lag). Each response addresses its question directly.

  6. Pass Headline and verdict agree

    Step b sentence: 'Are Seattle overdose deaths still over twice 2019's, with U.S. below? Yes.' The reader takeaway is that Seattle is still more than double 2019 and the U.S. is below 2019. That matches the claim, the True finding, and the answer. The claim is positive and the title is positive, so there is no polarity problem.

  7. Pass No Wikipedia or open wikis

    Every cited source is a King County government page or dataset, a CDC NCHS dataset or document, or a FactHub record. The institutions named in the evidence are King County, Public Health - Seattle & King County, CDC NCHS, and the U.S. Census Bureau. The Wikimedia Commons link is only the credited article image, which is not a cited source. No open wiki is cited.

Coverage: I could not open or read any source, dataset, API record, or snapshot, and I did not check any quote or figure against the web. Every takeaway in the sources block carries status 'not yet checked', and none is marked established. I therefore treated all takeaways as unchecked and did not rely on any of them for a finding. The keyPoints kind 'established' is the article's own label, not a source-audit result, so I did not treat it as confirmation. Checks 1 to 5 were judged from the article text: headline, claim, finding, answer, conclusion, key points, evidence summaries, objections, and the figures they quote. I checked those figures only for internal arithmetic consistency (478/200 = 2.4; 392/173 = 2.3; 207/297 = 30%; 413/811 = 50.9%; 65,724 vs 68,322 = 3.8% lower). Check 6 worked through all steps for this question title. Check 7 looked at every source's name, publisher, URL, and role, and at the evidence institutions. Noted but not reported as findings: the conclusion calls the King County residence-based count (408 to 848) 'the independent check' that the rise is 'not an artifact' of county counting, although ev-definitions says no Seattle residence count exists. That is a debatable judgment, not an obvious error. The U.S. 5% figure is CDC's reported count, which CDC marks as underreported; the article also gives the predicted 3.8%, so I treated that as disclosed nuance.

Skeptic

0 material, 2 minor, 5 of 7 passed · version 9 · claude-opus-5-5 · 14 November 2026

B: The reasoning holds, with several minor gaps, such as secondary objections left out or confidence a step high.

7 checks · 2 concerns
  1. Concern (minor) Confidence matches the evidence

    Say the 2025 U.S. count could end at or above 2019 once complete, and give CDC's predicted 2025 count, rather than saying final data 'could narrow' the gap.

    I would also give High confidence. The Seattle ratio is far above 2x on every measure, and an independent residence-based CDC count corroborates it. For the U.S., the latest 12 months are below 2019 even on CDC's predicted counts (3.8%). One limitation is understated, though. Key point 3 states as established that "U.S. overdose deaths in 2025 were 70,646, just under 2019's 71,130". The conclusion says the count and the rate "both put 2025 below 2019". The ev-us limitation says only that final data "could narrow that gap." The gap is 0.7%. By the article's own figures, CDC's late-report adjustment for the latest period is about 2% (64,433 reported against 65,724 predicted). An adjustment of under 1% for calendar 2025 would therefore close the gap or reverse it. The article does not give CDC's predicted count for the 12 months ending December 2025. The rate (5% lower) and the latest-12-month comparison still carry the verdict, so a reader's overall belief would not change. But on the count, 2025 is not reliably below 2019, and the article should say so. The title, description, and summary do not overstate the finding.

    About limitation:ev-us, key-point:3, conclusion, evidence:ev-us The finding

  2. Concern (minor) Strongest objections answered

    In obj-2026, address whether pending-toxicology cases in 2026, absent from confirmed 2019 counts, could move Seattle's 2.3x ratio or its 5% drop.

    The strongest objections are answered with evidence: - Residence versus occurrence (obj-residence): answered with King County's residence-based doubling. - "The U.S. is really about the same" (obj-us-barely): answered with the rate and the latest-12-month figures. - Population growth (obj-pop): answered with the age-adjusted rate. - "Seattle is just part of a regional fentanyl wave" (obj-west): answered with the 15 peer counties. - Provisional data (obj-2026): addressed. One secondary gap remains. The ev-seattle limitation and tk-pending say the 2026 Seattle count "includes suspected overdoses still waiting on toxicology tests." The 2019 comparison count is fully confirmed. The obj-2026 response answers a different problem: "Comparing January 1 to November 13 in each year keeps the windows the same." Matching the dates does not match case status. The article does not say how many 2026 Seattle cases are pending, or whether resolving them could move the 2.3x ratio. That also affects the 5% versus 17%-to-30% comparison in obj-pace. The full-year 2025 ratio of 2.4x carries the verdict regardless, so this is minor. The article also asserts that each place is "counted the same way in 2019 and since" (ev-definitions) without stating whether the medical examiner's case definition or toxicology practice changed over the period. The CDC residence count partly guards against that.

    About objection:obj-2026, limitation:ev-seattle, takeaway:tk-pending, conclusion The finding

  3. Pass Rates the real claim

    The claim is "Seattle's overdose deaths are still more than double their 2019 level, while U.S. overdose deaths have fallen below 2019." A reader who repeats it means two things: Seattle's current count is over twice 2019's, and the national count is now under 2019's. Method step 1 defines the claim the same way, as two comparisons with 2019, in 2025 and in the latest 2026 data. The answer and conclusion rate exactly that. The article does not swap in an easier version, such as rates only, or a harder one, such as "Seattle is uniquely bad." Two context points, the King County regions and the western peer counties, appear only as qualifiers: "Two facts cut the other way." They are not used to stretch the verdict. The editorial note also limits scope: it does not judge causes or policy. The Seattle count is by where deaths happened, which is how the county publishes city figures and so how the claim circulates. The article says so in key point 1 and the editorial note. The conclusion stays within the claim's places and its period, 2019 to November 2026.

  4. Pass Rating follows from the evidence

    Taking the evidence as stated, I would rate this True before reading the verdict. Seattle half: 200 deaths in 2019 against 478 in 2025 is 2.4 times. The January 1 to November 13 count is 392 against 173, or 2.3 times. The age-adjusted rate is 2.1 times. On a residence basis, King County's CDC count rose from 408 to 848, about 2.1 times. Every measure is above 2x, so "still more than double" holds. U.S. half: the 2025 count was 70,646 against 71,130, under 1% lower. The latest 12 months reported are 4.9% below the same months of 2019. CDC's predicted counts, which adjust for late reports, put them 3.8% below. The per-person rate was 5% lower in 2025. The claim's "have fallen below" is present tense, and the latest data plus the predicted-count adjustment settle it. The answer, "2.4 times as many... U.S. deaths have fallen just below 2019", matches the True finding. Nothing in the evidence points to Mostly true or lower.

  5. Pass Evidence is independent

    Several items share sources: - Seattle: ev-seattle, ev-regions, and the Seattle part of ev-percap all come from the King County dashboard. - U.S.: ev-us and the U.S. part of ev-percap share CDC's VSRR series. The article says this openly. The conclusion states: "Seattle's counts, age-adjusted rate and 2026 figures all come from the county's one dashboard, so they are one source, not three" and "For the U.S., the count and the per-person rate come from the same CDC figures, so they are one line of evidence." For the Seattle side it names a genuinely independent check: CDC's death-certificate counts by county of residence (ev-peers, King County 408 to 848). That system and counting basis differ from the medical examiner's occurrence-based dashboard. Only the county publishes a Seattle city figure, so a single source for Seattle is unavoidable, and the article does not present overlapping items as separate confirmation. No concern.

  6. Pass Lead questions hold up

    obj-peak, "Have Seattle's overdose deaths fallen about as much from their peak as U.S. deaths?", answered "Yes, by 37% vs 35%": 1. Surprising: a reader who has just seen "more than double 2019" would likely guess no. 2. Not the headline: the title and answer don't mention peaks. 3. Settled as worded: "about as much from their peak" naturally reads as a percentage decline, and 37% against 35% settles it. The different peak years (2023 vs 2022) don't make the wording false in either sense. 4. Neutral: it is a fact, not a talking point being corrected. 5. Stands alone: the first sentence, "Seattle's 2025 count was 37% below its 2023 peak of 763; U.S. deaths were 35% below their 2022 peak", uses everyday words. obj-pace, "Are Seattle's overdose deaths falling as fast in 2026 as the rest of King County's?", answered "No: 5% vs 17% to 30%": 1. Surprising: most readers would assume similar trends within one county. 2. Not the headline: the description mentions it, but the title and answer do not. 3. Settled as worded: "falling as fast" is settled by the year-on-year percentage changes. 4. Neutral. 5. Stands alone: the response opens "No." and immediately gives plain figures, with no unexplained terms or back-references. Both pass all five tests.

  7. Pass Rating states its terms

    The finding is True, so part (a) does not apply. Part (b) applies because a definition is in play: Seattle's count is by where the death happened, not where the person lived. Key point 1 handles this where the rating is seen. It says "Deaths in Seattle rose from 200 in 2019 to 478 in 2025" and gives the result under the other reading, as far as data allow: "No resident count exists; King County residents' deaths also doubled." For the U.S. side, the count and the per-person rate agree in direction. Key point 3 gives the count for 2025 and the latest 12 months, so no other reading flips the result. The definitional choice is stated in a key point, as required.

Coverage: I reviewed the full article payload: claim, answer, finding, confidence, summary, description, title, key points, context, conclusion, all six evidence items and their limitations, all eight objections (including lead objections obj-peak and obj-pace), method steps, editorial note, and the source and takeaway list. I took every evidence item as stated. Every source takeaway is marked "not yet checked", none "established". That includes the King County dashboard figures, the CDC reported and predicted counts, and the FactHub saved analyses such as the per-capita rates, regional changes, and western-county comparison. So my judgments rest on unverified figures. If the Source Checker finds a key number wrong, these results would need revisiting: the Seattle 200/478 and 173/392 counts, the U.S. 71,130/70,646 counts, the 3.8% predicted-count gap, or King County's 408 to 848 residence count. I did not read source snapshots or charts. I used outside reasoning only to raise two points: the pending-toxicology mismatch, and the possibility that late reports push the 2025 U.S. count to or above 2019. Both are framed as things the article should address, and the second relies on the article's own reported-versus-predicted figures.