FactHub curated dataset

Seattle Times I-638 coverage: departures from the record, coded by side

Every statement FactHub coded as departing from the documentary record in The Seattle Times' news coverage of Washington Initiative 638 and trans athletes, January 2025 to October 30, 2026, with which side of the I-638 debate each one favored. 47 coded items from 22 stories: 27 departures in the paper's own voice, 7 quotes the paper left unchecked, 10 errors that favored neither side and 3 items reviewed and excluded.

Coverage: January 2025 to October 30, 2026 (story publication dates)

Node: dataset:seattle-times-i638-departures:1

What this dataset measures

The Seattle Times is Washington's largest daily newspaper and ran more staff news stories on I-638 than any other Washington outlet. This release records each item a two-coder review and a red-team pass flagged as departing from the measure's text, official statistics, peer-reviewed research or the paper's own later reporting, so readers can check every item and every count.

Curated data tables

These related tables form this versioned dataset release. Each table has its own downloadable file and field definitions.

Coded departures from the record, Seattle Times I-638 news coverage

Every item flagged in review, including neutral and excluded items. Own-voice departures are the rows whose category is factual-firm, factual-arguable or framing.

Unit of observation: One coded statement or headline in one news story

View this table

seattle-times-i638-departures.csv text/csv · 22.9 KB (23,490 bytes)
11 field definitions
FieldTypeMeaningUnitsValue type
id string Stable item id. B = found by the blind second coder; P = found in the primary audit, the symmetry test or the red-team review. — observed
date date Publication date of the story (YYYY-MM-DD). — observed
url string The story's URL. For row B7 the section page is given instead, because the story URL contains a minor's name. — observed
quote string The coded wording, quoted or closely described from the story. Headlines and omissions are marked as such; ellipses mark cuts. — observed
whats_true string What the record shows, with the record named (measure text section, official statistic, study, or the paper's own later reporting). — derived
category string factual-firm: false in the paper's own voice on a plain reading of the record. factual-arguable: unsupported or materially imprecise in the paper's voice, with a defensible alternative reading. framing: the paper's wording overstates, hedges or contradicts what the record plainly shows. unchecked: a quote from a source that the story printed without checking a claim the record contradicts (not the paper's own words). neutral: an error that favors neither side. excluded: reviewed and not counted, with the reason in note. — derived
favors string Which side of the I-638 debate the item's error or framing helps: No (opponents of the measure), Yes (supporters) or Neutral. — derived
severity string major or minor: whether the item bears on a central argument of either campaign. — derived
correction string Whether the Times corrected the item as of October 30, 2026: Yes, No (with any correction note that covers something else), or n/a. — observed
found_by string Which review step found the item: blind coder, primary audit, symmetry test or red team. — observed
note string Adjudication notes, including the reason for each excluded item; empty when none. — derived

Saved analyses

Coded items by category and side favored

Question: How many coded items fall in each category, and which side did each favor?

Result: Factual, firm: 5 No, 0 Yes. Factual, arguable: 5 No, 0 Yes. Framing: 16 No, 1 Yes. Unchecked quotes: 3 No, 4 Yes. Neutral: 10. Excluded: 3. Total 47.

Operation: Count rows of the departures table grouped by category and favors.

Scope: All 47 rows.

View this analysis

  • Counts depend on the coders' category and side judgments; the rows let readers recount under their own reading.

Departures in the paper's own voice, by side favored

Question: Of the departures written in the paper's own voice, how many favored each side?

Result: 26 favored No and 1 favored Yes, across 16 stories.

Operation: Count rows with category factual-firm, factual-arguable or framing, grouped by favors; count distinct story URLs.

Scope: 27 own-voice rows from 16 stories.

View this analysis

  • Excludes neutral errors (10), unchecked quotes (7) and excluded items (3).
  • Framing items (17 of 27) are judgments; the firm factual errors alone run 5 to 0.

Robustness checks

Firm factual errors by side favored, and corrections

Question: How many firm factual errors favored each side, and how many did the Times correct?

Result: 5 firm factual errors, all favoring No and all overstating whom the restriction reaches. The Times corrected 3 (both 2025 Quilcene headlines and the October 14, 2026 Camas story); October 22 and October 30, 2026 were uncorrected as of October 30.

Operation: Count rows with category factual-firm, grouped by favors and by correction.

Scope: 5 rows (B4, B5, B27, B29, B36).

View this analysis

  • The October 14, 2026 pre-correction wording comes from an Internet Archive capture and the correction note.

Robustness check for: Departures in the paper's own voice, by side favored

Unchecked quotes by side favored

Question: When the paper printed a source's claim that the record contradicts without checking it, which side did the claim favor?

Result: 4 favored Yes and 3 favored No: roughly even. The Yes side's 'birth records / medical history suffice' claim was left unchecked 4 times.

Operation: Count rows with category unchecked, grouped by favors.

Scope: 7 rows (B16, B21, B22, B23, B24, B34, B37).

View this analysis

  • Counts only unchecked claims the record contradicts in a story about that claim; many other quotes from both sides went unchecked without being wrong.

Robustness check for: Departures in the paper's own voice, by side favored

Stories read and stories with coded items

Question: How many stories were reviewed, and how many contain coded items?

Result: 44 news stories read; 22 contain at least one coded item; 16 contain an own-voice departure; 47 items coded.

Operation: Count stories in scope; count distinct story URLs in the table overall and among own-voice rows.

Scope: Seattle Times staff, partner and freelance news stories on I-638 or Washington trans athletes, January 2025 to October 30, 2026.

View this analysis

  • The count of stories read (44) comes from the review's story list, which is not part of this table; the table records only stories with coded items.

Own-voice departures and unchecked quotes, by type and side

Question: How do departures in the paper's own voice compare by side with quotes it left unchecked?

Result: Own voice: 26 No, 1 Yes. Unchecked quotes: 3 No, 4 Yes.

Operation: Counts from counts-by-category-favors, relabeled for display: three own-voice categories plus unchecked quotes, each split by side.

Scope: 34 rows: 27 own-voice departures and 7 unchecked quotes.

View this analysis

  • Neutral and excluded items are not shown.

How firmly the paper rebutted each side's central verification claim

Question: When the paper checked each side's strongest claim about verification in its own voice, did it rebut it flatly or with a hedge?

Result: The No side's genital-exam claim was checked in four stories, five hedged sentences, never flatly. The Yes side's strong-form claim got two flat rebuttals, which were accurate.

Operation: List own-voice checks of the No side's 'I-638 requires genital exams' claim and of the Yes side's 'sports physicals already confirm sex' claim, coded flat or hedged.

Scope: Six Seattle Times news stories, January 9 to October 30, 2026.

View this analysis

  • Firmness is a reading of wording; the four hedged items are coded as framing rows B9, P2, P3 and B33. The flat rebuttals are accurate and are not counted as departures.

Other tests of the coverage, including those that found no tilt

Question: Do other measures of the coverage (wire selection, prominence, comments, sourcing, peers, other measures) show the same lean?

Result: Five tests found no tilt (wire selection, prominence, comment closures, funding, vocabulary); sourcing was close over the campaign but tilted No in the final week; the hedged-check pattern is specific to I-638.

Operation: Summary of separate follow-up reviews run alongside the coding.

Scope: Seattle Times I-638 coverage, 2025-2026; comparison sets as described per row.

View this analysis

  • These are summaries of separate reviews whose story-level coding is not part of this release. Treat them as context for the main table, not as independently checkable counts.

Sources

  1. The Seattle Times Primary record: the news stories whose wording is coded. Each row's url names its story.

    Retrieved October 30, 2026. All rights reserved (The Seattle Times). Short quotations only, for criticism and review; full story text is not redistributed.

    Reliability: News report. Washington's largest daily newspaper and the subject of this review. Its own published wording is the record being coded.

  2. Washington Secretary of State Reference record: the measure's filed text, used to check statements about what I-638 requires and whom it covers (Sections 1 and 3).

    Retrieved October 30, 2026. Public record. Public government record; linked, not hosted.

    Reliability: Government. Official filed text of Initiative Measure IL26-638 as published by the Secretary of State; the measure itself was written by its sponsors.

  3. American Academy of Pediatrics Reference record: what AAP well-visit guidelines list for girls aged 11-17, used to check statements about well visits.

    Retrieved October 30, 2026. Copyright American Academy of Pediatrics. Linked, not hosted.

    Reliability: Academic. The AAP's national guideline for preventive pediatric care (4th edition, 2017). The AAP opposes measures like I-638, but the guideline predates it and does not address sports.

  4. Data Resource Center for Child and Adolescent Health (NSCH) Reference record: Washington children without a personal doctor or nurse, 2023-24, used to check the paper's 18% figure.

    Retrieved October 30, 2026. Public data. Linked, not hosted.

    Reliability: Government. National Survey of Children's Health, funded by HRSA and fielded by the Census Bureau; query tool run by the Child and Adolescent Health Measurement Initiative.

  5. Hilton & Lundberg, Sports Medicine Reference record: review of strength and lean mass after testosterone suppression, used to check statements about the research. Free copy

    Retrieved October 30, 2026. CC BY 4.0. Linked, not hosted.

    Reliability: Peer-reviewed study. Peer-reviewed narrative review (Sports Med 2021;51:199-214). The authors have argued publicly for protecting the female category.

  6. Harper et al., British Journal of Sports Medicine Reference record: systematic review of strength, lean mass and muscle area after hormone therapy.

    Retrieved October 30, 2026. CC BY-NC 4.0. Linked, not hosted.

    Reliability: Peer-reviewed study. Systematic review of 24 studies (Br J Sports Med 2021;55:865-872). Lead author Joanna Harper supports transgender inclusion and is quoted by the Times.

Limitations and reproducibility

Partially reproducible: Every coded item gives the story URL, the coded wording and what the record shows, so anyone can re-check each item and recount the tables. The coding judgments themselves (especially framing) cannot be recomputed mechanically. Pre-correction wording for three corrected items comes from Internet Archive captures or the paper's correction notes.

  • Framing calls are judgments. Two independent coders and a red-team review reduce, but do not remove, the chance that another reader would code an item differently. Inter-coder agreement was not computed formally.
  • Scope: 44 Seattle Times staff, partner and freelance news stories on I-638 or Washington trans athletes, January 2025 to October 30, 2026. Opinion pieces, letters and wire copy are excluded, so the 12-to-2 opinion-page balance against the measure is not counted here.
  • Only items that depart from the record are listed. Accurate statements are not tallied, so the table measures the direction of errors and slanted framing, not their rate per sentence.
  • Quotes the paper printed without checking are tallied separately as 'unchecked', because they are not the paper's own words. They split roughly evenly: 4 favored Yes and 3 favored No.
  • Some pre-correction wording (for example the October 14, 2026 Camas story) comes from an Internet Archive capture and the paper's correction note, not from the current page.
  • One story URL contains the name of a minor athlete; for that row (B7) the url field gives the section page instead.
  • The Seattle Times was not asked to comment before publication.

Publications using this dataset

Version details

The exact citing versions, and the version of this dataset each cites.

  • Article seattle-times-i638-coverage-departures — version 2 (published) cites version 1
  • Article seattle-times-i638-coverage-departures — version 1 (published) cites version 1
  • Chart seattle-times-i638-departures-by-side — version 1 (published) cites version 1