BANNED BOOKS · GDELT VALIDATION · 4 OCTOBER 2026 · DATE AUDIT

Checking the news-to-district matches

A fixed sample of 36 records was reviewed from the 1,054-record candidate set. Three records were selected from each month, July 2022 through June 2023.

These are news-record classifications, not counts of bans. Nine records support at least one original district candidate; eight support none; nineteen remain unresolved. Relevant coverage can still be assigned to the wrong district.

October 3 follow-up

All 20 unresolved records received an access or corroboration follow-up. One original body was recovered: CNN’s Davis School District story. Nineteen selected records remain unresolved. Readable coverage elsewhere was retained as corroboration, without assuming it was identical to an inaccessible selected article.

Davis, Utah: a documented reinstatement decision

The district’s June 20, 2023 minutes record a 7–0 vote to reinstate the King James Bible at all school levels where it had been removed. The minutes describe eight copies, not eight ban instances. Public comments included residents and a school librarian. This establishes a board decision alongside public participation; it does not establish that opposition caused the decision.

The posted minutes are version 1 with blank signature fields; later approval was not separately checked. The record does not establish the initial removal date or when each copy returned to circulation. No duration is calculated. PEN row 3196 lists The Bible in Davis with a May 2023 month; CNN’s publication date is June 21. These dates describe different fields and remain separate.

District minutes, page 2 · CNN article. The official minutes and checksum are included in the research ZIP. The article text is not redistributed.

How the sample was selected

1. Freeze the current 1,054-record review ledger and record its SHA-256 checksum.
2. Group records by the month of their GDELT timestamp. This timestamp is not assumed to be the publication or restriction date.
3. Order each group by SHA-256 of a fixed seed and record ID, then select the first three records. Retain previous reviews, duplicate-looking stories and unavailable pages. Do not substitute easier records.
4. Read accessible article bodies, check geography against the official 2022–2023 NCES files, and classify each proposed district link separately.
5. Compare selected named titles with the preserved PEN index. Record differences in date, stage and scope without resolving them by assumption.

The fixed seed is banned-books-month-validation-v11-20261002. Monthly frame counts range from 29 to 247, so the equal-per-month sample is not proportional to news volume. The record-level CSV contains inclusion fractions and design weights, but this report gives sample counts only. No national accuracy estimate, recall estimate or confidence interval is presented. Independent human review remains pending.

Errors that a district-name search can make

What the PEN comparison established

The original 32 title probes produced 12 title–district correspondences. This follow-up adds a separate Bible–Davis correspondence and a primary board-decision record. These correspondences are not newly verified restriction onsets. Dearborn and Central Bucks reporting concerns stages that differ from the PEN entries. Portage lacks a verified onset date. Madison County, Mississippi, and Calvert County are absent from this preserved index; that absence cannot be recorded as zero bans.

For Indian River, both the reviewed article and PEN indicate March 2023. The earlier report incorrectly described PEN’s month as February; there is no month-level conflict in these two fields. Exact onset and classroom scope remain unverified. The Florida annual report corroborates the title and district without a month. For The Hill We Climb, PEN records March 2023, while the article refers to an April committee meeting. Their relationship remains unresolved.

Florida 2022–2023 objection report, page 9. The full official PDF, its checksum, and the row-level PEN comparisons are included in the research package.

Monthly review coverage

All 36 decisions

Each source link identifies the reviewed record. A failed or partial page remains unresolved even when its headline appears relevant. Encoded bodies were not decoded. Current-page text is not proof of its exact historical contents.

What this supports

The existing GDELT export contains useful school-book reporting alongside ambiguous names, public-library stories and unrelated keyword matches. Each district association needs article evidence. News timestamps alone cannot date a restriction. This sample adds validation evidence; it does not establish whether publicity caused bans or which state will act next.

RQ3: What happened when a district reviewed a challenged book with input from librarians and local residents?

Reporting in the sample describes different stages: proposals, ongoing reviews, conditional access, retained books and removals. These categories should remain distinct when answering RQ3. More complete primary access histories and independent review are needed before comparing restriction duration or outcomes.

Independent review

Use Independent_Review.html for a first pass without the earlier classifications, and export the completed JSON. No reviewer responses have been received. Comparison code checks completeness and lists disagreements; it does not establish who is correct or overwrite the research records.

Reproduce and review

Unzip the full package and run python run_all.py. The scripts review/build_sample_review.py and review/build_sample_followup.py recreate the fixed sample and apply the frozen follow-up. The final script, review/build_date_review.py, checks workbook cells and corrects the current prose. These scripts do not independently reproduce the judgment of reading an article. Use sample_review_v13.csv, sample_candidate_pairs_v12.csv and sample_title_probes_v13.csv to inspect the decisions.

Next evidence: resolve original-page access for the remaining 19 records, have a second reviewer check this fixed sample, and reconcile date/stage conflicts against district records. A separate sample from records outside the candidate set is needed to measure missed coverage.

Review version 13, October 4, 2026. No additional BigQuery query or paid service was used. Earlier versioned files are retained as audit history.