BANNED BOOKS REPRODUCIBILITY CORRECTION
30 September 2026

This notice supersedes any implication that the prior package reproduces every
finding in the report. It does not establish the cause of Barbara's discrepancy,
because her comparison data and output have not been supplied.

Corrections made
1. Added an offline, one-command review with source-row-level counting outputs.
2. Distinguish event rows, nonblank-title rows and unique titles.
3. Preserve differing sheet layouts and compare complete event multisets.
4. Do not concatenate sheet views or silently remove records.
5. Report both denominators for the top-three state share.
6. Share one workbook reader between reconciliation and forecast charts.
7. Correct generic provenance labels and include the original-report dependency
   needed to rebuild the historical affiliation figure.
8. Supply tests, readable code, execution results and file checksums.

Evidence status
2022-2023 workbook counts: reproducible from the retained workbook, with one
blank-title event explicitly reported. Retaining it is a documented row-count
policy, not confirmation of its missing title.
2024-2025 findings: arithmetic on supplied summary tables, not independent
event-level replication.
Historical regression, affiliations and survey results: not reproduced from
original inputs; required source files remain missing.
QGIS project and prior reports: retained as earlier submissions, not newly
validated by the Python tests.

No new forecast, causal claim or reconstructed survey data is supplied. The
website has not been changed. Passing software tests is not proof that all
source data or historical research conclusions are correct.
