Return to the investigation

Inside Mike Green’s survey draft

Word documents store details you don’t see on the page, including how the file was created. Several details in the draft Mike emailed Emily match a test document made with docx, a tool that builds Word files from computer code.

Original email and file reopened October 26, 2026. The preserved original exactly matches the attachment in Mike’s August 4 email.

Why this matters: this draft became part of a proposal for a public school-library review. People should be able to ask who prepared it, what tools they used and who checked the result before it reached the Board.

The file details provide strong evidence that the document was assembled from code. Claude can use the same tools, but so can other software or a person writing code. These matches alone cannot tell us whether Claude was used or who wrote the words.

The wording gives us a more direct finding: RatedBooks appears in the survey Mike returned and stays in Emily’s circulated revision. The comparison below shows that change first.

RatedBooks appears in the returned survey

The original timeline names review journals and rubrics. The version Mike returned turns that into a questionnaire and names RatedBooks. Emily’s circulated revision keeps the reference.

  1. The staff timeline Emily sent

    August 4, 2026, 10:25 a.m.

    “professional review journals (e.g., School Library Journal, Kirkus), or rubrics”

    RatedBooks does not appear in the timeline’s extracted text.

    Read the original timeline

  2. The survey Mike returned

    August 4, 2026, 12:17 p.m.

    “Content-rating rubrics or sites (identify, e.g., RatedBooks.org)”

    RatedBooks is named in Table C-1.

    Read Mike’s email and draft

  3. The revision Emily circulated

    August 4, 2026, that afternoon

    “Content-rating rubrics or sites (identify, e.g., RatedBooks.org)”

    The same row remains in her revision.

    Read the circulated survey

These are text excerpts from different document versions, not tracked changes. Both survey versions offer “Used regularly,” “Used occasionally” and “Not used.” The response cells are blank. The records show the reference entering the draft; they do not show a district selecting it.

Check the extracted rows, attachment match and file hashes.

The returned file beside a code-generated control

The control was made with the docx library for the September 25 technical check. It is a test file, not a captured Claude conversation.

Last edited by

Mike’s returned draft

Un-named

Code-generated control

Un-named

Exact value match

Revision

Mike’s returned draft

1

Code-generated control

1

Exact value match

Application details

Mike’s returned draft

Empty

Code-generated control

Empty

Both application-properties elements are empty

Contents field

Mike’s returned draft

TOC \h \o "1-2"

Code-generated control

TOC \h \o "1-2"

Exact field match

Update fields on opening

Mike’s returned draft

Enabled

Code-generated control

Enabled

Same setting

Inspect the original metadata

These are text extracted from the original file package, not a screenshot or an AI-detector result.

<?xml version="1.0" encoding="UTF-8" standalone="yes"?><cp:coreProperties xmlns:cp="http://schemas.openxmlformats.org/package/2006/metadata/core-properties" xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:dcterms="http://purl.org/dc/terms/" xmlns:dcmitype="http://purl.org/dc/dcmitype/" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"><dc:title>School Library Materials Acquisition and Procurement Review — Draft Survey Instrument</dc:title><dc:creator>Draft prepared for Board Member Emily Green</dc:creator><cp:lastModifiedBy>Un-named</cp:lastModifiedBy><cp:revision>1</cp:revision><dcterms:created xsi:type="dcterms:W3CDTF">2026-08-04T18:12:53.212Z</dcterms:created><dcterms:modified xsi:type="dcterms:W3CDTF">2026-08-04T18:12:53.231Z</dcterms:modified></cp:coreProperties>
Original core-properties XMLOriginal application-properties XMLRechecked comparison data and SHA-256 hashes

Where the Claude connection comes from

The local comparison also checked five formatting instructions in a saved copy of Claude’s Word-file guide. The draft follows each of these instructions.

FeatureSaved Claude instructionFound in Mike’s draft
US Letter page sizeSet explicit dimensions instead of using the library’s default size.The page dimensions are 12240 × 15840 in Word’s internal units.
Table widthsDefine the table grid and cell widths in DXA units.11 tables, 11 grids and 464 cell-width entries; every cell-width entry uses dxa.
Table shadingUse clear shading.All 49 shaded cells use clear.
Bulleted listsUse the numbering configuration for bullets.10 bullet formats; no literal bullet characters in document text.
HeadingsUse the built-in heading styles for the contents table.16 Heading1 and 13 Heading2 assignments.

The guide was saved September 25, 2026. The August 4 version was not checked. These are also valid ways for other software to use the same library.

Four minutes from file creation to email

The file records creation at 12:12:53 p.m. Mountain daylight time on August 4, 2026. Mike’s email was sent at 12:17:09 p.m., four minutes and sixteen seconds later.

That dates the file package and its transmission. It does not measure how long the contents took to research, draft or revise.

What would establish Claude use?

A saved Claude conversation, a generation log or a statement from Mike identifying the tool would distinguish Claude from other ways of producing the same file. None was located in the records checked for this view.

Read Mike’s email and the public attachment. The metadata shown here comes from the preserved original, not from an assumption that a later redacted copy retained its properties.

Based on the September 25 local comparison, COMPARISON-mike-draft-vs-docxjs-and-claude-skill.md. The raw metadata, counts and attachment match were checked again for this view.