Close the DocDown.Office artifact gap and run the full 125-set review campaign

[Malcolmnixon]

What this fixes

The DocDownOffice system and its Com subsystem were added to the SysML2 model pointing
at requirements, design and verification documents that were never written. Nothing validates model
document references — sysml2tools lint checks structure, not whether a cited path exists — so lint
stayed clean while the newest system in the repository had no requirements of its own.

That gap is closed, and the full formal review campaign across all 125 review-sets has now been
run on top of it.

What is here

Five new documents:

File	What it states
docs/reqstream/docdown-office.yaml	Registration is explicit and reflection-free; each format carries a probe authored in its own application
docs/reqstream/docdown-office/com.yaml	Availability is answered without launching Office; a delegate's rendering statement is suppressed when the COM run rendered
docs/design/docdown-office/com.md	Why the three shared types exist, and why ComDispatch is deliberately not shared
docs/verification/docdown-office.md	System verification: exact-set registration, probe presence
docs/verification/docdown-office/com.md	Com verification: the unavailable path is what CI exercises

Review campaign

125 of 125 review-sets reviewed. 68 defects found and fixed.

Area	Sets	Defects
Office	60	42
Core	18	21
Repository-wide	2	4
OTS	20	1
Pdf / Pdf.Rendering	15	0
Tool	10	0

Essentially every defect was the same species: prose that had stopped matching the code. Counts that
no longer counted the right things, cited tests that had been renamed or merged, option lists that
still named options which had been removed, and layout diagrams describing a folder tree that had
since been restructured. None were code defects — which is the point of running the campaign after
the consolidation rather than before it.

Verification

• All cited tests checked against three sources before being accepted: the xUnit method, the
  .fileassert.yaml entry, and the docdown --validate self-validation case. Citing tests that do
  not exist has been this repository's most repeated defect, so this was checked rather than assumed.
• Requirements 589 → 593, satisfied 539 → 543.
• build.ps1 — 1937 tests, 0 failed, none skipped.
• lint.ps1 — clean.
• reviewmark --plan — every file covered by a review-set.
• Every model reference — source, design, verification, requirements — resolves. 0 dangling, both
  directions.

Also corrected

Stale claims found while in the area:

• 41 documents still named test projects that were merged into DemaConsulting.DocDown.Office.Tests.
• Six verification documents were headed "Fixtures are generated, never committed", which
  stopped being true when the self-test probes were committed. They now distinguish the suite's
  generated fixtures — which must cover structures one document cannot — from the one committed probe
  per backend.
• 46 dead Pandoc input paths across two definition.yaml files, which were failing the docs
  build with withBinaryFile: does not exist. Nothing validates those paths either.

[Copilot]

🟡 Changes recommended

Unresolved traceability, document-structure, and verification-description issues remain.

Get a fresh assessment by requesting another Copilot review.

Pull request overview

Adds requirements, design, and verification artifacts for DocDown.Office and its Com subsystem.

Changes:

• Registers new Office and COM documentation.
• Updates traceability and review-set definitions.
• Corrects test-project references and fixture descriptions.

File summaries

File	Summary
requirements.yaml	Adds Office requirements.
docs/verification/docdown-pdf.md	Updates PDF fixture guidance.
docs/verification/docdown-pdf-rendering.md	Updates PDF rendering fixture guidance.
docs/verification/docdown-office/word/word-doc-down-builder-extensions.md	Corrects test-project reference.
docs/verification/docdown-office/word/open-xml/word-open-xml-reader.md	Corrects test-project reference.
docs/verification/docdown-office/word/open-xml/word-open-xml-image-reader.md	Corrects test-project reference.
docs/verification/docdown-office/word/open-xml/word-open-xml-extractor.md	Corrects test-project reference.
docs/verification/docdown-office/word/open-xml.md	Corrects test-project reference.
docs/verification/docdown-office/word/markdown/word-table-writer.md	Corrects test-project reference.
docs/verification/docdown-office/word/markdown/word-markdown-writer.md	Corrects test-project reference.
docs/verification/docdown-office/word.md	Updates test-project and fixture documentation.
docs/verification/docdown-office/visio/visio-doc-down-builder-extensions.md	Corrects test-project reference.
docs/verification/docdown-office/visio/open-xml/visio-package-reader.md	Corrects test-project reference.
docs/verification/docdown-office/visio/open-xml/visio-image-reader.md	Corrects test-project reference.
docs/verification/docdown-office/visio/markdown/visio-shape-labeler.md	Corrects test-project reference.
docs/verification/docdown-office/visio/markdown/visio-content-emitter.md	Corrects test-project reference.
docs/verification/docdown-office/visio/markdown.md	Corrects test-project reference.
docs/verification/docdown-office/visio/com/visio-com-extractor.md	Corrects test-project reference.
docs/verification/docdown-office/visio/com/visio-com-availability.md	Corrects test-project reference.
docs/verification/docdown-office/visio/com.md	Corrects test-project reference.
docs/verification/docdown-office/visio.md	Updates test-project and fixture documentation.
docs/verification/docdown-office/powerpoint/powerpoint-doc-down-builder-extensions.md	Corrects test-project reference.
docs/verification/docdown-office/powerpoint/open-xml/powerpoint-open-xml-reader.md	Corrects test-project reference.
docs/verification/docdown-office/powerpoint/open-xml/powerpoint-open-xml-image-reader.md	Corrects test-project reference.
docs/verification/docdown-office/powerpoint/open-xml/powerpoint-open-xml-extractor.md	Corrects test-project reference.
docs/verification/docdown-office/powerpoint/open-xml.md	Corrects test-project reference.
docs/verification/docdown-office/powerpoint/markdown/powerpoint-content-emitter.md	Corrects test-project reference.
docs/verification/docdown-office/powerpoint/markdown.md	Corrects test-project reference.
docs/verification/docdown-office/powerpoint/com/powerpoint-com-extractor.md	Corrects test-project reference.
docs/verification/docdown-office/powerpoint/com/powerpoint-com-availability.md	Corrects test-project reference.
docs/verification/docdown-office/powerpoint/com.md	Corrects test-project reference.
docs/verification/docdown-office/powerpoint.md	Updates test-project and fixture documentation.
docs/verification/docdown-office/excel/open-xml/excel-open-xml-reader.md	Corrects test-project reference.
docs/verification/docdown-office/excel/open-xml/excel-open-xml-image-reader.md	Corrects test-project reference.
docs/verification/docdown-office/excel/open-xml/excel-open-xml-extractor.md	Corrects test-project reference.
docs/verification/docdown-office/excel/open-xml/excel-drawing-text-reader.md	Corrects test-project reference.
docs/verification/docdown-office/excel/open-xml/excel-chart-reader.md	Corrects test-project reference.
docs/verification/docdown-office/excel/open-xml.md	Corrects test-project reference.
docs/verification/docdown-office/excel/markdown/excel-content-emitter.md	Corrects test-project reference.
docs/verification/docdown-office/excel/markdown/excel-chart-writer.md	Corrects test-project reference.
docs/verification/docdown-office/excel/markdown.md	Corrects test-project reference.
docs/verification/docdown-office/excel/excel-doc-down-builder-extensions.md	Corrects test-project reference.
docs/verification/docdown-office/excel.md	Updates test-project and fixture documentation.
docs/verification/docdown-office/com.md	Adds COM subsystem verification design.
docs/verification/docdown-office.md	Adds Office system verification design.
docs/verification/definition.yaml	Registers verification documents.
docs/reqstream/docdown-office/powerpoint.yaml	Updates test-project documentation.
docs/reqstream/docdown-office/com.yaml	Adds COM requirements.
docs/reqstream/docdown-office.yaml	Adds Office requirements.
docs/design/docdown-office/com.md	Adds COM subsystem design.
docs/design/definition.yaml	Registers the design document.
.reviewmark.yaml	Adds artifacts to review sets.

Review details

Suppressed comments (7)

.reviewmark.yaml:734

• This new subsystem requirement is added to a review-set whose context contains only the parent design. The reviewmark standard requires subsystem reviews to include both the parent system design and parent system requirements; add docs/reqstream/docdown-office.yaml to the context so this decomposition is reviewed against its parent.

      - "docs/reqstream/docdown-office/com.yaml"                                       # subsystem requirements

docs/design/docdown-office/com.md:3

• This new design document links to OfficeComView.svg, but docs/sysml2/views/design-views.sysml defines no OfficeComView and no such SVG is present. The rendered design therefore has a broken diagram link; add the corresponding view and generated artifact, or remove this image reference.

![DocDown.Office Com Structure](OfficeComView.svg)

docs/reqstream/docdown-office.yaml:16

• The without reflection or assembly scanning clause is an implementation constraint, not observable requirement behavior, and the listed tests can only verify the registered set and chaining. Keep the requirement focused on the explicit registration contract and document the reflection-free mechanism in the design/risk controls instead.

          The system shall register every Office backend it ships through one explicit call, without
          reflection or assembly scanning.

docs/reqstream/docdown-office.yaml:43

• The requirement promises probes authored by the producing applications, but its only cited test, Probe_EmbeddedResource_LoadsAsAnOpenPackage, loads bytes and checks only non-empty content plus the PK header. That evidence cannot distinguish an application-authored readable probe from a synthesized or malformed package; add an extraction/authorship-level check or narrow the requirement to embedded package presence.

      - id: DocDownOffice-SelfTestProbes
        title: >-
          The system shall carry, for each format it reads, a probe document authored in the
          application that produces that format.
        tags: [system]
        justification: |
          A self-test exists to answer whether this deployment can read what the real application
          emits. A backend that synthesized its own document could only ever prove that a library
          agreed with itself, and would keep reporting success while failing on every genuine
          document. Carrying the probe inside the package also means the check does not depend on
          the file system, the working directory, or anything the host happens to have installed.
        tests:
          - Probe_EmbeddedResource_LoadsAsAnOpenPackage

docs/verification/docdown-office.md:10

• The additional root-level tests are OfficeDocDownBuilderExtensionsTests and OfficeProbeResourceTests, which are package/unit assertions rather than a DocDownOffice system-integration test suite; the format system tests are already named in the preceding clause. Calling these tests system-level overstates the verification level. Describe them as package-level tests or add a true system-level test.

`DocDown.Office` is verified through the per-format system tests described in the Word, Excel,
PowerPoint, and Visio subsystem verification documents, plus the system-level tests in
`DemaConsulting.DocDown.Office.Tests`, running on xUnit v3 across net8.0, net9.0, and net10.0.

docs/verification/docdown-office.md:17

• The final clause is grammatically incorrect and does not identify what the four calls register. This should say that AddOffice() must register the same backends as those four calls.

`AddOffice()` stands in for four separate calls, so it is only safe if it registers what those four
registered. The test asserts the **whole set** of extractor identifiers, not that particular ones are

docs/verification/docdown-office/word.md:40

• No fixture is committed. immediately contradicts the next sentence's committed self-test binary and makes the new heading misleading. Say that no generated test fixture is committed, then keep the explicit exception for the embedded probe.

• Files reviewed: 52/52 changed files
• Comments generated: 12
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

🟡 Changes recommended

Unresolved requirement, design, and verification inconsistencies remain.

Get a fresh assessment by requesting another Copilot review.

Review details

Suppressed comments (16)

Previously missed (2) — in code that hasn't changed since the last review.

docs/reqstream/docdown-office/com.yaml:35

• This requirement says the contradictory fact is suppressed only when the COM backend has rendered pages, but ComposingDelegatedSink filters the matching unavailable fact during the delegated run, before RenderPagesAsync runs and without receiving any render-success state. If every page render fails, the fact is still suppressed (and the extractor still records its renderer fact); either gate suppression and the authoritative fact on actual rendered pages or weaken this requirement and add the failure case to verification.
  docs/verification/docdown-office/word.md:43
• This sentence says no fixture is committed and is immediately followed by the statement that one committed binary is the self-test probe. That is internally contradictory; the generated test fixtures are not committed, but the embedded probe is. Please qualify the sentence as applying to generated fixtures.

docs/design/docdown-office/com.md:1

• This is a subsystem companion under docs/design/docdown-office/; sibling subsystem designs use a ## top heading (for example docs/design/docdown-office/powerpoint/com.md:1). Using # makes this artifact look like a second system document and breaks the established heading hierarchy.

# DocDown.Office Com Subsystem Design

docs/reqstream/docdown-office.yaml:25

• DocDownOffice-Com-Availability is unrelated to registration, so making it a child of DocDownOffice-Registration misrepresents the requirement decomposition. System registration requirements in the existing Office requirement files decompose into registration-unit requirements; remove this cross-behavior child link or add a dedicated system requirement for availability.

        children:
          - DocDownOffice-Com-Availability

docs/reqstream/docdown-office.yaml:43

• The requirement says each probe is authored in its producing application, but Probe_EmbeddedResource_LoadsAsAnOpenPackage only checks that bytes are non-empty and begin with PK\x03\x04 (the verification text explicitly calls this a packaging assertion). That test cannot establish the authoring/application-specific content; cite the four format self-test cases or weaken the requirement to embedded-package presence.

        tests:
          - Probe_EmbeddedResource_LoadsAsAnOpenPackage

docs/reqstream/docdown-office/com.yaml:25

• DocDownOffice-Com-DelegatedComposition is an independent behavior, not a decomposition of availability; nesting it under DocDownOffice-Com-Availability gives the ReqStream hierarchy the wrong meaning. Keep the two requirements as siblings under Com Requirements, as the existing PowerPoint and Visio Com requirement files do.

            children:
              - DocDownOffice-Com-DelegatedComposition

docs/verification/docdown-office.md:62

• The test sorts engine.Extractors before comparing it, so it proves an exact set but cannot prove an ordered set or registration order. Either describe this as an exact set or stop sorting if order is part of the verification claim.

Builds an engine through `AddOffice()` and asserts the ordered set of extractor identifiers is
exactly `excel-openxml`, `powerpoint-com`, `powerpoint-openxml`, `visio-com`, `visio-openxml`, and
`word-openxml`. Evidence for `DocDownOffice-Registration`.

docs/verification/docdown-office.md:83

• Probe_EmbeddedResource_LoadsAsAnOpenPackage only checks non-empty bytes and the PK\x03\x04 prefix; an arbitrary or truncated blob with that prefix would pass. That does not establish the acceptance criterion that all four probes are Open XML packages, so validate the archive/package structure or weaken the criterion and wording.

**Test**: `Probe_EmbeddedResource_LoadsAsAnOpenPackage`

Loads each of the four embedded probes from the shipped assembly and asserts it is non-empty and
begins with the Zip signature. Evidence for `DocDownOffice-SelfTestProbes`.

docs/verification/docdown-office.md:17

• The sentence is grammatically incomplete: what those four registered should identify the calls and use the conditional verb.

registered. The test asserts the **whole set** of extractor identifiers, not that particular ones are

docs/verification/docdown-office/com.md:53

• The acceptance criterion repeats a condition the implementation does not enforce: the sink suppresses the delegate's unavailable page-rendering fact before rendering is attempted, and the tests cover only the all-pages-rendered path. Add a failed-render scenario and make suppression/pages.renderer reflect whether at least one page actually rendered, or change the criterion to match the current behavior.

- A delegated managed run's "rendering not provided" fact is suppressed when the COM run rendered,
  and nothing else the delegate wrote is altered.

docs/verification/docdown-office/com.md:1

• This is a subsystem verification companion under docs/verification/docdown-office/; the neighboring subsystem documents use a ## top heading (for example docs/verification/docdown-office/powerpoint/com.md:1). Using # breaks the established document hierarchy.

# DocDown.Office Com Verification Design

docs/verification/docdown-office/excel.md:41

• The new heading says a probe is committed, but the following paragraph still says no binary .xlsx or .xls is committed. src/DemaConsulting.DocDown.Office/Resources/probe.xlsx is an embedded committed binary; distinguish generated fixtures from that probe.

### Test fixtures are generated; the self-test probe is committed

docs/verification/docdown-office/powerpoint.md:52

• The new heading says a probe is committed, but the following paragraph still says no binary .pptx or .ppt is committed. src/DemaConsulting.DocDown.Office/Resources/probe.pptx is an embedded committed binary; distinguish generated fixtures from that probe.

### Test fixtures are generated; the self-test probe is committed

docs/verification/docdown-office/visio.md:52

• The new heading says a probe is committed, but the following paragraph still says no binary .vsdx or .vsd is committed. src/DemaConsulting.DocDown.Office/Resources/probe.vsdx is an embedded committed binary; distinguish generated fixtures from that probe.
  docs/verification/docdown-pdf-rendering.md:44
• The new heading claims a committed probe, but this section still says every PDF used by the suite is generated and never explains the embedded probe. The rendering package does ship Resources/probe.pdf; state that only generated test fixtures are excluded from the repository so the document does not contradict the package contents.
  docs/verification/docdown-pdf.md:27
• This new heading says a self-test probe is committed, but the paragraph below still says no binary PDF is committed. src/DemaConsulting.DocDown.Pdf/Resources/probe.pdf is embedded and loaded by PdfDocumentExtractor, so the verification design is contradictory; distinguish generated test fixtures from the committed probe.

• Files reviewed: 66/66 changed files
• Comments generated: 3
• Review effort level: Lite

[Copilot]

🟡 Changes recommended

Unresolved documentation and traceability inconsistencies remain across the model, designs, requirements, and verification artifacts.

Get a fresh assessment by requesting another Copilot review.

Review details

Suppressed comments (29)

Previously missed (2) — in code that hasn't changed since the last review.

docs/verification/docdown-office.md:18

• The sentence says AddOffice() must register “what those four registered,” which is grammatically unclear and leaves the referent implicit. State explicitly that the method must register the same backends as the four per-format calls.
  test/DemaConsulting.DocDown.Core.Tests/Detection/FormatSnifferTests.cs:213
• The renamed test now checks only Basis, but this file still describes “basis-and-confidence reporting” at the class level and repeatedly mentions 0.9 or zero confidence in assertion comments. Update those narratives so the test documentation matches the current record, which has no confidence value.

docs/design/docdown-core/detection.md:56

• This design now documents FormatDetection as format plus basis, but the SysML companion at docs/sysml2/model/docdown-core/detection.sysml:31-37 still says it carries a confidence score. Update the model artifact in the same change so the architecture model and design contract do not disagree.

- **`DetectionBasis`** — `Extension` or `ContentSignature`.
- **`FormatDetection`** — the identified format and the basis it was identified by.

docs/design/docdown-core/detection/format-sniffer.md:28

• This changed detection design still describes confidence scores in its algorithm bullets at lines 39-43 (0.9, 1.0, and 0.0), even though the current FormatDetection no longer has that field and the verification/requirements artifacts now remove it. Update those bullets as part of this document change so the design is internally consistent.

The extension table maps `.pdf`, `.docx`/`.doc`, `.xlsx`/`.xls`, `.pptx`/`.ppt`, `.vsdx`/`.vsdm`/`.vsd`,
`.html`/`.htm`, and `.txt`/`.text`/`.log` to their formats — the `.doc`, `.xls`, `.ppt`, and `.vsd`
entries are the legacy binary Office formats, detectable but not extractable. Recognizing an
extension here does not imply an extractor is registered for it — that is `ExtractorSelector`'s

docs/design/docdown-office/com.md:1

• This is a subsystem design document, but its heading hierarchy is one level too high: existing subsystem designs use ## Com Subsystem followed by ### sections (for example docs/design/docdown-office/powerpoint/com.md:1-5), as required by .github/standards/design-documentation.md:59-66. Change the root and section headings so the generated documentation structure remains consistent.

# DocDown.Office Com Subsystem Design

docs/design/docdown-office/excel/open-xml/excel-open-xml-extractor.md:70

• The replacement of SpreadsheetDocument.Create left the wrapped noun on the next line, so this now reads “the managed reader this backend is built on workbook.” Complete the sentence with the embedded probe instead.

- **DocumentFormat.OpenXml** (OTS) — the managed reader this backend is built on
  workbook.

docs/design/docdown-office/powerpoint/com/powerpoint-com-availability.md:24

• “Use in common” is not idiomatic US English here; the shared helper is something PowerPoint and Visio share.

The probe itself lives in the shared `OfficeComAvailability`, which PowerPoint and Visio use in
common: the two probes differed only in the application name and the ProgID, so those are arguments

docs/design/docdown-office/powerpoint/com/powerpoint-com-extractor.md:56

• The new ProbeResourceName bullet says the render case uses an embedded .pptx, but the preceding RunRender description still says it builds a synthetic deck in a work folder and deletes that deck. After this change the case should render the embedded probe (and must not describe a synthesized temporary input). The companion SysML extractor description also still says synthetic.

- **`ProbeResourceName`** (private const) — names the embedded deck the render case rasterizes: a
  real `.pptx` authored in Microsoft PowerPoint, read through `SelfTestProbe.Load`. The synthesizer
  that once stood here built a theme, color scheme, font scheme, format scheme, slide master and
  layout, because PowerPoint refuses to open a deck without them.

docs/design/docdown-office/powerpoint/open-xml/powerpoint-open-xml-extractor.md:69

• The replacement of PresentationDocument.Create left the wrapped noun on the next line, so this now reads “the managed reader this backend is built on deck.” Complete the sentence with the embedded probe instead.

- **DocumentFormat.OpenXml** (OTS) — the managed reader this backend is built on
  deck.

docs/design/docdown-office/visio/com/visio-com-availability.md:23

• “Use in common” is not idiomatic US English here; the shared helper is something Visio and PowerPoint share.

The probe itself lives in the shared `OfficeComAvailability`, which Visio and PowerPoint use in
common: the two probes differed only in the application name and the ProgID, so those are arguments

docs/design/docdown-office/visio/com/visio-com-extractor.md:54

• The new ProbeResourceName bullet says the render case uses an embedded .vsdx, but the preceding visio.com.render description still says it builds a synthetic drawing in a work folder and deletes it. After this change the case should render the embedded probe (and must not describe a synthesized temporary input). The companion SysML extractor description also still says synthetic.

- **`ProbeResourceName`** (private const) — names the embedded drawing the render case rasterizes: a
  real `.vsdx` authored in Microsoft Visio, read through `SelfTestProbe.Load`. The synthesizer that
  once stood here had to write a window part and real shape geometry, because Visio refuses to open a
  package without the first and draws nothing for a shape without the second.

docs/design/docdown-office/word/open-xml.md:44

• Replacing the writer call left the sentence as “builds a embedded document.” The self-test now reads the probe, and the article should be “an” if this sentence is retained.

embedded document authored in Microsoft Word, reads it with the reader, and passes

docs/reqstream/docdown-office.yaml:25

• DocDownOffice-Com-Availability is not a refinement of DocDownOffice-Registration: its title and justification govern environment probing, not explicit backend registration. This children link makes the requirement tree report availability under registration while leaving the format-registration decomposition unrepresented; move it under an appropriate parent or make it a separate root requirement, following the focused registration children used in docs/reqstream/docdown-pdf.yaml:67-74.

        children:
          - DocDownOffice-Com-Availability

docs/reqstream/docdown-office.yaml:29

• AddOffice_NullBuilder_Throws is an acceptance criterion and test in docs/verification/docdown-office.md, but this registration requirement cites only the other three tests. Without a citation here, the new system traceability omits one of its stated acceptance tests; add the test or model a child requirement for null rejection.

          - AddOffice_OnBuilder_RegistersEveryOfficeBackend
          - AddOffice_ReturnsSameBuilderForChaining
          - AddExcel_Alone_RegistersOnlyTheExcelBackend

docs/reqstream/docdown-office/com.yaml:23

• This justification first says the reader can act on the probe's reason and then immediately says the reader cannot act on it. Rephrase the sentence so it consistently describes an environmental fact rather than an installation instruction.

              registry, never throws, and answers with a declarative reason the reader can act on -
              or rather, one the reader cannot act on, which is why the reasons state a fact about
              the environment instead of instructing an installation.

docs/verification/docdown-office/com.md:1

• This subsystem verification document uses a # root and ## sections, unlike the repository's subsystem verification layout, which uses a ## root and ### sections (see docs/verification/docdown-office/powerpoint/com.md:1-6 and .github/standards/design-documentation.md:59-66). Adjust the heading levels throughout this new document so verification sections are nested correctly.
  docs/verification/docdown-office/excel.md:52
• After acknowledging the embedded probe, this paragraph still says the repository stays text-only. probe.xlsx is a committed binary, so limit this statement to the generated test fixtures; otherwise the verification claim is false.
  docs/verification/docdown-office/powerpoint.md:64
• The new paragraph says no .pptx is committed, but this package now commits Resources/probe.pptx for the self-test. Restrict the statement to test fixtures; as written, the verification document contradicts its preceding probe description.
  docs/verification/docdown-pdf-rendering.md:44
• The new heading says a self-test probe is committed, but the following paragraph still says the repository stays text-only and never identifies the committed probe. PdfPageRenderingExtractor loads the embedded Resources/probe.pdf, so this verification document contradicts the packaging it is meant to describe; distinguish generated test fixtures from the committed self-test probe.
  docs/verification/docdown-pdf.md:34
• The new paragraph says a probe binary is committed, but this document's Test Environment still says “no committed binaries” at line 47. Update that input description to distinguish generated test fixtures from the embedded self-test probe; otherwise the verification design contradicts itself.
  src/DemaConsulting.DocDown.Core/Output/ContentWriter.cs:214
• ContentPart currently exposes only Sheet and Chart kinds, so the unchanged summary's example of a slide is inaccurate. Update the summary with this parameter description; otherwise the public API docs describe a kind the enum does not provide.
  src/DemaConsulting.DocDown.Office/Excel/OpenXml/ExcelOpenXmlExtractor.cs:147
• Although this lower remark now says the case reads an embedded Excel workbook, the GetSelfTestCases remark above still says it reads a workbook “it builds itself” (lines 112-114). Update that earlier contract comment too, or the self-test documentation remains contradictory.
  src/DemaConsulting.DocDown.Office/PowerPoint/OpenXml/PowerPointOpenXmlExtractor.cs:136
• Although this lower remark now says the case reads an embedded PowerPoint deck, the GetSelfTestCases remark above still says it reads a deck “it builds itself” (lines 101-103). Update that earlier contract comment too, or the self-test documentation remains contradictory.
  src/DemaConsulting.DocDown.Office/Visio/OpenXml/VisioOpenXmlExtractor.cs:135
• This implementation now loads SelfTestProbe, but docs/design/docdown-office/visio/open-xml/visio-open-xml-extractor.md:32-35 and :51 still describe a VisioPackageBuilder-generated self-test drawing. Update that companion design artifact so the new application-authored probe is documented consistently.
  src/DemaConsulting.DocDown.Office/Visio/OpenXml/VisioOpenXmlExtractor.cs:135
• Although this lower remark now says the case reads an embedded Visio drawing, the GetSelfTestCases remark above still says it reads a drawing “it builds itself” (lines 101-102). Update that earlier contract comment too, or the self-test documentation remains contradictory.
  src/DemaConsulting.DocDown.Office/Word/OpenXml/WordOpenXmlExtractor.cs:141
• The source now reads the embedded Word probe, but the shared OTS design still says DocumentFormat.OpenXml provides the writer used to build the parse-round-trip probe (docs/design/ots/openxml.md:12-14). Update that companion description so it no longer attributes the removed writer path to the self-test.
  src/DemaConsulting.DocDown.Office/Word/OpenXml/WordOpenXmlExtractor.cs:141
• Although this lower remark now says the case reads an embedded Word document, the GetSelfTestCases remark above still says it reads a document “it builds itself” (lines 103-105). Update that earlier contract comment too, or the public self-test documentation remains contradictory.
  src/DemaConsulting.DocDown.Pdf/PdfDocumentExtractor.cs:241
• The implementation now reads the embedded PDF probe, but its companion design still says GetSelfTestCases builds and rereads a one-page PDF in memory (docs/design/docdown-pdf/pdf-document-extractor.md:42-44). Update that design artifact in the same change so the published self-test contract matches the code.
  src/DemaConsulting.DocDown.Pdf/PdfDocumentExtractor.cs:241
• Although this lower remark now says the case reads an embedded PDF, the GetSelfTestCases remark above still says it parses a document “it builds itself” (lines 120-122). Update that earlier contract comment too, or the self-test documentation remains contradictory.

• Files reviewed: 94/94 changed files
• Comments generated: 5
• Review effort level: Lite

[Copilot]

🟡 Changes recommended

Final review comments identify unresolved documentation and requirement-structure inconsistencies.

Get a fresh assessment by requesting another Copilot review.

Review details

Suppressed comments (28)

Previously missed (3) — in code that hasn't changed since the last review.

docs/design/docdown-office/excel/open-xml/excel-open-xml-extractor.md:70

• The edit leaves the old wrapped noun on the next line, so the dependency now reads “the managed reader this backend is built on workbook.” Remove the orphaned workbook. and end the dependency description after “built on.”
  docs/design/docdown-office/powerpoint/open-xml/powerpoint-open-xml-extractor.md:69
• The edit leaves the old wrapped noun on the next line, so the dependency now reads “the managed reader this backend is built on deck.” Remove the orphaned deck. and end the dependency description after “built on.”
  docs/design/docdown-office/word/open-xml.md:44
• The preceding text still says the case “builds a” document, so this edit produces “builds a embedded document” and continues to describe the removed in-memory construction. Reword the two-line sentence to say that the case reads the embedded document.

docs/design/docdown-core/detection/format-sniffer.md:26

• Although this hunk updates the extension table, this design document still says the sniffer reports confidence and still documents extension results as Extension / 0.9 elsewhere. FormatDetection now contains only the format and basis, so this artifact remains inconsistent with the updated source and verification document; remove the confidence claims throughout the file.

The extension table maps `.pdf`, `.docx`/`.doc`, `.xlsx`/`.xls`, `.pptx`/`.ppt`, `.vsdx`/`.vsdm`/`.vsd`,
`.html`/`.htm`, and `.txt`/`.text`/`.log` to their formats — the `.doc`, `.xls`, `.ppt`, and `.vsd`

docs/design/docdown-office/com.md:1

• This new subsystem design document is nested under docs/design/docdown-office, but its top-level heading is level 1. Existing nested Office subsystem documents use a ## title, so this renders as a second system-level document instead of a child of DocDown.Office.

# DocDown.Office Com Subsystem Design

docs/design/docdown-office/com.md:5

• This is a subsystem design companion, but it uses a system-level # title and ## sections, and it does not provide the mandatory subsystem Overview, Interfaces, and Design sections. Sibling subsystem designs such as docs/design/docdown-office/powerpoint/com.md use a ## title with ### sections; align this new artifact with that hierarchy.

# DocDown.Office Com Subsystem Design

![DocDown.Office Com Structure](OfficeComView.svg)

The Com subsystem holds what the PowerPoint and Visio COM automation backends share. It exists

docs/design/docdown-office/excel/open-xml.md:38

• The preceding text still says the case “builds a”, so this edit produces “builds a embedded workbook” and leaves the self-test described as constructing a fixture. Reword the two-line sentence to say that it reads the embedded workbook.

embedded workbook authored in Microsoft Excel, reads it with the reader, and

docs/design/docdown-office/excel/open-xml/excel-open-xml-extractor.md:46

• These lines now contradict the embedded-probe description above: they still claim that building rather than embedding keeps the case free of a shipped payload, even though the self-test loads the embedded workbook. Replace this with the reason the application-authored probe is intentionally embedded.

    worksheet. Building rather than embedding a fixture keeps the case free of a shipped binary
    payload.

docs/design/docdown-office/powerpoint/open-xml.md:40

• The preceding text still says the case “builds a”, so this edit produces “builds a embedded deck” and leaves the self-test described as constructing a fixture. Reword the two-line sentence to say that it reads the embedded deck.

embedded deck authored in Microsoft PowerPoint, reads it with the reader, and passes

docs/design/docdown-office/powerpoint/open-xml/powerpoint-open-xml-extractor.md:44

• This sentence still claims that building rather than embedding keeps the case free of a shipped payload, contradicting the new embedded-deck description and ProbeResourceName implementation. Replace it with the reason the application-authored probe is intentionally embedded.

    slide. Building rather than embedding a fixture keeps the case free of a shipped binary payload.

docs/design/docdown-office/visio/open-xml/visio-package-reader.md:57

• The changed sentence leaves an ungrammatical construction: it says the extractor self-test reads the probe, then uses “and both ... and” to join the fixture clause. Split the extractor self-test from the generated-fixture statement.

`VisioOpenXmlExtractor.ExtractAsync` calls `Read()` on the buffered drawing stream, the
extractor's self-test reads the embedded probe drawing, and the test fixtures build packages
with `VisioPackageBuilder` in the test project for it to read back.

docs/reqstream/docdown-office.yaml:29

• The verification design lists AddOffice_NullBuilder_Throws as evidence for the registration criterion and the acceptance criteria explicitly require null rejection, but this requirement's test list omits it. Add the test here so that behavior remains traceable in ReqStream like the per-format registration requirements.

          - AddExcel_Alone_RegistersOnlyTheExcelBackend

docs/reqstream/docdown-office.yaml:25

• DocDownOffice-Registration is a registration requirement, but it decomposes into DocDownOffice-Com-Availability, which is a separate subsystem responsibility rather than a child of registration. Existing Office requirement files keep registration and COM availability as separate requirement branches (for example, docs/reqstream/docdown-office/powerpoint/com.yaml:15-52). Remove this child edge and keep the two requirements as siblings.

        children:
          - DocDownOffice-Com-Availability

docs/reqstream/docdown-office/com.yaml:25

• Availability and delegated-output composition are independent COM responsibilities, but DocDownOffice-Com-Availability currently makes DocDownOffice-Com-DelegatedComposition its child. The existing PowerPoint and Visio COM requirement files model availability, rendering, and automation as sibling branches; remove this false decomposition and keep both requirements at the subsystem level.

            children:
              - DocDownOffice-Com-DelegatedComposition

docs/verification/docdown-office.md:17

• This sentence says AddOffice() registers “what those four registered”, which is grammatically incomplete and obscures that the four calls are the subject. Use “what those calls register” before describing the exact-set assertion.
  docs/verification/docdown-office/com.md:1
• This new subsystem verification document is nested under docs/verification/docdown-office, but its top-level heading is level 1. Existing nested Office verification documents use a ## title, so this breaks the document hierarchy.
  docs/verification/docdown-office/excel.md:52
• The new paragraph still says the repository stays text-only, but probe.xlsx is a committed binary embedded by the Office project. Limit this claim to the generated test fixtures; otherwise the verification document contradicts the committed self-test probe described immediately above.
  docs/verification/docdown-office/powerpoint.md:64
• This new paragraph says no .pptx is committed and the repository stays text-only, but probe.pptx is a committed embedded resource. Limit the statement to the suite's generated fixtures so it does not contradict the self-test probe paragraph above.
  docs/verification/docdown-office/powerpoint/com/powerpoint-com-extractor.md:9
• This document is being updated for the consolidated test project, but its release-time paragraph still says docdown --validate builds a synthetic single-slide deck. The implementation now writes the embedded probe.pptx to the work folder, so the verification description remains false after this update.
  docs/verification/docdown-office/visio/com/visio-com-extractor.md:9
• This document is being updated for the consolidated test project, but its release-time paragraph still says docdown --validate builds a synthetic single-page drawing. The implementation now writes the embedded probe.vsdx to the work folder, so the verification description remains false after this update.
  docs/verification/docdown-pdf-rendering.md:44
• The heading now acknowledges a committed self-test probe, but the following paragraph still says every PDF is generated at test time and that the repository stays text-only. The backend now reads a committed embedded PDF probe, so distinguish that probe from the generated test fixtures here.
  src/DemaConsulting.DocDown.Office/Excel/OpenXml/ExcelOpenXmlExtractor.cs:147
• This updated remark correctly says the self-test reads an embedded Excel workbook, but the enclosing GetSelfTestCases remarks above still say it reads a workbook the backend builds itself. Update that outer description so the public XML documentation matches the shipped probe.
  src/DemaConsulting.DocDown.Office/PowerPoint/OpenXml/PowerPointOpenXmlExtractor.cs:136
• This updated remark correctly says the self-test reads an embedded PowerPoint deck, but the enclosing GetSelfTestCases remarks above still say it reads a deck the backend builds itself. Update that outer description so the public XML documentation matches the shipped probe.
  src/DemaConsulting.DocDown.Office/Visio/OpenXml/VisioOpenXmlExtractor.cs:134
• This new remark is grammatically broken: it ends with “joined by a glued” and the next line starts “it back”, so it reads as “glued it back” instead of describing the glued connector and the read operation. Please reword the two lines to describe the embedded drawing and the resolved connection.
  src/DemaConsulting.DocDown.Office/Visio/OpenXml/VisioOpenXmlExtractor.cs:135
• As with the other extractors, this new embedded-probe remark conflicts with the enclosing GetSelfTestCases remarks above, which still say the backend reads a drawing it builds itself. Update the outer description to match the embedded probe.
  src/DemaConsulting.DocDown.Office/Word/OpenXml/WordOpenXmlExtractor.cs:141
• This updated remark correctly says the self-test reads an embedded Word document, but the enclosing GetSelfTestCases remarks above still say it reads a document the backend builds itself. Update that outer description so the public XML documentation does not describe the old probe implementation.
  src/DemaConsulting.DocDown.Pdf/PdfDocumentExtractor.cs:241
• This updated remark correctly says the self-test reads an embedded PDF, but the enclosing GetSelfTestCases remarks above still say it parses a document the backend builds itself. The two descriptions now contradict each other; update the outer remark to describe the embedded probe as well.
  test/DemaConsulting.DocDown.Core.Tests/Detection/FormatSnifferTests.cs:213
• The renamed test no longer claims to verify confidence, but this file still describes confidence throughout, including the class summary and assertion comments (for example, the old 0.9 contract below). FormatDetection has no confidence member; remove the stale confidence claims so the tests do not document a deleted contract.

• Files reviewed: 94/94 changed files
• Comments generated: 4
• Review effort level: Lite

[Copilot]

🟡 Changes recommended

Unresolved moderate documentation and requirement-traceability inconsistencies remain.

Get a fresh assessment by requesting another Copilot review.

Review details

Suppressed comments (34)

Previously missed (8) — in code that hasn't changed since the last review.

docs/design/docdown-office/excel/open-xml/excel-open-xml-extractor.md:70

• The dependency bullet is now truncated to built on workbook, so the design document no longer states a valid dependency. End the sentence after built on (or explicitly say it is used by the self-test) instead of leaving the old workbook fragment behind.
  docs/design/docdown-office/powerpoint/open-xml/powerpoint-open-xml-extractor.md:69
• The dependency bullet is now truncated to built on deck, so the design document no longer states a valid dependency. End the sentence after built on (or explicitly say it is used by the self-test) instead of leaving the old deck fragment behind.
  docs/design/docdown-office/word/open-xml.md:44
• The updated sentence leaves the old builds a wording in place, producing builds a embedded and redundantly saying the document is read afterward. Describe the operation as reading the embedded Word probe instead.
  docs/verification/docdown-office.md:17
• The registration explanation has a grammatical error: what those four registered should use the present-tense verb register.
  src/DemaConsulting.DocDown.Office/Excel/OpenXml/ExcelOpenXmlExtractor.cs:145
• The lower remark now says this case reads an embedded Excel probe, but the earlier GetSelfTestCases summary still says it builds the workbook itself. Update that summary as well so the API documentation describes the actual self-test input.
  src/DemaConsulting.DocDown.Office/PowerPoint/OpenXml/PowerPointOpenXmlExtractor.cs:135
• The lower remark now says this case reads an embedded PowerPoint probe, but the earlier GetSelfTestCases summary still says it builds the deck itself. Update that summary too; otherwise the same public documentation gives two different accounts of the self-test.
  src/DemaConsulting.DocDown.Office/Word/OpenXml/WordOpenXmlExtractor.cs:139
• The lower remark now says this case reads an embedded Word probe, but this earlier GetSelfTestCases documentation still says it builds the document itself. That leaves the public self-test description internally contradictory; update the earlier summary to describe the embedded probe too.
  src/DemaConsulting.DocDown.Pdf/PdfDocumentExtractor.cs:239
• The lower remark now says the PDF case reads an embedded Word-exported probe, but the earlier GetSelfTestCases summary still says it builds a document itself. Update that earlier summary too, or the self-test documentation remains inconsistent with the implementation.

docs/design/docdown-core/detection/format-sniffer.md:25

• This file's algorithm list still reports 0.9, 1.0, and 0.0 confidence values at lines 39–43, contradicting the API change documented elsewhere in the PR that FormatDetection carries no confidence. Remove those score fields from the algorithm description too.

The extension table maps `.pdf`, `.docx`/`.doc`, `.xlsx`/`.xls`, `.pptx`/`.ppt`, `.vsdx`/`.vsdm`/`.vsd`,

docs/design/docdown-core/output/manifest-writer.md:74

• This PR updates the manifest documentation to say ManifestWriter runs after ContentWriter and before MetadataWriter, but the ContentWriter companion document still says the old order with ManifestWriter after MetadataWriter. Update the companion statement as part of the ordering change so the two design artifacts do not contradict each other.

`DocDownEngine` calls `ManifestWriter` after `ContentWriter` and before `MetadataWriter`.

docs/design/docdown-office/com.md:1

• This is a subsystem document nested under the DocDown.Office system, but it uses a level-one heading. Existing nested subsystem design documents use a level-two top heading (for example docs/design/docdown-office/visio/com.md:1); use ## here so the document hierarchy is valid.

# DocDown.Office Com Subsystem Design

docs/design/docdown-office/com.md:63

• The sentence ends with because it does, which has no grammatical antecedent and does not explain the cost claim. Remove the fragment or replace it with the intended rationale so this new design constraint is readable.

- The availability probe is cheap enough to run on every extraction, because it does.

docs/design/docdown-office/excel/open-xml.md:38

• The changed self-test description still says it builds a fixture, and the next sentence says that building avoids a shipped payload. The implementation now reads an embedded probe, so this paragraph both has the a embedded grammar error and contradicts the shipped-resource design.

embedded workbook authored in Microsoft Excel, reads it with the reader, and

docs/design/docdown-office/excel/open-xml/excel-open-xml-extractor.md:46

• The new probe wording is followed by the stale claim that building rather than embedding keeps the case free of a shipped binary. This self-test reads ProbeResourceName, so remove that claim and keep the description consistent with the embedded workbook.

  - `excel.openxml.parseRoundTrip`, which reads the embedded workbook authored in Microsoft Excel
    and passes when the model carries at least one
    worksheet. Building rather than embedding a fixture keeps the case free of a shipped binary
    payload.

docs/design/docdown-office/powerpoint/com/powerpoint-com-extractor.md:54

• The new ProbeResourceName entry is contradicted by the preceding RunRender description, which still says the case builds a synthetic deck and ships no fixture. The implementation writes the embedded probe to the work folder; update that earlier paragraph as well.

- **`ProbeResourceName`** (private const) — names the embedded deck the render case rasterizes: a
  real `.pptx` authored in Microsoft PowerPoint, read through `SelfTestProbe.Load`. The synthesizer

docs/design/docdown-office/powerpoint/open-xml.md:40

• The changed self-test description still says it builds a fixture, and the next sentence says that building avoids a shipped payload. The implementation now reads an embedded probe, so this paragraph both has the a embedded grammar error and contradicts the shipped-resource design.

embedded deck authored in Microsoft PowerPoint, reads it with the reader, and passes

docs/design/docdown-office/powerpoint/open-xml/powerpoint-open-xml-extractor.md:44

• The new probe wording is followed by the stale claim that building rather than embedding keeps the case free of a shipped binary. This self-test reads ProbeResourceName, so remove that claim and keep the description consistent with the embedded deck.

  - `powerpoint.openxml.parseRoundTrip`, which reads the embedded deck authored in Microsoft PowerPoint
    and passes when the model carries at least one
    slide. Building rather than embedding a fixture keeps the case free of a shipped binary payload.

docs/design/docdown-office/visio/com/visio-com-extractor.md:52

• The new ProbeResourceName entry is contradicted by the preceding RunRender description, which still says the case builds a synthetic drawing and ships no fixture. The implementation writes the embedded probe to the work folder; update that earlier paragraph as well.

- **`ProbeResourceName`** (private const) — names the embedded drawing the render case rasterizes: a
  real `.vsdx` authored in Microsoft Visio, read through `SelfTestProbe.Load`. The synthesizer that

docs/design/docdown-office/visio/open-xml/visio-open-xml-extractor.md:51

• The dependency list names SelfTestProbe, but the self-test description still says it builds a drawing at runtime. The implementation now reads the embedded Visio probe, so change this description to match the actual case.

- **`SelfTestProbe`** (Core) — loads the embedded drawing the self-test reads.

docs/reqstream/docdown-office.yaml:25

• The COM availability requirement is not a decomposition of the system registration requirement: backend registration and runtime availability are independent behaviors and are tested separately. Existing package-level requirement files keep registration and availability as sibling requirements; remove this child link rather than making availability appear to implement registration.

        children:
          - DocDownOffice-Com-Availability

docs/reqstream/docdown-office/com.yaml:25

• DocDownOffice-Com-DelegatedComposition is an independent COM behavior, not a decomposition of availability: the availability probe can succeed while composition is still wrong, and the two requirements have separate test sets. The existing PowerPoint and Visio COM requirement files model rendering, availability, and automation as sibling requirements; remove this children link so the new shared requirements have the same traceability.

            children:
              - DocDownOffice-Com-DelegatedComposition

docs/verification/docdown-office/com.md:1

• This is a subsystem verification document nested under the DocDown.Office system, but it uses a level-one heading. Existing nested subsystem verification documents use a level-two top heading (for example docs/verification/docdown-office/visio/com.md:1); use ## here.
  docs/verification/docdown-office/excel.md:49
• This section acknowledges a committed workbook probe, but it still says every workbook is generated and later says the repository stays text-only. The probe is a shipped .xlsx, so scope the generated statements to test fixtures and remove the text-only claim.
  docs/verification/docdown-office/powerpoint.md:61
• This section now says a probe binary is committed, but it also says every deck is generated and explicitly claims no .pptx is committed and the repository is text-only. The shipped probe makes those statements false; distinguish test fixtures from the probe throughout this paragraph.
  docs/verification/docdown-office/powerpoint/com/powerpoint-com-extractor.md:9
• The updated test-project reference leaves the release-time description at lines 100–102 saying docdown --validate builds a synthetic deck and ships no fixture. The COM self-test now writes the embedded PowerPoint probe; update that scenario so the verification document does not describe the old implementation.
  docs/verification/docdown-office/visio.md:52
• The new heading distinguishes fixtures from the probe, but this unchanged sentence still claims every drawing used by the suite is generated. The COM/Open XML self-tests read the embedded application-authored probe, so limit this claim to test fixtures.
  docs/verification/docdown-office/visio/com/visio-com-extractor.md:9
• The updated test-project reference leaves the release-time description at lines 96–98 saying docdown --validate builds a synthetic drawing and ships no fixture. The COM self-test now writes the embedded Visio probe; update that scenario so the verification document does not describe the old implementation.
  docs/verification/docdown-office/visio/open-xml/visio-open-xml-extractor.md:18
• This test-environment entry still says the self-test round-trip uses a drawing built at test time, although the implementation and the updated design now read the embedded Visio probe. Distinguish the embedded self-test input from the generated reader/integration fixtures.
  docs/verification/docdown-office/word.md:35
• The new heading distinguishes fixtures from the probe, but this unchanged sentence still claims every document used by the suite is generated. The self-test reads the embedded application-authored document, so this should say test fixture rather than document.
  docs/verification/docdown-pdf-rendering.md:44
• The heading now acknowledges a committed probe, but the following paragraph still says every PDF is generated and that the repository stays text-only. That makes the verification design contradict the embedded rendering probe; distinguish generated test fixtures from the committed self-test input here.
  docs/verification/docdown-pdf.md:27
• This sentence now says what is true of generated fixtures, but not of the newly committed self-test probe: the probe is also a document used by the self-test. Limit the claim to test fixtures so it does not contradict the next sentence.
  src/DemaConsulting.DocDown.Core/Output/ContentWriter.cs:214
• This update narrows the parameter description to the two current enum values, but the preceding summary still says a part can be a slide even though ContentPartKind only contains Sheet and Chart. Please update the summary at the same time so the public XML documentation is internally consistent.
  src/DemaConsulting.DocDown.Office/Visio/OpenXml/VisioOpenXmlExtractor.cs:134
• The sentence is malformed after the probe change: glued is left hanging and it back is the remainder of the removed "reads it back" phrase. This XML comment should describe the embedded drawing and the topology assertion without the orphaned text.
  test/DemaConsulting.DocDown.Core.Tests/Detection/FormatSnifferTests.cs:213
• This test was renamed to assert only the detection basis, but the file still documents and comments a confidence contract in the class remarks and several assertions, including the new assertion comment below. FormatDetection no longer has confidence, so remove those remaining claims throughout this test file; otherwise its documentation describes an API that no longer exists.

• Files reviewed: 94/94 changed files
• Comments generated: 1
• Review effort level: Lite

[Copilot]

🟡 Changes recommended

Unresolved documentation and traceability inconsistencies remain across the reviewed design, requirements, model, and source comments.

Get a fresh assessment by requesting another Copilot review.

Review details

Suppressed comments (23)

Previously missed (4) — in code that hasn't changed since the last review.

docs/design/docdown-core/detection.md:56

• This lead-in says the supporting value types remain unchanged, but the following bullets document the changed FormatDetection contract after confidence was removed. The section now contradicts itself; change the lead-in to introduce the current type definitions.
  docs/design/docdown-office/excel/open-xml/excel-open-xml-extractor.md:70
• The replacement leaves the old workbook. fragment on the next line, so the dependency entry now reads “the managed reader this backend is built on workbook.” Remove the orphaned fragment and finish this bullet as a complete dependency description.
  docs/design/docdown-office/powerpoint/open-xml/powerpoint-open-xml-extractor.md:69
• The replacement leaves the old deck. fragment on the next line, so the dependency entry now reads “the managed reader this backend is built on deck.” Remove the orphaned fragment and finish this bullet as a complete dependency description.
  docs/design/docdown-office/word/open-xml.md:44
• This sentence still says the parse case “builds a embedded” document. The case now reads the shipped probe, so the article is grammatically wrong and the verb contradicts the implementation.

docs/design/docdown-office/excel/open-xml.md:38

• This sentence still says the parse case “builds a embedded” workbook. The case now reads the shipped probe, so the article is grammatically wrong and the verb contradicts the implementation.

**The self-test set.** Two cases are exposed: an `excel.openxml.parseRoundTrip` case that reads the
embedded workbook authored in Microsoft Excel, reads it with the reader, and

docs/design/docdown-office/excel/open-xml/excel-open-xml-extractor.md:46

• The new probe description is immediately followed by the old claim that building rather than embedding keeps the case free of a shipped binary. That is now false: this self-test reads an embedded .xlsx; remove the obsolete sentence.

  - `excel.openxml.parseRoundTrip`, which reads the embedded workbook authored in Microsoft Excel
    and passes when the model carries at least one
    worksheet. Embedding a probe Excel itself authored is what makes the case prove this deployment
    can read what the real application emits, rather than that a library agrees with itself.

docs/design/docdown-office/excel/open-xml/excel-open-xml-extractor.md:44

• The design now says this case reads the embedded Excel probe, but its linked requirement still justifies the self-test as one that "builds and reads a workbook" (docs/reqstream/docdown-office/excel/open-xml/excel-open-xml-extractor.yaml:82-85). Update that requirement too so the source, design, and ReqStream statement describe the same probe.

  - `excel.openxml.parseRoundTrip`, which reads the embedded workbook authored in Microsoft Excel
    and passes when the model carries at least one

docs/design/docdown-office/powerpoint/com/powerpoint-com-availability.md:25

• “PowerPoint and Visio use in common” is not grammatical English. Replace it with “PowerPoint and Visio share” so the explanation reads naturally.

The probe itself lives in the shared `OfficeComAvailability`, which PowerPoint and Visio use in
common: the two probes differed only in the application name and the ProgID, so those are arguments
rather than two copies of the logic. This type names PowerPoint's own entry point and supplies them.

docs/design/docdown-office/powerpoint/com/powerpoint-com-extractor.md:48

• The embedded probe carries two slides, and PowerPointComExtractor requires slides.Count == ProbeSlideCount with ProbeSlideCount = 2; this design text still says the self-test passes when exactly one slide returns. That documents a false pass criterion. Describe the expected two slides instead.

  writes the embedded probe deck into the self-test work folder, renders it through the real
  `PowerPointAutomation` at 96 DPI, and passes only when exactly one slide came back carrying

docs/design/docdown-office/powerpoint/com/powerpoint-com-extractor.md:48

• This new description is not synchronized with the SysML2 unit model: docs/sysml2/model/docdown-office/powerpoint/com/powerpoint-com-extractor.sysml:3 still says powerpoint.com.render builds a synthetic single-slide deck, while the source and requirement now use an embedded two-slide probe. Update the model in the same change so the source, design, and model do not disagree.

  writes the embedded probe deck into the self-test work folder, renders it through the real
  `PowerPointAutomation` at 96 DPI, and passes only when exactly one slide came back carrying

docs/design/docdown-office/powerpoint/open-xml.md:40

• This sentence still says the parse case “builds a embedded” deck. The case now reads the shipped probe, so the article is grammatically wrong and the verb contradicts the implementation.

**The self-test set.** Two cases are exposed: a `powerpoint.openxml.parseRoundTrip` case that reads
the embedded deck authored in Microsoft PowerPoint with the reader, and passes

docs/design/docdown-office/powerpoint/open-xml/powerpoint-open-xml-extractor.md:44

• The new probe description is immediately followed by the old claim that building rather than embedding keeps the case free of a shipped binary. That is now false: this self-test reads an embedded .pptx; remove the obsolete sentence.

  - `powerpoint.openxml.parseRoundTrip`, which reads the embedded deck authored in Microsoft PowerPoint
    and passes when the model carries at least one
    slide. Embedding a probe PowerPoint itself authored is what makes the case prove this deployment

docs/design/docdown-office/powerpoint/open-xml/powerpoint-open-xml-extractor.md:43

• The design now says this case reads the embedded PowerPoint probe, but its linked requirement still justifies the self-test as one that "builds and reads a deck" (docs/reqstream/docdown-office/powerpoint/open-xml/powerpoint-open-xml-extractor.yaml:80-83). Update that requirement too so the source, design, and ReqStream statement describe the same probe.

  - `powerpoint.openxml.parseRoundTrip`, which reads the embedded deck authored in Microsoft PowerPoint
    and passes when the model carries at least one

docs/design/docdown-office/visio/com/visio-com-availability.md:24

• “Visio and PowerPoint use in common” is not grammatical English. Replace it with “Visio and PowerPoint share” so the explanation reads naturally.

The probe itself lives in the shared `OfficeComAvailability`, which Visio and PowerPoint use in
common: the two probes differed only in the application name and the ProgID, so those are arguments
rather than two copies of the logic. This type names Visio's own entry point and supplies them.

docs/design/docdown-office/visio/com/visio-com-extractor.md:45

• This new description is not synchronized with the SysML2 unit model: docs/sysml2/model/docdown-office/visio/com/visio-com-extractor.sysml:3 still says visio.com.render builds a synthetic single-page drawing, while the source and requirement now use an embedded probe. Update the model in the same change so the source, design, and model do not disagree.

  `visio.com.render` walks through the door the availability case only knocks on: it writes the
  embedded probe drawing into the self-test work folder, renders it through the real

docs/design/docdown-office/word/open-xml/word-open-xml-extractor.md:40

• The design now says this case reads the embedded Word probe, but its linked requirement still justifies the self-test as one that "builds and reads a document" (docs/reqstream/docdown-office/word/open-xml/word-open-xml-extractor.yaml:77-79). Update that requirement too so the source, design, and ReqStream statement describe the same probe.

  - `word.openxml.parseRoundTrip`, which reads the embedded document authored in Microsoft Word
    and passes when the body carries content.

docs/reqstream/docdown-office.yaml:27

• The verification design treats AddOffice_NullBuilder_Throws as part of DocDownOffice-Registration, but this ReqStream requirement lists only the other three scenarios. That leaves the null-builder acceptance behavior unlinked in the requirements trace; add this test here or introduce a dedicated child requirement for the null-input contract.

        tests:
          - AddOffice_OnBuilder_RegistersEveryOfficeBackend
          - AddOffice_ReturnsSameBuilderForChaining
          - AddExcel_Alone_RegistersOnlyTheExcelBackend

docs/reqstream/docdown-office/visio/open-xml/visio-open-xml-extractor.yaml:78

• Updating this nested requirement to say the managed Visio self-test reads the embedded probe exposes a stale parent statement at docs/reqstream/docdown-office/visio.yaml:251-254, which still says the round-trip builds a drawing. Update the parent requirement too so both ReqStream levels describe the same self-test.
  docs/reqstream/ots/openxml.yaml:50
• The tests list still uses WordOpenXmlExtractor_SelfValidation_ReportsCases as evidence for the SDK's document-creation requirement, but that self-test now reads the embedded Word probe and no longer exercises DocumentFormat.OpenXml's writer. The linked verification text also still says it builds a document. Remove that self-test from this OTS evidence or replace the trace with a writer-backed fixture test.
  docs/reqstream/ots/pdfpig.yaml:86
• The tests list still uses PdfDocumentExtractor_GetSelfTestCases_DeployedBackend_ReturnsParseAndRenderingCases as evidence for PdfPig document creation, but that self-test now reads an embedded PDF and no longer exercises PdfPig's writer. The linked verification text still says the self-test builds a document. Remove that self-test from this OTS evidence or replace the trace with a writer-backed fixture test.
  src/DemaConsulting.DocDown.Core/Output/ContentWriter.cs:214
• The new parameter text narrows ContentPart to worksheets/charts, but the summary immediately above still says a part can be a sheet or slide, and the ContentPartKind remarks below likewise describe decks/slides even though the enum only has Sheet and Chart. Please update the surrounding XML documentation to one consistent vocabulary; otherwise the public API docs contradict the actual type.
  src/DemaConsulting.DocDown.Office/PowerPoint/Com/PowerPointComExtractor.cs:308
• Although this hunk correctly describes the embedded probe, the class still documents RenderAndInspect's input as a synthetic deck and retains a probe-size remark describing a single slide, while ProbeSlideCount is 2. The updated XML documentation therefore still contradicts the implementation; update those remaining probe descriptions.
  src/DemaConsulting.DocDown.Office/Visio/Com/VisioComExtractor.cs:290
• This hunk switches the case to the embedded probe, but the same class still emits for a drawing the self-test built with exactly one. from DescribeRenderShortfall (and its render-helper parameter still calls the input synthetic). Those descriptions are now false and can mislead release-time failure output; update the remaining provenance text.

• Files reviewed: 106/106 changed files
• Comments generated: 2
• Review effort level: Lite

[Copilot]

🟡 Changes recommended

Unresolved documentation and contract inconsistencies remain in the reviewed changes.

Get a fresh assessment by requesting another Copilot review.

Review details

Suppressed comments (15)

Previously missed (1) — in code that hasn't changed since the last review.

docs/design/docdown-office/com.md:65

• The added ending “because it does” makes this bullet grammatically and logically incomplete: running the probe on every extraction is not the reason it is cheap. Remove the dangling clause or state the intended reason explicitly.

docs/design/docdown-office/powerpoint/com/powerpoint-com-availability.md:25

• "Which PowerPoint and Visio use in common" is not grammatical English in this newly added paragraph. Use "which PowerPoint and Visio share" so the description of the shared helper is clear.

The probe itself lives in the shared `OfficeComAvailability`, which PowerPoint and Visio use in
common: the two probes differed only in the application name and the ProgID, so those are arguments
rather than two copies of the logic. This type names PowerPoint's own entry point and supplies them.

docs/design/docdown-office/powerpoint/com/powerpoint-com-extractor.md:54

• This design update now says the COM self-test renders an embedded PowerPoint deck, but the companion SysML unit model still documents powerpoint.com.render as rendering a synthetic single-slide deck (docs/sysml2/model/docdown-office/powerpoint/com/powerpoint-com-extractor.sysml:3). Update that model description in the same change so the architecture model does not contradict its linked design and source.

- **`ProbeResourceName`** (private const) — names the embedded deck the render case rasterizes: a
  real `.pptx` authored in Microsoft PowerPoint, read through `SelfTestProbe.Load`. The synthesizer

docs/design/docdown-office/visio/com/visio-com-availability.md:24

• "Which Visio and PowerPoint use in common" is not grammatical English in this newly added paragraph. Use "which Visio and PowerPoint share" so the description of the shared helper is clear.

The probe itself lives in the shared `OfficeComAvailability`, which Visio and PowerPoint use in
common: the two probes differed only in the application name and the ProgID, so those are arguments
rather than two copies of the logic. This type names Visio's own entry point and supplies them.

docs/design/docdown-office/visio/com/visio-com-extractor.md:52

• This design update now says the COM self-test renders an embedded Visio drawing, but the companion SysML unit model still documents visio.com.render as rendering a synthetic single-page drawing (docs/sysml2/model/docdown-office/visio/com/visio-com-extractor.sysml:3). Update that model description in the same change so the architecture model does not contradict its linked design and source.

- **`ProbeResourceName`** (private const) — names the embedded drawing the render case rasterizes: a
  real `.vsdx` authored in Microsoft Visio, read through `SelfTestProbe.Load`. The synthesizer that

docs/design/docdown-pdf/pdf-document-extractor.md:44

• This now says the self-test reads the embedded probe but then "rereads it," which is still the old build-and-read behavior. The implementation only loads and parses the embedded bytes once; please describe this as a parse case rather than a round trip.

  parse round trip that reads the embedded probe document and rereads it in memory, and a
  page-rendering case that

docs/verification/docdown-office.md:17

• This new sentence is grammatically incorrect: “what those four registered” has no noun for registered to modify. Say that AddOffice() must register the same backends as the four calls.
  docs/verification/docdown-office.md:32
• The probe set includes Visio, which is read through the Open Packaging/OPC path rather than the Open XML SDK. Calling the ZIP header an invariant of "every Open XML package" is too narrow for the four-probe assertion; use OPC/package wording here.
  src/DemaConsulting.DocDown.Core/Output/ContentWriter.cs:214
• The parameter text now describes worksheets/charts, but the summary immediately above still says “sheet or slide,” while ContentPartKind exposes Sheet and Chart only. Update the summary with this correction so the public XML documentation is internally consistent.
  src/DemaConsulting.DocDown.Office/Excel/OpenXml/ExcelOpenXmlExtractor.cs:148
• The preceding sentence was updated from a generated workbook to the embedded probe, but this line still claims the operation is a round trip. The code only loads the probe into memory, so this remark is now inaccurate.
  src/DemaConsulting.DocDown.Office/PowerPoint/Com/PowerPointComExtractor.cs:308
• The outer self-test documentation now says the embedded probe is written to the work folder, but the RenderAndInspect parameter documentation below still calls that input a synthetic deck. Update the helper documentation too so the source no longer describes a document that is not synthesized.
  src/DemaConsulting.DocDown.Office/Visio/Com/VisioComExtractor.cs:290
• The outer self-test documentation now says the embedded probe is written to the work folder, but the RenderAndInspect parameter documentation below still calls that input a synthetic drawing. Update the helper documentation too so the source no longer describes a document that is not synthesized.
  src/DemaConsulting.DocDown.Office/Word/OpenXml/WordOpenXmlExtractor.cs:142
• The preceding sentence was updated from a generated document to the embedded probe, but this line still claims the operation is a round trip. The code only loads the probe into memory, so this remark is now inaccurate.
  src/DemaConsulting.DocDown.Pdf/PdfDocumentExtractor.cs:242
• The preceding sentence was updated from building a PDF to reading the embedded probe, but this line still claims the operation is a round trip. The code only loads and parses the embedded bytes, so this remark is now inaccurate.
  test/DemaConsulting.DocDown.Core.Tests/Detection/FormatSnifferTests.cs:213
• This hunk removes confidence from one test's contract, but the same test class still documents confidence throughout (for example the class remarks and the PDF, HTML, VSDX, VSDM, legacy-extension, and unknown-format assertions). Leaving those claims in place makes the test documentation contradict the current FormatDetection API; update the remaining references as part of this contract change.

• Files reviewed: 106/106 changed files
• Comments generated: 2
• Review effort level: Lite

[Copilot]

🟡 Changes recommended

Outstanding documentation and evidence corrections remain in the final review comments.

Get a fresh assessment by requesting another Copilot review.

Review details

Suppressed comments (21)

Previously missed (2) — in code that hasn't changed since the last review.

docs/design/docdown-pdf/pdf-document-extractor.md:45

• This design bullet still says the parse case reads and then rereads the probe, but the implementation only loads the embedded PDF and parses it once. Calling it a round trip with a reread preserves the old generated-fixture behavior that this PR is removing; describe it as a single in-memory probe read.
  src/DemaConsulting.DocDown.Core/Output/ContentWriter.cs:214
• The ContentPartKind enum currently exposes only Sheet and Chart (see ContentWriter.cs:236-249), but the summary immediately above still advertises a slide. This new parameter text narrows the description without fixing that stale summary, leaving the public XML documentation inconsistent; update the summary and parameter together.

docs/design/docdown-office/com.md:65

• ExtractorRegistry.GetCandidates() caches availability per registry and re-probes only after RefreshAvailability (src/DemaConsulting.DocDown.Core/Extraction/ExtractorRegistry.cs:118-125), so this constraint is inaccurate: the probe does not run on every extraction. Please document the lazy first probe/cache and explicit refresh behavior instead, otherwise the design overstates registry and OS/registry I/O.

- The availability probe is cheap enough to run on every extraction, because it does.

docs/design/docdown-office/com.md:51

• The same design section still says an Office probe would run on every extraction, but ExtractorRegistry caches availability per registry and only re-probes after RefreshAvailability (src/DemaConsulting.DocDown.Core/Extraction/ExtractorRegistry.cs:118-125). Please describe the cached result and explicit refresh boundary so the operational behavior is accurate.

- **The probe never launches an application.** The engine probes every registered backend before
  selecting one, so a probe that started Microsoft Office would do so on every extraction, including
  on machines that cannot run it.

docs/reqstream/docdown-office.yaml:33

• DocDownOffice-ComAutomation is a requirement with a combined behavior but has no tests: entry; only its child requirements carry tests. That leaves the parent without direct ReqStream evidence and differs from the neighboring parent requirements, which link a test. Add a direct system-level test link or restructure this as a grouping node rather than a separately verified requirement.

      - id: DocDownOffice-ComAutomation
        title: >-
          The system shall report whether its COM automation backends can run in the current
          environment, and shall keep a delegated run's output consistent with what it produced.
        tags: [system]

docs/reqstream/docdown-office.yaml:50

• The sole test linked to DocDownOffice-SelfTestProbes checks only that four resources are non-empty ZIPs. That cannot establish the requirement's stronger condition that each probe was authored by the corresponding Office application; a DocDown-generated package would pass the same test. Either weaken the requirement to an embedded valid probe or add traceable provenance evidence for the authorship claim.

      - id: DocDownOffice-SelfTestProbes
        title: >-
          The system shall carry, for each format it reads, a probe document authored in the
          application that produces that format.

docs/reqstream/docdown-office/com.yaml:33

• This new requirement makes suppression conditional on the COM run having rendered pages, but the current extractors wrap the sink before rendering and report pages.renderer available before RenderSlidesAsync/RenderPagesAsync. If every page fails, the delegated pageRendering fact is still suppressed even though no page was rendered. Make the suppression depend on at least one successful render (with an all-failed test), or define this requirement as a capability statement rather than a rendered-output statement.

          - id: DocDownOffice-Com-DelegatedComposition
            title: >-
              The subsystem shall suppress a delegated managed backend's statement that it does not
              render pages, when the COM backend has rendered them.

docs/reqstream/docdown-office/com.yaml:20

• ExtractorRegistry caches each backend's availability after the first candidate query and re-probes only after RefreshAvailability; it is not probed “on every extraction.” This makes the justification factually inaccurate and overstates the side effect being prevented. Please describe the cached probe pass instead.

              The engine probes every registered backend before it selects one, so a probe that
              launched Microsoft Office would start an application on a machine that may not even be
              able to run it, on every extraction. A probe reads the operating system and the

docs/reqstream/docdown-office/excel/open-xml/excel-open-xml-extractor.yaml:86

• This justification still says the embedded workbook is read back in a round trip, but the self-test only loads the committed Excel-authored probe and parses it. Replace the round-trip claim with the actual probe-parse behavior so the requirement evidence remains truthful.
  docs/reqstream/docdown-office/powerpoint/open-xml/powerpoint-open-xml-extractor.yaml:84
• The requirement justification says the embedded probe is read back, which implies a writer/read round trip that this implementation no longer performs. It only loads and parses the application-authored deck; revise this justification so the requirement does not claim evidence from a nonexistent write step.
  docs/reqstream/docdown-office/visio/open-xml/visio-open-xml-extractor.yaml:81
• This justification says the embedded Visio drawing is read back in a round trip, but the self-test only loads and parses the committed application-authored probe. That overstates what the test proves; revise it to describe the direct topology parse.
  docs/reqstream/docdown-office/word/open-xml/word-open-xml-extractor.yaml:80
• The requirement justification still describes reading the embedded Word probe back after a write, but the implementation only loads and parses the application-authored document. Update this evidence statement to describe the actual embedded-probe parse.
  docs/verification/docdown-office/visio/open-xml.md:30
• This subsystem input description says there are no committed binary fixtures, but the managed self-test reads the committed application-authored probe.vsdx. Distinguish generated suite packages from the embedded probe; otherwise the verification setup omits a required input and contradicts the implementation.
  src/DemaConsulting.DocDown.Core/Detection/FormatSniffer.cs:162
• The implementation treats a recognized extension as the primary signal and returns before reading content, so calling the extension "weaker evidence" than a content signature is misleading now that confidence scores are gone. Please describe the extension as the primary/short-circuiting signal and the signature as the fallback instead.
  src/DemaConsulting.DocDown.Office/Excel/OpenXml/ExcelOpenXmlExtractor.cs:148
• The updated description says the case reads an embedded Excel workbook, but the last sentence still says the whole round trip fits in memory. This case no longer creates a workbook or performs a round trip; describe the embedded probe as being loaded and parsed in memory instead.
  src/DemaConsulting.DocDown.Office/PowerPoint/OpenXml/PowerPointOpenXmlExtractor.cs:137
• This hunk now says the self-test only reads the embedded deck, but the immediately preceding summary and case ID still call it a “parse round-trip,” and the exception path below still reports a round trip. Since this code never writes a deck, update those remaining labels/messages so the self-test documentation and failure output describe the operation that actually runs.
  src/DemaConsulting.DocDown.Office/Visio/OpenXml/VisioOpenXmlExtractor.cs:136
• Although this hunk now loads the embedded drawing, it still says the drawing is parsed “back” and leaves the method summary/case name and exception text as a parse round-trip. Nothing is written before the reader runs, so this describes a different operation and makes failures misleading; update the remaining round-trip wording consistently.
  src/DemaConsulting.DocDown.Office/Word/OpenXml/WordOpenXmlExtractor.cs:142
• The updated description says the case reads an embedded Word document, but the last sentence still says the whole round trip fits in memory. This case no longer creates a document or performs a round trip; describe the embedded probe as being loaded and parsed in memory instead.
  src/DemaConsulting.DocDown.Pdf.Rendering/PdfPageRenderingExtractor.cs:295
• The implementation now rasterizes a page from an embedded PDF, not a PDF it creates and rereads, but the method summary and catch message still call this a render round trip. That leaves the self-test's description and reported failures factually wrong; replace the remaining round-trip wording with render-specific wording.
  src/DemaConsulting.DocDown.Pdf/PdfDocumentExtractor.cs:242
• The first sentence now correctly describes a shipped probe, but the final sentence still calls this a round trip. The self-test no longer builds and rereads a document; it loads one embedded PDF and parses it, so this remark misstates what the case proves.
  test/DemaConsulting.DocDown.Office.Tests/Visio/OpenXml/VisioOpenXmlExtractorTests.cs:47
• The updated test description still says the self-test “reads back” the embedded drawing, which implies a prior write that no longer occurs. The test only loads the committed Visio-authored probe and parses it; remove “back” so the evidence describes the actual check.

• Files reviewed: 111/111 changed files
• Comments generated: 2
• Review effort level: Lite

[Copilot]

🟡 Changes recommended

Unresolved moderate traceability and documentation issues remain.

Get a fresh assessment by requesting another Copilot review.

Review details

Suppressed comments (21)

Previously missed (1) — in code that hasn't changed since the last review.

src/DemaConsulting.DocDown.Office/Visio/Com/VisioComExtractor.cs:282

• These updated remarks now say the render writes the embedded probe, but DescribeRenderShortfall still emits “for a drawing the self-test built with exactly one.” That failure message is false for the current implementation; update it to identify the embedded probe drawing instead.

docs/design/docdown-core/output/manifest-writer.md:74

• This caller note is unconditional, but unreadable runs go through WriteFailureAsync and call ProduceResultAsync with no ContentWriter result; only produced runs invoke ContentWriter before ManifestWriter. Clarify the failure-path exception so this design sequence matches the implementation.

`DocDownEngine` calls `ManifestWriter` after `ContentWriter` and before `MetadataWriter`.

docs/design/docdown-office/excel/open-xml/excel-open-xml-extractor.md:51

• The self-test now only loads and parses the embedded workbook; it does not build or read back a workbook, so calling this the “round-trip case” is inaccurate and conflicts with the preceding probe description.

- **`RunParseRoundTrip`** (private) — the round-trip case body. It reads the embedded workbook named

docs/design/docdown-office/powerpoint/com.md:65

• This replacement drops the PowerPointComDispatch supporting-type bullet and leaves its continuation as an orphaned sentence, so the section no longer documents the dispatch helper and appears to attribute watchdog/COM details to the shared sink. Keep the shared-type bullet and re-add PowerPointComDispatch as its own bullet before the section ends.

- **`ComposingDelegatedSink`** and **`DelegatedExtractionContext`** — not owned by this
  subsystem. They live in the shared `DocDown.Office.Com` namespace, because Visio and PowerPoint
  needed byte-equivalent copies of both. The sink suppresses the delegated managed backend's
  `powerpoint.pageRendering` unavailable fact so it does not contradict the rendering this run performed; the fact
  key is a constructor argument, which was the only real difference between the two former copies.

docs/design/docdown-office/powerpoint/open-xml/powerpoint-open-xml-extractor.md:51

• The self-test now only loads and parses the embedded deck; it does not build or read back a deck, so calling this the “round-trip case” is inaccurate and conflicts with the preceding probe description.

- **`RunParseRoundTrip`** (private) — the round-trip case body. It reads the embedded deck named by

docs/design/docdown-office/visio/com.md:52

• The shared-type replacement removes VisioComDispatch from the subsystem's supporting-type list, even though this design doc is the model's documented design location for that unit and the adapter still depends on it. Add the low-level dispatch helper as its own bullet after the shared types.

- **`ComposingDelegatedSink`** and **`DelegatedExtractionContext`** — not owned by this
  subsystem. They live in the shared `DocDown.Office.Com` namespace, because Visio and PowerPoint
  needed byte-equivalent copies of both. The sink suppresses the delegated managed backend's
  `visio.pageRendering` unavailable fact so it does not contradict the rendering this run performed; the fact
  key is a constructor argument, which was the only real difference between the two former copies.

docs/design/docdown-office/word/open-xml/word-open-xml-extractor.md:49

• The self-test now only loads and parses the embedded document; it no longer performs a write/read round trip, so describing this resource as what the “round-trip case” reads is inaccurate.

- **`ProbeResourceName`** (private const) — names the embedded document the round-trip case reads: a

docs/reqstream/docdown-office.yaml:45

• This new system requirement has child requirements but no tests: entry, so ReqStream has no direct system-level evidence for the stated COM behavior. The repository's requirements standard requires every requirement to link a passing test; add a system-level scenario (or otherwise restructure this as a non-requirement grouping) so the requirement is traceable.

      - id: DocDownOffice-ComAutomation
        title: >-
          The system shall report whether its COM automation backends can run in the current
          environment, and shall keep a delegated run's output consistent with what it produced.
        tags: [system]
        justification: |
          Two of the six backends this package ships drive a Microsoft Office application over COM,
          and neither can run where that application is absent - which is every CI runner and most
          Linux and macOS hosts. The system must therefore answer whether such a backend is usable
          before it is selected, and must do so without launching the application. Where a COM
          backend does run, it delegates content extraction to its managed counterpart and adds
          rendering on top, so the composed report must not carry the delegate's statement that no
          rendering was performed. Both behaviors are shared by the COM backends and decompose into
          the Com subsystem.
        children:
          - DocDownOffice-Com-Availability
          - DocDownOffice-Com-DelegatedComposition

docs/reqstream/docdown-office.yaml:59

• The linked Probe_EmbeddedResource_LoadsAsAnOpenPackage test only checks that each resource is a non-empty ZIP; a DocDown-synthesized package would pass it too. Nothing in the automated evidence proves the stronger requirement that the probes were authored by Word, Excel, PowerPoint, and Visio, so either add verifiable provenance evidence or narrow the requirement to carrying embedded probes.

      - id: DocDownOffice-SelfTestProbes
        title: >-
          The system shall carry, for each format it reads, a probe document authored in the
          application that produces that format.
        tags: [system]
        justification: |
          A self-test exists to answer whether this deployment can read what the real application
          emits. A backend that synthesized its own document could only ever prove that a library
          agreed with itself, and would keep reporting success while failing on every genuine
          document. Carrying the probe inside the package also means the check does not depend on
          the file system, the working directory, or anything the host happens to have installed.
        tests:
          - Probe_EmbeddedResource_LoadsAsAnOpenPackage

docs/verification/docdown-office/excel.md:44

• This sentence says every workbook used by the suite is generated, but the extractor self-test exercised by the suite reads the committed embedded Excel probe. Qualify the sentence as referring to test fixtures so it does not contradict the probe distinction immediately below.
  docs/verification/docdown-office/powerpoint.md:55
• This sentence says every deck used by the suite is generated, but PowerPointOpenXmlExtractor_SelfTestCases_RoundTripPassesAndRenderingSkipped runs against the committed embedded PowerPoint probe. Qualify it as referring to test fixtures to preserve the distinction this section is intended to document.
  docs/verification/docdown-office/visio.md:55
• This sentence says every drawing used by the suite is generated, but VisioOpenXmlExtractor_SelfTestCases_RoundTripPassesAndRenderingSkipped reads the committed embedded Visio probe. Qualify it as referring to test fixtures so it matches the probe distinction stated in the next paragraphs.
  docs/verification/docdown-office/word.md:38
• This sentence says every document used by the suite is generated, but WordOpenXmlExtractor_SelfValidation_ReportsCases executes the self-test that reads the committed embedded Word probe. Since the section is distinguishing fixtures from probes, qualify this sentence as referring to test fixtures.
  docs/verification/docdown-pdf-rendering.md:49
• The paragraph calls the embedded probe an exception to the rule that every PDF used by the suite is generated, but the probe is not a generated test fixture and is consumed by the validation test. Phrase the rule around test fixtures and describe the probe as a separate self-test input instead.
  docs/verification/docdown-pdf.md:31
• This sentence says every PDF used by the suite is generated, but PdfDocumentExtractor_GetSelfTestCases_DeployedBackend_ReturnsParseAndRenderingCases runs the self-test against the committed embedded probe. Qualify this as referring to PDF fixtures so the paragraph does not contradict its own probe exception.
  src/DemaConsulting.DocDown.Office/Excel/OpenXml/ExcelOpenXmlExtractor.cs:148
• This self-test now only loads and parses the embedded Excel workbook; it no longer creates a workbook, so the remaining “whole round trip” wording is stale. Describe the embedded probe being loaded and parsed in memory instead.
  src/DemaConsulting.DocDown.Office/PowerPoint/Com/PowerPointComExtractor.cs:300
• These updated remarks correctly identify the embedded probe, but the adjacent MaximumPlausiblePixels documentation still says the deck is a single slide even though ProbeSlideCount is 2. Please update that bound's explanation to describe the per-slide limit without the contradictory single-slide claim.
  src/DemaConsulting.DocDown.Office/Word/OpenXml/WordOpenXmlExtractor.cs:142
• This self-test now only loads and parses the embedded Word document; it no longer creates a document, so the remaining “whole round trip” wording is stale. Describe the embedded probe being loaded and parsed in memory instead.
  src/DemaConsulting.DocDown.Pdf.Rendering/PdfPageRenderingExtractor.cs:316
• The changed failure message now identifies the embedded document, but the exception path below still reports that a round trip failed. This render self-test only loads and rasterizes probe bytes, so the operator-facing error should describe failure to rasterize the embedded document.
  src/DemaConsulting.DocDown.Pdf/PdfDocumentExtractor.cs:242
• This self-test now only loads and parses an embedded PDF; it no longer creates a document, so describing the operation as a round trip is inaccurate. Replace the final sentence with the in-memory embedded-probe behavior so the API documentation matches the implementation.
  src/DemaConsulting.DocDown.Pdf/PdfDocumentExtractor.cs:257
• The changed success-path message now identifies the embedded probe, but the exception path below still tells the operator that a round trip failed. Since this case no longer builds anything, the failure message should describe inability to read the embedded document rather than a nonexistent round trip.

• Files reviewed: 111/111 changed files
• Comments generated: 1
• Review effort level: Lite

[Copilot]

🔵 Needs a closer look

Unresolved documentation and requirement inconsistencies remain, including one moderate finding.

Review details

Suppressed comments (13)

Previously missed (1) — in code that hasn't changed since the last review.

docs/reqstream/docdown-office/com.yaml:33

• The shared sink suppresses the delegate's unavailable fact when it wraps the managed extraction, before the COM render result is known; even an all-pages-failed render follows that path. This requirement therefore promises conditional behavior the implementation does not provide and conflicts with the design's description of filtering the unavailable fact. Phrase the requirement around the COM backend owning the rendering path, or change the implementation to defer suppression.

docs/design/docdown-pdf/pdf-document-extractor.md:43

• RunParseRoundTrip now loads the embedded probe once; it does not reread it. This sentence still describes the removed writer-based round trip, so the design document contradicts the implementation it was just updated to describe.

  parse round trip that reads the embedded probe document and rereads it in memory, and a

docs/reqstream/docdown-office.yaml:13

• The PR description's 589 → 593 requirement count is not supported by the checked-in ReqStream: this tree contains 592 requirement IDs. This change adds five Office IDs, removes two OTS IDs, and renames one Core ID, for a net increase of three from 589; please reconcile the campaign summary with the actual count.

      - id: DocDownOffice-Registration

docs/reqstream/ots/openxml.yaml:42

• Removing DocDown-OTS-OpenXml-DocumentCreation leaves the companion verification document's Required Functionality section still stating that the SDK must build documents suitable for test fixtures, while the suite still generates those fixtures at runtime. Either retain this requirement and its evidence, or remove/update that claim in the companion documents so the requirements and verification artifacts do not disagree.
  docs/reqstream/ots/pdfpig.yaml:78
• Removing DocDown-OTS-PdfPig-DocumentCreation leaves the companion verification document's Required Functionality section still stating that PdfPig must build documents suitable for test fixtures, while the suite still generates those fixtures at runtime. Either retain this requirement and its evidence, or remove/update that claim in the companion documents so the requirements and verification artifacts do not disagree.
  docs/verification/docdown-office/com.md:53
• This acceptance criterion has the same mismatch as the requirement: ComposingDelegatedSink filters the managed unavailable fact before rendering and does so even when the adapter returns only per-page failures. It should describe suppression while the COM backend owns the rendering path, or the implementation and tests need to be changed to make suppression conditional on successful rendering.
  docs/verification/ots/openxml.md:70
• Removing the DocDown-OTS-OpenXml-DocumentCreation requirement leaves this verification document's Required Functionality section still claiming that the SDK must build test-fixture documents. Fixture generation is test infrastructure rather than a remaining OTS requirement, so this section now overstates the contract; remove the writer clause there while retaining the Test Environment description of how fixtures are generated.
  docs/verification/ots/pdfpig.md:113
• Removing the DocDown-OTS-PdfPig-DocumentCreation requirement leaves this verification document's Required Functionality section still claiming that PdfPig must build test-fixture documents. The writer is test infrastructure, not a capability required by the parser contract, so remove that clause from the required functionality statement while retaining the Test Environment description of generated fixtures.
  src/DemaConsulting.DocDown.Core/Detection/FormatSniffer.cs:162
• This still says an extension is weaker evidence and that only the decoder can prove the bytes agree with it, but Detect returns a recognized extension result before checking content (FormatSniffer.cs:106-110), and the tests explicitly assert that an extension wins over contradictory bytes. Update this remark to describe the actual contract: extension is primary and content signature is fallback.
  src/DemaConsulting.DocDown.Office/Excel/OpenXml/ExcelOpenXmlExtractor.cs:148
• The updated description correctly says the self-test reads an embedded Excel workbook, but the final sentence still calls it a "round trip." No workbook is created or reread here; the probe is loaded and parsed once.
  src/DemaConsulting.DocDown.Office/Visio/Com/VisioComExtractor.cs:290
• These new remarks say the self-test writes and renders the embedded probe, but DescribeRenderShortfall still reports a wrong page count as coming from a drawing “the self-test built.” That failure message now contradicts the actual input and the updated self-test description; update it to refer to the embedded probe drawing.
  src/DemaConsulting.DocDown.Office/Word/OpenXml/WordOpenXmlExtractor.cs:142
• The updated description correctly says the self-test reads an embedded Word document, but the final sentence still calls it a "round trip." No document is created or reread here; the probe is loaded and parsed once.
  src/DemaConsulting.DocDown.Pdf/PdfDocumentExtractor.cs:242
• This remark now describes loading the embedded PDF, but it still says that the "whole round trip" fits in memory. The self-test no longer builds or rereads a document; it loads the probe and parses it once, so this leaves the new behavior inaccurately documented.

• Files reviewed: 113/113 changed files
• Comments generated: 0 new
• Review effort level: Lite

[Copilot]

🟡 Changes recommended

One or more issues must be addressed before approval.

Get a fresh assessment by requesting another Copilot review.

Review details

Suppressed comments (7)

docs/design/ots/pdftoimage.md:22

• Adding GetPageCount here makes PDFtoImage usage no longer confined to PageRenderer: the implementation calls it directly from PdfPageRenderingExtractor.SelectPageNumbers. That contradicts the single-file native-interop/containment statements in this design and the PageRenderer requirement; route page counting through PageRenderer under its gate, or update the containment and synchronization contract together.
  docs/design/ots/pdftoimage.md:21
• PDFtoImage.GetPageCount is not the API used by the implementation; the call is PDFtoImage.Conversion.GetPageCount(...). As written, this feature list points readers to a nonexistent type/member combination.
  docs/reqstream/docdown-office.yaml:27
• The verification acceptance criteria include AddOffice_NullBuilder_Throws (and list it as a scenario), but this requirement's test list omits it. The null-builder behavior is therefore tested and documented but not traceable to DocDownOffice-Registration; please add it to this list.
  docs/verification/docdown-pdf-rendering.md:48
• This section now identifies the committed probe as the self-test input, but the same document's self-test scenario at lines 186-187 still says the backend “builds itself”; line 85 also still counts PDFium and SkiaSharp as separate OTS items. Leaving those statements unchanged makes the verification document contradict the new probe and OTS structure.
  docs/verification/ots.md:45
• Removing PDFium and SkiaSharp as standalone OTS items here leaves contradictory companion evidence: docs/verification/introduction.md still lists both as OTS items and docs/verification/docdown-pdf-rendering.md plus page-renderer.md still call them separate OTS items. Update those artifacts as well so the canonical inventory and the cited verification sets agree.
  src/DemaConsulting.DocDown.Office/PowerPoint/Com/PowerPointComExtractor.cs:300
• The embedded PowerPoint probe is now a two-slide deck, but the existing MaximumPlausiblePixels remark still describes a single 13.3-by-7.5-inch slide. Update that explanation to match the two-slide probe (or explain the bound per slide), otherwise the source documents contradictory probe geometry.
  src/DemaConsulting.DocDown.Office/Visio/Com/VisioComExtractor.cs:282
• The changed self-test now renders the embedded Visio probe, but the shortfall text in DescribeRenderShortfall still says the drawing was "built with exactly one." A page-count mismatch will report false provenance; change that message to identify the embedded probe drawing.

• Files reviewed: 128/128 changed files
• Comments generated: 3
• Review effort level: Lite

[Copilot]

🟡 Changes recommended

Unresolved moderate issues remain in PDF rendering test coverage and OTS/design documentation.

Get a fresh assessment by requesting another Copilot review.

Review details

Suppressed comments (6)

docs/design/docdown-pdf-rendering/page-renderer.md:45

• PageRenderer now also calls Conversion.GetPageCount to choose pages, but this dependency entry still lists only SavePng and RenderOptions. Update the design dependency list so the new native page-count call is documented alongside the rendering path.

- **PDFtoImage** (OTS) — `Conversion.SavePng` and `RenderOptions`, which rasterize the page and write
  its PNG bytes in one call. No type from the transitive native stack is named here. See *PDFtoImage*
  under the OTS integration design.

docs/design/introduction.md:133

• This rationale uses public type naming as the OTS boundary, but the repository standard classifies external third-party components, not only directly named types, as OTS (.github/standards/software-items.md:87-100). PDFium and SkiaSharp still ship transitively here and provide the native raster/PNG functionality; this claim therefore leaves the design inconsistent with the repository's OTS classification and remaining verification references.

- **PDFtoImage** — managed page-rasterization API, the runtime dependency of DocDown.Pdf.Rendering;
  it delivers a native rasterizer and 2D backend (PDFium and SkiaSharp) transitively, and those are
  the only native binaries DocDown ships. No DocDown type names either of them, so neither is a
  listed OTS item

docs/design/ots.md:24

• The OTS inventory now drops PDFium and SkiaSharp solely because DocDown does not name their types, but both remain third-party components delivered in this package's runtime dependency graph and provide the native raster/PNG functionality. The repository standard defines external third-party components as OTS (.github/standards/software-items.md:87-100), and docs/verification/introduction.md still lists both as OTS items. Keep their individual traceability artifacts, or update the categorization standard/model and every remaining cross-reference consistently.

**PdfPig, PDFtoImage, the Open XML SDK, and TestResults are the exceptions, and deliberately so.**
PdfPig is a runtime library that `DocDown.Pdf` depends on and that therefore flows to consumers of
that package; PDFtoImage (with its transitive PDFium and SkiaSharp native stack) is the runtime
library the optional `DocDown.Pdf.Rendering` package depends on to rasterize pages; the Open XML SDK

docs/verification/ots.md:41

• Removing PDFium and SkiaSharp from this OTS inventory leaves the verification documentation inconsistent: docs/verification/docdown-pdf-rendering.md:85-86, docs/verification/docdown-pdf-rendering/page-renderer.md:12-13, and docs/verification/introduction.md:165-173 still list them as separate OTS items. Update those references in the same change so the declared OTS inventory and its evidence agree.
  test/DemaConsulting.DocDown.Pdf.Rendering.Tests/PdfPageRenderingExtractorTests.cs:129
• SkipWhenRendererUnavailable() probes the native stack and skips this test when it is absent, but the injected pageCount delegate throws before PageRenderer.Render is ever called. The count-fault isolation path therefore is not exercised on hosts without the native renderer; remove this skip so the new behavior is covered everywhere.
  test/DemaConsulting.DocDown.Pdf.Rendering.Tests/PdfPageRenderingExtractorTests.cs:152
• The new test is named and documented as keeping the delegated content, but ContractAssert.LayoutPresent only checks summary.txt and manifest.json; it never checks content.md. A regression that loses the delegated text while preserving the top-level layout would still pass, so assert the content file (and a fixture phrase) here.

• Files reviewed: 133/133 changed files
• Comments generated: 1
• Review effort level: Lite

[Copilot]

🟡 Changes recommended

Outstanding documentation and traceability inconsistencies require updates before approval.

Get a fresh assessment by requesting another Copilot review.

Review details

Suppressed comments (12)

docs/design/ots.md:50

• This inventory now says the Open XML SDK reads Office documents, but docs/design/ots/openxml.md still describes only DocDown.Word/WordprocessingML, still lists WordprocessingDocument.Create as a used feature, and points at the removed DemaConsulting.DocDown.Word.csproj. Update that companion design (and its requirements/verification companions) so the consolidated OTS entry is internally consistent.

| Open XML SDK        | Reads Office documents for the DocDown.Office extraction package         |

docs/design/ots.md:53

• This inventory now lists PDFtoImage as the sole PDF-rendering OTS item, but the changed PDF rendering verification document still enumerates PDFium and SkiaSharp as separate OTS items at docs/verification/docdown-pdf-rendering.md:85. The OTS inventory and evidence set are therefore inconsistent; update that narrative in this consolidation.

| PDFtoImage          | Managed page-rasterization API for the DocDown.Pdf.Rendering package     |

docs/reqstream/docdown-office.yaml:51

• DocDownOffice-SelfTestProbes requires each probe to be authored by the application that produces its format, but its only cited test checks that each resource is non-empty and starts with PK; the verification document explicitly calls that a packaging assertion, not a content or provenance assertion. Link the four backend self-test cases that read the embedded probes, or narrow this requirement to the property the cited test actually proves.
  docs/verification/docdown-office.md:35
• The only cited probe test checks that each resource is a non-empty ZIP, and this section explicitly calls that a packaging—not content—assertion. That cannot evidence the requirement's distinguishing claim that each probe was authored by the producing application; either add format-specific checks that reject synthetic probes or narrow the requirement and its justification.
  docs/verification/ots.md:38
• The broadened inventory entry is not reflected in docs/verification/ots/openxml.md, which still describes only the Word test suite and DocDown.Word even though the SDK now serves the consolidated Office package. Update the companion verification document before treating the OTS campaign as complete.
  src/DemaConsulting.DocDown.Office/Excel/OpenXml/ExcelOpenXmlExtractor.cs:148
• This self-test now reads the embedded Excel probe rather than building a workbook, so it is no longer a round trip. The final sentence still describes a round trip that fits in memory; update it to explain that the probe is loaded directly from the assembly.
  src/DemaConsulting.DocDown.Office/Word/OpenXml/WordOpenXmlExtractor.cs:142
• This self-test now reads the embedded Word probe rather than building a document, so it is no longer a round trip. The final sentence still describes a round trip that fits in memory; update it to explain that the probe is loaded directly from the assembly.
  src/DemaConsulting.DocDown.Pdf.Rendering/PageRenderer.cs:126
• This new GetPageCount seam is not described in docs/design/docdown-pdf-rendering/page-renderer.md's Key Methods section, which still lists only Render and ProbeAvailability. Because page counting now drives selection and shares RenderGate, the design and its verification evidence should document this behavior too.
  src/DemaConsulting.DocDown.Pdf.Rendering/PdfPageRenderingExtractor.cs:250
• The new page-count exception path is not reflected in the companion ReqStream/verification artifacts: they still cover only per-page render faults. This leaves the new Produced/note behavior untraced. Add a count-failure requirement/scenario, or explicitly extend the existing failure requirement to cover it.
  src/DemaConsulting.DocDown.Pdf.Rendering/PdfPageRenderingExtractor.cs:61
• _pageCount is now a second unit field and the source of page selection, but docs/design/docdown-pdf-rendering/pdf-page-rendering-extractor.md still says _render is the only field and that PdfPig supplies the page count. Please update the companion design to describe the PDFtoImage count delegate and the injected count-fault constructor; otherwise the unit's architecture and dependency documentation are false.
  src/DemaConsulting.DocDown.Pdf/PdfDocumentExtractor.cs:242
• After switching this self-test from building a document to loading the embedded probe, it no longer performs a round trip. The updated remark still claims that the whole round trip fits in memory; describe the assembly-loaded probe instead so the documentation matches the implementation.
  test/DemaConsulting.DocDown.Pdf.Rendering.Tests/PageRendererTests.cs:13
• This updated test remark correctly attributes the evidence to PDFtoImage, but the companion verification documents still describe standalone PDFium and SkiaSharp OTS items, and docs/verification/introduction.md still lists both. Update those artifacts so the OTS verification inventory agrees with the consolidated PDFtoImage item.

• Files reviewed: 133/133 changed files
• Comments generated: 2
• Review effort level: Lite

[Copilot]

🟡 Changes recommended

Four moderate findings remain unresolved, along with five documentation nits.

Get a fresh assessment by requesting another Copilot review.

Review details

Suppressed comments (8)

Previously missed (1) — in code that hasn't changed since the last review.

test/DemaConsulting.DocDown.Pdf.Rendering.Tests/PdfPageRenderingExtractorTests.cs:152

• The test name and assertion comment promise to keep delegated content, but ContractAssert.LayoutPresent only checks summary/manifest existence and mandatory blocks; it never inspects content.md. A regression that drops the delegated text would still pass. Assert that content.md contains text from RenderingFixtures.SimpleText() (for example, the quick-brown-fox line).

docs/design/docdown-office/com.md:65

• This sentence is incomplete: “because it does” never states what makes the probe cheap, leaving the design constraint unreadable. Finish the rationale, such as the fact that it only checks the OS and ProgID registration.

- The availability probe is cheap enough to run on every extraction, because it does.

docs/design/introduction.md:133

• These lines remove PDFium and SkiaSharp from the OTS inventory solely because DocDown does not name their types. The repository's OTS standard covers third-party components that provide functionality (.github/standards/software-items.md:22-27), and the PDFtoImage design identifies these as independently delivered native components (docs/design/ots/pdftoimage.md:12-14); transitive delivery and public-surface containment do not remove their traceability obligations. Restore separate OTS requirements/design/verification/review entries, or introduce and apply a documented bundling rule consistently.

  the only native binaries DocDown ships. No DocDown type names either of them, so neither is a
  listed OTS item

docs/reqstream/docdown-office.yaml:50

• The requirement requires each probe to be authored by the originating application, but its only linked test checks that the embedded bytes are non-empty and start with a ZIP header. A synthetic package would pass those assertions, so the distinguishing requirement is not verified; add content/provenance evidence or narrow the requirement to an embedded valid package.
  docs/reqstream/docdown-office.yaml:28
• The new system verification document lists AddOffice_NullBuilder_Throws as evidence for DocDownOffice-Registration, but the requirement's tests list omits it. Add the named test here (or stop attributing that scenario to this requirement) so the ReqStream trace and verification document agree.
  docs/verification/docdown-pdf-rendering.md:42
• The system verification still documents only the injected per-page render fault, while the PR adds PdfPageRenderingExtractor_ExtractAsync_PageCountFaults_ReportsNoteAndKeepsDelegatedContent. Add a system scenario and acceptance criterion for the count-failure note and preserved delegated output.
  docs/verification/docdown-pdf-rendering.md:87
• The sentence is grammatically incomplete: either has no object. Complete it as either component so this verification document is readable.
  docs/verification/introduction.md:174
• This entry attributes the verification of System.IO.Packaging only to DocDown.Word, but VisioPackageReader calls the API directly and the consolidated Office tests cover that path. Update the evidence description to include the direct Visio usage rather than implying Word-only evidence.

• Files reviewed: 135/135 changed files
• Comments generated: 1
• Review effort level: Lite