For qualitative

For quantitative

**Project type:** This skill works for **all project types**:
- **Qualitative**: Verifies participant quotes against transcripts
- **Quantitative**: Verifies statistical claims against data/scripts
- **Mixed**: Handles both verification types

Updates `progress.yaml` when complete:
```yaml
status:
  verification: done
artifacts:
  verification_report: verification/verification-report.md

1. Stage and commit current state:

   git add [files] && git commit -m "verifier: Phase N complete"

2. Then proceed with modifications.

-v2

-final

-working

1. Extract all direct quotes and verifiable claims from a manuscript
2. Map each item to its purported source (interview, article, document)
3. Verify each item is present in the source using efficient search
4. Escalate to deep reading (haiku agent) when fast search fails
5. Report verification status with issues flagged for review

• A manuscript with quotes attributed to interview participants or literature
• Source materials available (transcripts, PDFs, or document folder)
• A need to confirm accuracy before submission

• Final check before journal submission
• After revisions that moved or edited quotes
• When using quotes from secondary sources
• Quality assurance for interview-based findings sections

• Participant quotes with pseudonyms
• Claims about what participants said/did/felt
• Aggregate claims ("Most participants...")
• Paraphrased attributions

• Direct quotes from cited works
• Paraphrased claims with citations
• Data or statistics attributed to sources
• Theoretical claims linked to specific authors

• Read the manuscript (or specified sections)
• Identify source type: interviews, literature, or mixed
• Locate source materials (folder path, Zotero collection, or file list)
• Confirm verification scope (all quotes, specific sections, etc.)
• Count approximate items to verify

## Scope Summary

verification-report.md

Pause: User confirms scope and source locations.

• Identify direct quotes (text in quotation marks with attribution)
• Identify paraphrased claims with source attribution
• Identify aggregate claims about participants or literature
• For each item, extract:
  • The quote or claim text
  • The attributed source (participant name, author/year)
  • Location in manuscript (section, approximate position)
  • Verification level (exact, near, paraphrase, aggregate)
• Create extraction database

## Verification Items

verification-report.md

Pause: User reviews extracted items. Can mark items to skip.

• For interview quotes: Match participant pseudonym to transcript file
• For literature: Match citation to PDF/document or Zotero item
• Flag unmapped items:
  • Participant not found in transcript list
  • Citation not found in source materials
  • Ambiguous source references
• Create source-to-item mapping

## Verification Items

verification-report.md

Pause: User resolves unmapped items.

1. Fast search (Grep tool):
   • For exact quotes: Search for distinctive phrase (8-15 words)
   • For paraphrases: Search for key terms that must appear
   • If found: Mark VERIFIED with source location
2. If not found, fuzzy search:
   • Try variations (punctuation, spacing, common OCR errors)
   • Search for partial matches (beginning/end of quote)
   • If found: Mark VERIFIED with notes on variation
3. If still not found, deep reading (haiku agent):
   • Spawn haiku agent with source document and search target
   • Agent reads document looking for semantic match
   • Agent returns: FOUND (with location), NOT FOUND, or PARTIAL MATCH
4. Record result:
   • VERIFIED: Exact or acceptable match found
   • PARTIAL: Quote/claim partially matches source
   • NOT FOUND: Could not locate in purported source
   • NEEDS REVIEW: Ambiguous or requires human judgment

## Verification Results

verification-report.md

Pause: After each batch of ~20 items to show progress.

• Summarize verification results:
  • Total items verified
  • Items by status (verified, partial, not found, needs review)
  • Breakdown by source type
• Detail issues:
  • NOT FOUND items with context and recommendations
  • PARTIAL matches with specific discrepancies
  • NEEDS REVIEW items with decision prompts
• Provide fix recommendations:
  • Quote corrections with source text
  • Missing attribution suggestions
  • Items to remove or rewrite

## Verification Report

verification-report.md

"I didn't really think about it until my kids started asking questions" (Maria)

1. Primary: "didn't really think about it until my kids"
2. Fallback: "think about it until" AND "kids" AND "questions"
3. Fuzzy: "did not really think" OR "didnt really think"

• Exact quotes must be exact: Even small changes should be marked with [...] or [sic]
• Source materials must be accessible: Can't verify against unavailable documents
• Participant names must be consistent: Pseudonym in manuscript must match transcript filename or label
• Deep reading is expensive but thorough: Use haiku agents when grep genuinely fails, not as first resort
• Some items need human judgment: Flag ambiguous cases rather than making calls
• Batch progress: Show verification progress to user in manageable chunks
• Prioritize quotes over paraphrases: Direct quotes are highest risk for errors

1. Ask for the manuscript:

   "Please share the path to your manuscript (or the specific section you want verified)."

2. Identify source type:

   "Are we verifying quotes from interviews, cited literature, or both?"

3. Locate sources:

   "Where are your source materials? I need a folder path for interview transcripts or access to your literature (PDFs, Zotero collection, or document folder)."

4. Confirm scope:

   "Should I verify all quotes and claims, or focus on a specific section (e.g., Findings only)?"

5. Proceed with Phase 0 to assess verification scope.