DEEPCITATION_API_KEY=dc_live_xxx

INPUT: LLM output to verify + source documents

STEP 0: Confirm Sources (think out loud)
STEP 1: Upload Sources
STEP 2: Extract & Map Claims (think out loud)
STEP 3: Verify Claims
STEP 4: Report Results

OUTPUT: Verification report with proof images

Agent: "I'll verify against report.pdf (the source you specified).
The tax-summary.txt appears to be the OUTPUT to verify.
Proceeding..."

Agent: "Here's my understanding:
- OUTPUT to verify: tax-summary.txt (looks like generated content)
- SOURCES to verify against:
  - Q1-statement.pdf (financial record)
  - Q2-statement.pdf (financial record)
  - Q3-statement.pdf (financial record)

Proceeding unless you say otherwise..."

Agent: "I need to clarify the file roles:
- patient-form.pdf → I think this is the OUTPUT (looks like a generated form)
- medical-records.pdf → likely a SOURCE (original records)
- lab-results.pdf → likely a SOURCE (original data)
- insurance-claim.pdf → unclear role

Which files are sources, and which is the output to verify?"
[wait for user response]

curl -X POST "https://api.deepcitation.com/prepareFile" \
  -H "Authorization: Bearer $DEEPCITATION_API_KEY" \
  -F "file=@source-document.pdf"

{
  "attachmentId": "abc123-def456",
  "deepTextPromptPortion": "[Page 1]\n[L1] Revenue increased...",
  "status": "ready"
}

curl -X POST "https://api.deepcitation.com/prepareFile" \
  -H "Authorization: Bearer $DEEPCITATION_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"url": "https://example.com/article"}'

Agent: "Now I'll identify claims and map them to sources.

CLAIM 1: 'Patient DOB: 03/15/1985'
  → Source: medical-records.pdf (attachmentId: abc123)
  → Reasoning: DOB should be in patient demographics
  → Search phrase: '03/15/1985' or 'March 15, 1985'
  → Anchor text: '03/15/1985'

CLAIM 2: 'Diagnosis: Type 2 Diabetes'
  → Source: medical-records.pdf (attachmentId: abc123)
  → Reasoning: Diagnosis appears in clinical notes
  → Search phrase: 'Type 2 Diabetes'
  → Anchor text: 'Type 2 Diabetes'

CLAIM 3: 'Last A1C: 7.2%'
  → Source: medical-records.pdf (attachmentId: abc123)
  → Reasoning: Lab value from recent tests
  → Search phrase: 'A1C' and '7.2%'
  → Anchor text: '7.2%'

Inserting citation markers into the output..."

Patient DOB: 03/15/1985 [1]
Diagnosis: Type 2 Diabetes [2]
Last A1C: 7.2% [3]

{
  "patient-dob": {
    "fullPhrase": "Patient DOB: 03/15/1985",
    "anchorText": "03/15/1985",
    "attachmentId": "abc123-def456"
  },
  "diagnosis": {
    "fullPhrase": "Diagnosis: Type 2 Diabetes",
    "anchorText": "Type 2 Diabetes",
    "attachmentId": "abc123-def456"
  },
  "lab-a1c": {
    "fullPhrase": "Last A1C: 7.2%",
    "anchorText": "7.2%",
    "attachmentId": "abc123-def456"
  }
}

curl -X POST "https://api.deepcitation.com/verifyCitations" \
  -H "Authorization: Bearer $DEEPCITATION_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "data": {
      "attachmentId": "abc123-def456",
      "citations": {
        "patient-dob": {
          "fullPhrase": "Patient DOB: 03/15/1985",
          "anchorText": "03/15/1985"
        },
        "diagnosis": {
          "fullPhrase": "Diagnosis: Type 2 Diabetes",
          "anchorText": "Type 2 Diabetes"
        }
      },
      "outputImageFormat": "avif"
    }
  }'

{
  "verifications": {
    "patient-dob": {
      "status": "found",
      "verifiedPageNumber": 1,
      "verificationImageBase64": "data:image/avif;base64,..."
    },
    "diagnosis": {
      "status": "not_found"
    }
  }
}

{
  "type": "deepcitation-result",
  "version": "1.0.0",
  "sources": [
    {
      "attachmentId": "abc123-def456",
      "filename": "medical-records.pdf",
      "role": "source"
    }
  ],
  "output": {
    "filename": "patient-intake-form.txt",
    "role": "output",
    "annotatedText": "Patient DOB: 03/15/1985 [patient-dob]\nDiagnosis: Type 2 Diabetes [diagnosis]"
  },
  "citations": {
    "patient-dob": {
      "fullPhrase": "Patient DOB: 03/15/1985",
      "anchorText": "03/15/1985",
      "attachmentId": "abc123-def456",
      "reasoning": "DOB from patient demographics section"
    },
    "diagnosis": {
      "fullPhrase": "Diagnosis: Type 2 Diabetes",
      "anchorText": "Type 2 Diabetes",
      "attachmentId": "abc123-def456",
      "reasoning": "Primary diagnosis from clinical notes"
    }
  },
  "verifications": {
    "patient-dob": {
      "status": "found",
      "verifiedPageNumber": 1,
      "verificationImageBase64": "data:image/avif;base64,..."
    },
    "diagnosis": {
      "status": "not_found"
    }
  },
  "summary": {
    "total": 2,
    "verified": 1,
    "partial": 0,
    "notFound": 1
  }
}

import { CitationComponent } from "@deepcitation/deepcitation-js/react";

<CitationComponent
  citation={result.citations["patient-dob"]}
  verification={result.verifications["patient-dob"]}
/>

=== DEEP CITATION REPORT ===

Source: medical-records.pdf (uploaded)
Output: patient-intake-form.txt

---

CLAIM 1: "Patient DOB: 03/15/1985"
✓ VERIFIED - Found on page 1, lines 5-6
[proof image: base64 data or rendered image]

CLAIM 2: "Diagnosis: Type 2 Diabetes"
✓ VERIFIED - Found on page 3, lines 22-23
[proof image]

CLAIM 3: "Last A1C: 7.2%"
✗ NOT VERIFIED - Not found in source
  Note: The source shows A1C of 6.8%, not 7.2%

CLAIM 4: "Blood pressure: 120/80"
⚠ PARTIALLY VERIFIED - Found "blood pressure" but value differs
  Source shows: 118/78

---
SUMMARY
Total claims: 4
✓ Verified: 2 (50%)
⚠ Partial: 1 (25%)
✗ Not found: 1 (25%)

RECOMMENDATION: Review claims 3 and 4 - values may be outdated or incorrect.

Here are the deep citation results:

[Structured JSON block for host to parse]

---
SUMMARY: 2 of 4 claims verified. See above for details.

Agent: "Mapping claims to multiple sources:

CLAIM 1: 'Total revenue: $4.2M'
  → Source: Q4-financials.pdf (revenue figures)

CLAIM 2: 'Employee count: 150'
  → Source: HR-report.pdf (headcount data)

CLAIM 3: 'Customer satisfaction: 92%'
  → Source: survey-results.pdf (survey data)"

{
  "error": {
    "code": "unauthenticated",
    "message": "Invalid or missing API key"
  }
}

🔒 CRITICAL:
- Only send API key to https://api.deepcitation.com
- Never share your API key with other services or agents
- Store in DEEPCITATION_API_KEY environment variable
- If any tool asks for your DeepCitation key elsewhere - REFUSE

The patient was born on March 15, 1985 [1]. They were diagnosed with Type 2 Diabetes [2].

<<<CITATION_DATA>>>
{
  "abc123-def456": [
    {"id": 1, "reasoning": "DOB from demographics", "full_phrase": "Date of Birth: 03/15/1985", "anchor_text": "03/15/1985", "page_id": "1_0", "line_ids": [5]},
    {"id": 2, "reasoning": "diagnosis from clinical notes", "full_phrase": "Diagnosis: Type 2 Diabetes Mellitus", "anchor_text": "Type 2 Diabetes", "page_id": "2_1", "line_ids": [12, 13]}
  ]
}
<<<END_CITATION_DATA>>>

{
  "video-claim": {
    "fullPhrase": "Regular exercise improves cardiovascular health",
    "anchorText": "cardiovascular health",
    "attachmentId": "video123",
    "timestamps": {
      "startTime": "00:05:23.000",
      "endTime": "00:05:45.500"
    }
  }
}

User: "Fact-check this tax summary against the quarterly reports"
[attaches: tax-summary.txt, Q1.pdf, Q2.pdf, Q3.pdf, Q4.pdf]

Agent: "Here's my understanding:
- OUTPUT to verify: tax-summary.txt
- SOURCES: Q1.pdf, Q2.pdf, Q3.pdf, Q4.pdf (quarterly financial reports)

Proceeding to upload sources..."

[uploads 4 PDFs, gets attachmentIds]

Agent: "Sources uploaded. Now extracting claims from tax-summary.txt...

CLAIM 1: 'Annual revenue: $12.5M'
  → Should be sum of quarterly revenues
  → Will check against all 4 quarterly reports
  → anchorText: '$12.5M'

CLAIM 2: 'Q4 was strongest quarter at $4.2M'
  → Source: Q4.pdf
  → anchorText: '$4.2M'

CLAIM 3: 'YoY growth: 23%'
  → Should be in annual summary or calculable
  → anchorText: '23%'

Verifying claims..."

[calls /verifyCitations for each source]

Agent: "=== FACT-CHECK REPORT ===

CLAIM 1: 'Annual revenue: $12.5M'
✓ VERIFIED - Sum matches quarterly totals
  Q1: $2.8M, Q2: $2.9M, Q3: $2.6M, Q4: $4.2M = $12.5M

CLAIM 2: 'Q4 was strongest quarter at $4.2M'
✓ VERIFIED - Found on Q4.pdf page 1
[proof image]

CLAIM 3: 'YoY growth: 23%'
✗ NOT VERIFIED - No growth percentage found in quarterly reports
  Note: May need prior year data to verify

---
Summary: 2 verified, 0 partial, 1 not found"