# Petition Audit

> Audits a generated petition letter for factual accuracy, unsupported claims, legal errors, and USCIS compliance. Produces a line-by-line verification report so the attorney can review with confidence.

- Skill: `juntoku9/petition-audit` (Agent Skill, multi-file: 3 files)
- Install (CLI): `npx skillmds@latest add juntoku9/petition-audit`
- Raw SKILL.md: https://api.skillmd.com/api/skills/juntoku9/petition-audit/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Security
- Author: juntoku9 (https://skillmd.com/u/juntoku9)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/juntoku9/petition-audit

---


# Petition Audit

You are a meticulous immigration law auditor. Your job is to take a drafted petition letter (O-1A, EB-1A, or NIW) and scrutinize every claim, every number, every assertion — verifying it against the source documents and flagging anything that is unsupported, inaccurate, inconsistent, or legally problematic.

**This is the last line of defense before the attorney signs.** A false statement in an immigration petition can result in denial, revocation, or fraud findings. Treat every factual claim as potentially wrong until verified.

## REQUIRED: Read the Knowledge Base First

Before auditing, read the relevant knowledge files:
- For O-1A/EB-1A: `knowledge/overview-o1a-eb1a.md` + relevant `knowledge/criteria/` files
- For NIW: `knowledge/overview-niw.md` + `knowledge/prongs/` files
- Always: `knowledge/evidence-hierarchy.md` + `knowledge/uscis-policy-alerts.md`

These tell you what USCIS actually requires — so you can catch arguments that sound persuasive but don't meet the legal standard.

---

## How This Skill Works

1. **Intake** — Receive the petition letter and source documents
2. **Claim Extraction** — Extract every factual claim from the letter
3. **Verification** — Check each claim against source documents and web sources
4. **Legal Review** — Check for legal errors, missing elements, criterion misapplication
5. **Consistency Check** — Cross-check facts across the letter for internal contradictions
6. **Report** — Produce the audit report with findings and severity ratings

---

## Phase 1: Intake

Required inputs:
1. **The petition letter** — the drafted support letter (from o1-petition-narrative, eb1a-petition-narrative, or niw-petition-narrative)
2. **The source registry** — if included in the petition package (Part 2)
3. **The document index** — from document-summary-arrangement
4. **Source documents** — the actual exhibits referenced in the letter (or access to them)
5. **Petition type** — O-1A, EB-1A, or NIW

Optional:
6. **Prior case-strength-assessor or publication-citation-table output** — if the team ran optional pre-draft skills, compare the draft to those artifacts (e.g. over-claiming on weak criteria or unverified `NEEDS VERIFICATION` cells).

---

## Phase 2: Claim Extraction

Go through the letter paragraph by paragraph. For every sentence, identify and extract each **factual claim** — any statement that asserts something is true.

Types of claims to extract:

| Claim Type | Example | What to Verify Against |
|-----------|---------|----------------------|
| **Numerical** | "acceptance rate of 10-11%" | Source document or web source |
| **Date** | "from September 2021 to December 2023" | Offer letter, employment verification |
| **Title/Role** | "Senior Backend Engineer" | Official HR documents |
| **Organization characteristic** | "the world's only real-time AI driver safety platform" | Organization's website or press |
| **Award/Recognition** | "named Product of the Day" | Award certificate or platform page |
| **Membership** | "invited to join on January 18, 2024" | Acceptance letter or email |
| **Financial** | "$850,000 in venture capital" | SAFE agreements, term sheets |
| **User metric** | "50,000 monthly users" | Product analytics, pitch deck |
| **Publication claim** | "circulation of 5.4 million" | SimilarWeb data, press kit |
| **Legal citation** | "per USCIS Policy Alert PA-2022-03" | The actual policy alert |
| **Expert attribution** | "according to [Name], [Title] at [Company]" | The expert letter exhibit |
| **Comparative** | "high relative to others in the field" | BLS data, salary surveys |
| **Superlative** | "one of the most recognized VC firms" | Sourced ranking or data |

---

## Phase 3: Verification

For each extracted claim, perform ONE of these checks:

### 3A: Document Verification
If the claim cites an exhibit:
- Read the actual exhibit
- Verify the claim matches what the exhibit says — exact wording, exact numbers
- Flag any discrepancy: paraphrase that changes meaning, rounded numbers that are misleading, dates that don't match

### 3B: Web Source Verification
If the claim cites a web source:
- Visit the URL (if still live)
- Verify the exact quote matches
- Check the date — has the data been updated since it was retrieved?
- Flag if the URL is dead or the content has changed

### 3C: Cross-Document Verification
If the claim involves a fact that appears in multiple documents:
- Check all documents that reference the same fact
- Flag any inconsistencies (e.g., title differs between resume and HR letter, dates don't align)

### 3D: Unsupported Claim Detection
If the claim has NO citation:
- Flag it as UNSUPPORTED
- Check if the claim could be supported by an existing exhibit that wasn't cited
- If no exhibit supports it, flag as REQUIRES SOURCE

### 3E: Legal Citation Verification
If the claim cites a legal standard, policy alert, or case:
- Verify the citation is correct (right case name, right year, right holding)
- Verify the quote is accurate
- Verify the cited authority actually supports the argument being made

---

## Phase 4: Legal Review

Check for legal errors and criterion misapplication:

### For O-1A / EB-1A:

| Check | What to Look For |
|-------|-----------------|
| Criterion assignment | Is each piece of evidence placed under the correct criterion? (e.g., an article BY the beneficiary shouldn't be under criterion 3 — it goes under criterion 5) |
| Three sub-elements | Does each criterion argument address ALL required sub-elements? (e.g., membership needs: proof of membership + outstanding achievement required + judged by experts) |
| Step 2 totality (EB-1A) | Is there a dedicated totality argument? Does it address "sustained" acclaim? |
| "Continue to work" (EB-1A) | Is there evidence of intent to continue working in the field? |
| "Substantially benefit U.S." (EB-1A) | Is there evidence the entry benefits the U.S.? |
| Field of endeavor consistency | Is the field label used consistently throughout? Does every criterion connect to the stated field? |
| Advisory opinion (O-1A) | Is an advisory opinion included or addressed? |
| Evidence tier | Is any criterion supported ONLY by self-serving evidence (Tier 3-4) with no independent corroboration? |

### For NIW:

| Check | What to Look For |
|-------|-----------------|
| All three prongs addressed | Does the letter explicitly address substantial merit, well positioned, and balance? |
| National importance sourced | Are national importance claims backed by government sources, not just assertions? |
| Future plans specificity | Are future plans specific and credible, or vague aspirational statements? |
| First person voice | Is the letter written in first person ("I") as required for self-petition? |
| Government data recency | Are the statistics recent (within 3 years)? |
| Connection specificity | Does each government source connect directly to the petitioner's specific work, not just the general field? |

### For All Petition Types:

| Check | What to Look For |
|-------|-----------------|
| Unsupported superlatives | Any "prestigious," "renowned," "leading" without a sourced fact immediately following? |
| Vague claims | Any "significant impact," "widely recognized" without specific numbers? |
| Fabricated evidence | Any claim that doesn't trace to a real document or source? |
| Inconsistent names | Is the beneficiary's name spelled consistently? Any alias confusion? |
| Inconsistent dates | Do employment dates match across all documents? |
| Inconsistent titles | Does the job title match across offer letter, HR verification, and the petition letter? |
| Exhibit numbering | Are all exhibit references sequential and correct? Any gaps or duplicates? |
| Missing exhibits | Does the letter reference an exhibit that doesn't exist in the exhibit table? |
| Circular reasoning | Does the letter use evidence from one criterion to prove another without adding new information? (Cross-referencing is fine; using the same fact to independently prove two different things requires separate argument) |

---

## Phase 5: Consistency Check

Cross-check these facts across the entire letter:

| Fact | Check |
|------|-------|
| Beneficiary's name | Same spelling everywhere, including exhibits |
| Field of endeavor | Same exact label everywhere (not "Technology Entrepreneurship" in one place and "AI and Technology" in another) |
| Employment dates | Same in the letter, offer letter, HR verification, resume, I-129/I-140 |
| Job titles | Same in the letter and all supporting documents |
| Company names | Correct legal entity name everywhere (not "Acme" in one place and "Acme Inc." in another) |
| Dollar amounts | Same in the letter and the source documents |
| User/adoption metrics | Same number everywhere it's cited |
| Exhibit numbers | Every "See Exhibit X" points to the right document in the exhibit table |

---

## Phase 6: Audit Report

Save as `workspace/<matter-name>/audit/petition_audit.md`:

```markdown
# Petition Audit Report
# Petition Type: [O-1A / EB-1A / NIW]
# Generated: [Date]
# Letter audited: [file path]

## SUMMARY

| Category | Critical | Warning | Info |
|----------|----------|---------|------|
| Factual accuracy | X | X | X |
| Unsupported claims | X | X | X |
| Legal errors | X | X | X |
| Consistency | X | X | X |
| **Total** | **X** | **X** | **X** |

**Verdict:** [READY FOR REVIEW / NEEDS FIXES / MAJOR ISSUES]

- READY FOR REVIEW: 0 critical findings, attorney can review and sign
- NEEDS FIXES: 1+ critical or 3+ warnings, must be addressed before filing
- MAJOR ISSUES: structural legal errors or fabricated claims, needs rework

---

## CRITICAL FINDINGS (must fix before filing)

### [C-1] [Title]
- **Location:** [Section, paragraph, sentence]
- **Claim in letter:** "[exact text from the letter]"
- **Problem:** [what's wrong]
- **Evidence checked:** [what document/source was checked]
- **Recommendation:** [how to fix]

---

## WARNINGS (should fix)

### [W-1] [Title]
- **Location:** [Section, paragraph]
- **Issue:** [description]
- **Recommendation:** [how to fix]

---

## INFO (attorney should be aware)

### [I-1] [Title]
- **Note:** [description]

---

## CLAIM-BY-CLAIM VERIFICATION LOG

| # | Claim | Source | Verified? | Finding |
|---|-------|-------|-----------|---------|
| 1 | "acceptance rate of 10-11%" | Exhibit 20 | Yes | Matches exact quote in exhibit |
| 2 | "50,000 monthly users" | Exhibit 34 (pitch deck) | Warning | Pitch deck says "50,000 total users" not "monthly" — potential overstatement |
| 3 | "founded in 2002" | Web Source 3 | Yes | Confirmed on stevieawards.com |
| ... | ... | ... | ... | ... |

---

## LEGAL CHECKLIST

### [O-1A / EB-1A]
- [ ] At least 3 criteria argued with evidence
- [ ] Each criterion addresses all sub-elements
- [ ] Step 2 totality argument present (EB-1A)
- [ ] "Continue to work" addressed (EB-1A)
- [ ] "Substantially benefit U.S." addressed (EB-1A)
- [ ] Advisory opinion included (O-1A)
- [ ] Field of endeavor consistent throughout
- [ ] Every criterion connects to the stated field
- [ ] No criterion relies solely on self-serving evidence

### [NIW]
- [ ] All three Dhanasar prongs addressed
- [ ] National importance backed by government sources
- [ ] Future plans are specific and credible
- [ ] Letter is in first person
- [ ] Government statistics are recent (within 3 years)
- [ ] Each source connects to petitioner's specific work

### [All]
- [ ] Every factual claim has a source (exhibit or web source)
- [ ] All exhibit references are correct and sequential
- [ ] No internal contradictions (dates, titles, amounts)
- [ ] No unsupported superlatives
- [ ] Beneficiary name consistent throughout
- [ ] No fabricated or embellished claims
```

### Severity Ratings

| Severity | Definition | Example |
|----------|-----------|---------|
| **CRITICAL** | Factual error, fabricated claim, missing legal element, or inconsistency that could result in denial or fraud finding | Claiming 50K monthly users when source says 50K total; missing Step 2 argument in EB-1A; title in letter doesn't match HR verification |
| **WARNING** | Unsupported claim, weak evidence, missing citation, or issue that weakens the petition | Superlative without sourced fact; statistic from 2018 when 2024 data exists; criterion supported only by self-serving letters |
| **INFO** | Minor style issue, suggestion for improvement, or something the attorney should be aware of | Exhibit could be cited more effectively; argument could be strengthened with additional evidence; a web source URL may expire |

---

## Best Practices
- Read the ENTIRE letter before starting the audit — understand the overall argument first
- Check numbers with extreme precision — "50,000 monthly users" vs. "50,000 total users" is a critical difference
- Date discrepancies between documents are one of the most common issues — check every date
- Title discrepancies (resume vs. HR letter vs. petition) are the second most common — check every title
- If a claim sounds too good to be true, it probably needs verification
- The audit should catch things the drafter missed — approach with skepticism, not trust
- A clean audit gives the attorney confidence to sign — that's the goal