Rich Narrative Generator (v2)

---
name: rich_narrative_v3
version: 3.1.0
description: Investment memo-style rich narrative with inline citations
last_updated: 2026-03-15
---

# Rich Narrative with Citations

Generate a comprehensive, citation-backed evaluation for a charity. This narrative
elaborates on the baseline evaluation with detailed analysis and source attribution.

## CRITICAL: Immutable Fields

The following fields are INHERITED from the baseline evaluation and MUST NOT be
regenerated or modified. They will be injected after your generation:

- headline (use exactly as provided)
- at_a_glance (use exactly as provided)
- amal_scores (use exactly as provided)
- zakat_guidance (use exactly as provided)

DO NOT include these fields in your output. Focus only on the enhanced narrative sections.

## Citation Requirements - Wikipedia Style

Use INLINE LINKED TEXT, not trailing citation markers. Wrap the specific claim text
in a cite tag with the citation ID:

- Format: `<cite id="1">linked claim text</cite>`
- The linked text should be the SPECIFIC CLAIM being cited (2-8 words typically)
- Minimum: 10 citations per narrative
- Target: 15-20 citations

### Correct (Wikipedia-style):
"Islamic Relief USA has maintained <cite id="1">a four-star rating from Charity Navigator</cite>
for over <cite id="2">15 consecutive years</cite>, reflecting strong financial health.
The organization reported <cite id="3">$165 million in revenue in 2023</cite> with
<cite id="4">83% directed to program expenses</cite>."

### WRONG (academic-style - DO NOT USE):
"Islamic Relief USA has maintained a four-star rating from Charity Navigator [1]..."

### Deep Linking Requirements

Citation URLs MUST be as specific as possible - link to the exact location where
the evidence can be found:

GOOD deep links:
- `https://www.charitynavigator.org/ein/954453134#accountability` (specific section)
- `https://projects.propublica.org/nonprofits/organizations/954453134` (org page)
- `https://irusa.org/zakat/#eligibility` (specific page section)

BAD shallow links:
- `https://www.charitynavigator.org/` (homepage only)
- `https://irusa.org` (no specific page)

NEVER use these URLs (broken/deprecated):
- Any guidestar.org URLs (site has been deprecated, URLs don't work)
- Any candid.org profile URLs (require internal IDs we don't have)

For Candid/GuideStar data (e.g., Gold Seal of Transparency):
- You can MENTION Candid seals in the narrative text
- Do NOT wrap Candid mentions in <cite> tags (no working URL available)
- Example: "The charity holds a Gold Seal of Transparency from Candid" (no citation)

For Charity Navigator data:
- ALWAYS cite CN ratings with the CN URL: https://www.charitynavigator.org/ein/{EIN}
- Example: <cite id="X">4-star rating from Charity Navigator</cite>

When constructing source_url in citations:
1. Use the FULL URL path to the specific page
2. Add anchor fragments (#section) when citing a specific part of a page
3. Include query parameters if they help locate the data
4. For Form 990s, link to the specific tax year filing, not the org overview

### Matching Claims to Deep Links

Choose the most specific source URL for each claim type:

| Claim Type | Best Source URL Pattern |
|------------|------------------------|
| Revenue, expenses, assets, financial data | Form 990 PDF or ProPublica filing |
| CEO/executive compensation | Form 990 PDF (most authoritative) |
| Ratings, accountability scores | Charity Navigator /ein/... |
| Transparency, organizational profile | Charity Navigator or Website /about |
| Program statistics, beneficiaries, outcomes | Annual Report PDF or Website /our-impact |
| Mission, history, founding year | Website /about or /about-us |
| Zakat eligibility, zakat policy | Website /zakat |
| School counts, student numbers, impact metrics | Impact Report PDF or Website /our-impact |
| Audited financial details | Audited Financials PDF |

### Source Priority (highest to lowest confidence)
1. **PDFs** (Form 990 PDF, Annual Report PDF, Audited Financials) - most authoritative
2. **Form 990 filing page** on ProPublica - official IRS data
3. **Rating agency pages** (Charity Navigator, Candid) - third-party verified
4. **Specific website pages** (/programs, /our-impact, /about-us) - charity's own claims

NEVER use the homepage (e.g., https://dil.org) for specific claims - always use the
specific page where that data can be found (e.g., https://dil.org/our-impact for
"214 schools" claim).

## Writing Style

**Readability (8th-10th grade target)**
- Keep sentences under 20 words. Break long sentences into two.
- Use common words: "use" not "utilize", "help" not "facilitate", "buy" not "procure"
- One idea per sentence. If a sentence has "and" connecting two ideas, split it.
- Active voice: "The charity served 5,000 families" not "5,000 families were served"
- For metric explanations: First sentence states the fact. Second sentence explains why it matters. Don't combine them with "which means" or "reflecting" — use a period.

**Human voice**
- Write like a journalist covering a beat, not a chatbot summarizing data.
- Have opinions. Name tradeoffs: "This focus on depth over breadth means..."
- Never use: "It's important to note", "Furthermore", "Moreover", "comprehensive
  approach", "robust framework", "holistic", "leveraging", "synergies",
  "multifaceted", "a testament to", "at the forefront", "serves as a beacon"

**No internal score leakage**
- Do NOT reveal internal assessment scores: no "AMAL score of 81", "impact score 37/50"
- DO use real financial data: "$907 cost per beneficiary", "80% program ratio",
  "100/100 Charity Navigator rating" — these are public data, not internal ratings
- Describe quality qualitatively: "strong financial health" not "scored 81/100"
- FORBIDDEN PHRASES (never write these): "AMAL score", "overall score of X/100", "impact score", "credibility score", "alignment score", "assessment score", "X/100 overall", "scored X out of 100", "rating of X/100" when referring to internal assessments
- SELF-CHECK: Before finishing, scan your entire narrative for any phrase containing "score" or "X/100" — if it's NOT a public rating (Charity Navigator, BBB, GiveWell), delete it

**Nuance and differentiation**
- Every metric must answer "so what?" — explain the real-world consequence
- Compare to sector norms: "80% program ratio beats the 75% sector average"
- Name tradeoffs: "High per-person cost reflects intensive clinical care, but limits reach"
- Differentiate: what does THIS charity do that similar ones don't?

## Output Structure

Generate a JSON object with ONLY these fields:

```json
{
  "summary": "500-800 word narrative with <cite id=\"1\">inline linked claims</cite> throughout...",

  "strengths": [
    {
      "point": "Brief strength description",
      "detail": "1-2 sentence elaboration with <cite id=\"1\">specific cited claims</cite>",
      "citation_ids": ["[1]", "[2]"]
    }
  ],

  "strengths_deep_dive": [
    "Detailed paragraph with <cite id=\"2\">inline citations</cite> on key claims..."
  ],

  "areas_for_improvement": [
    {
      "area": "Area needing improvement",
      "context": "Why this matters: <cite id=\"3\">specific data point</cite>",
      "citation_ids": ["[1]", "[2]"]
    }
  ],

  "amal_score_rationale": "The score reflects <cite id=\"1\">strong transparency ratings</cite> and <cite id=\"2\">efficient program delivery</cite>.",

  "dimension_explanations": {
    "credibility": {
      "explanation": "Donors can trust this charity due to <cite id=\"1\">perfect accountability scores</cite>, <cite id=\"2\">independent audits</cite>, and <cite id=\"3\">outcome tracking via Candid</cite>.",
      "improvement": "To strengthen credibility further, the charity could <cite id=\"7\">complete BBB Wise Giving Alliance accreditation</cite> and publish board meeting minutes.",
      "citation_ids": ["[1]", "[2]", "[3]", "[7]"]
    },
    "impact": {
      "explanation": "Operations show <cite id=\"4\">83% program expense ratio</cite> with <cite id=\"5\">low overhead</cite> and direct service delivery.",
      "improvement": "Impact measurement would benefit from <cite id=\"8\">tracking cost-per-beneficiary</cite> across individual programs and commissioning independent evaluations.",
      "citation_ids": ["[4]", "[5]", "[8]"]
    },
    "alignment": {
      "explanation": "Strong fit for Muslim donors with <cite id=\"6\">verified zakat eligibility</cite> and high cause urgency.",
      "improvement": "",
      "citation_ids": ["[6]"]
    }
  },

  "case_against": {
    "summary": "Limited concerns: <cite id=\"7\">no third-party impact evaluation</cite> yet.",
    "risk_factors": ["<cite id=\"7\">No RCT-level evidence available</cite>"],
    "mitigation_notes": "Organization is expanding outcome measurement.",
    "citation_ids": ["[7]"]
  },

  "peer_comparison": {
    "peer_group": "International humanitarian organizations",
    "differentiator": "Distinguished by <cite id=\"8\">40+ country reach</cite>.",
    "similar_orgs": ["Organization 1", "Organization 2"],
    "citation_ids": ["[8]"]
  },

  "donor_fit_matrix": {
    "zakat_status": "ZAKAT-ELIGIBLE or SADAQAH-ONLY",
    "zakat_asnaf_served": ["fuqara (poor)", "masakin (needy)"],
    "cause_area": "HUMANITARIAN",
    "geographic_focus": ["Palestine", "Yemen", "Syria"],
    "giving_style": "Direct service",
    "evidence_rigor": "Grade C - Some outcome tracking through Candid",
    "tax_deductible": true,
    "citation_ids": ["[1]", "[2]"]
  },

  "ideal_donor_profile": {
    "best_for_summary": "Best for donors who want to provide emergency humanitarian relief to Muslim communities in crisis zones with zakat-eligible giving.",
    "donor_motivations": [
      "Want to fulfill zakat obligations with confidence",
      "Care deeply about humanitarian crises affecting Muslims globally",
      "Value established organizations with proven delivery capacity",
      "Prefer direct aid over advocacy or capacity building"
    ],
    "giving_considerations": [
      "Large-scale operations mean individual gifts join a pooled fund",
      "Programs operate in regions with political instability and access challenges",
      "Organization has strong institutional presence but limited RCT-level evidence"
    ],
    "not_ideal_for": "Donors seeking hyperlocal US community impact or those who require rigorous academic evidence of effectiveness",
    "citation_ids": ["[1]", "[6]"]
  },

  "impact_evidence": {
    "theory_of_change": "IMPLICIT",
    "theory_of_change_summary": "How the charity creates impact...",
    "evidence_grade": "C",
    "evidence_grade_explanation": "Grade C means some outcome measurement but no rigorous evaluation...",
    "rct_available": false,
    "external_evaluations": ["Candid outcomes data"],
    "outcome_tracking_years": 3,
    "why_evidence_matters": "Evidence helps ensure donations create real impact...",
    "citation_ids": ["[3]", "[4]"]
  },

  "financial_deep_dive": {
    "annual_revenue": 165000000,
    "program_expense_ratio": 0.83,
    "admin_ratio": 0.10,
    "fundraising_ratio": 0.07,
    "reserves_months": 3.5,
    "peer_program_ratio_median": 0.90,
    "industry_program_ratio": 0.75,
    "peer_count": 18,
    "yearly_financials": [
      {"year": 2023, "revenue": 165000000, "expenses": 155000000, "net_assets": 50000000},
      {"year": 2022, "revenue": 128000000, "expenses": 130000000, "net_assets": 45000000},
      {"year": 2021, "revenue": 136000000, "expenses": 125000000, "net_assets": 47000000}
    ],
    "revenue_cagr_3yr": 10.0,
    "cn_financial_score": 99.0,
    "citation_ids": ["[5]", "[6]"]
  },

  "long_term_outlook": {
    "room_for_funding": "HIGH",
    "room_for_funding_explanation": "Large-scale humanitarian needs exceed current capacity...",
    "founded_year": 1993,
    "years_operating": 32,
    "maturity_stage": "Institution",
    "revenue_growth_3yr": 10.0,
    "strategic_priorities": ["Expand emergency response", "Strengthen domestic programs"],
    "citation_ids": ["[7]"]
  },

  "organizational_capacity": {
    "employees_count": 250,
    "volunteers_count": 5000,
    "ceo_name": "Anwar Khan",
    "ceo_compensation": 350000,
    "ceo_compensation_pct_revenue": 0.21,
    "board_size": 9,
    "independent_board_pct": 0.78,
    "has_conflict_policy": true,
    "has_financial_audit": true,
    "geographic_reach": "40+ countries, 8 major program areas",
    "programs_count": 8,
    "payroll_to_revenue_pct": 15.0,
    "staff_per_million_revenue": 1.5,
    "citation_ids": ["[8]", "[9]"]
  },

  "similar_organizations": [
    {"name": "Helping Hand for Relief", "differentiator": "Larger scale"},
    {"name": "Muslim Aid USA", "differentiator": "Higher program efficiency"},
    {"name": "ICNA Relief", "differentiator": "More domestic focus"}
  ],

  "grantmaking_profile": {
    "is_significant_grantmaker": true,
    "total_grants": 2500000,
    "domestic_grants": 1500000,
    "foreign_grants": 1000000,
    "grant_count": 45,
    "top_recipients": ["Partner Org 1 ($500K)", "Partner Org 2 ($300K)"],
    "regions_served": ["Yemen", "Palestine", "Syria"],
    "grant_strategy": "Primarily funds local implementing partners in crisis zones",
    "citation_ids": ["[10]"]
  },

  "bbb_assessment": {
    "meets_all_standards": true,
    "standards_met": 20,
    "standards_not_met": [],
    "governance_status": "PASS",
    "effectiveness_status": "PASS",
    "finances_status": "PASS",
    "audit_type": "audited",
    "summary": "Meets all 20 BBB Wise Giving Alliance Standards for Charity Accountability",
    "citation_ids": ["[11]"]
  },

  "data_confidence": {
    "form_990_tax_year": 2023,
    "ratings_last_updated": "2024-12",
    "website_last_crawled": "2025-01",
    "known_gaps": ["Missing 2024 Form 990", "No third-party impact evaluation"],
    "total_citations": 15,
    "unique_sources": 8,
    "confidence_score": 85
  },

  "all_citations": [
    {
      "id": "[1]",
      "claim": "four-star rating from Charity Navigator",
      "source_name": "Charity Navigator - Ratings",
      "source_url": "https://www.charitynavigator.org/ein/954453134#ratings",
      "source_type": "rating",
      "quote": "This charity has earned a 4/4 Star rating",
      "access_date": "2026-01-09",
      "confidence": 0.95
    },
    {
      "id": "[2]",
      "claim": "$165 million in revenue in 2023",
      "source_name": "Form 990 (2023)",
      "source_url": "https://projects.propublica.org/nonprofits/organizations/954453134/202312319349300211/full",
      "source_type": "form990",
      "quote": "Total Revenue: $165,235,589",
      "access_date": "2026-01-09",
      "confidence": 1.0
    },
    {
      "id": "[3]",
      "claim": "zakat eligibility verified",
      "source_name": "Islamic Relief USA - Zakat",
      "source_url": "https://irusa.org/zakat/#eligibility",
      "source_type": "website",
      "quote": "100% of Zakat donations used only for Zakat-eligible purposes",
      "access_date": "2026-01-09",
      "confidence": 0.9
    }
  ],

  "citation_stats": {
    "total_count": 15,
    "by_source_type": {"rating": 3, "form990": 4, "news": 2, "website": 6},
    "high_confidence_count": 12,
    "unique_sources": 8
  },

  "confidence": {
    "score": 85,
    "level": "HIGH",
    "data_gaps": ["Missing 2024 financials"],
    "data_freshness": "Most data from 2023",
    "sources_used": ["Form 990", "Charity Navigator", "Website"],
    "notes": "Strong data availability from multiple sources"
  }
}
```

## Writing Guidelines

### Summary Section (500-800 words)
- Open with organization's core mission and scale
- Use inline citations: `<cite id="X">specific claim</cite>` for all facts
- Every number, rating, or factual claim must be wrapped in a cite tag
- The linked text should be concise (2-8 words) - the specific claim being cited

### Cite Tag Usage Examples
CORRECT:
- "The organization has <cite id="1">$165 million in annual revenue</cite>."
- "They maintain <cite id="2">a perfect 100 accountability score</cite>."
- "Programs reach <cite id="3">over 40 countries</cite> worldwide."

WRONG (don't cite entire sentences):
- "<cite id="1">The organization has $165 million in annual revenue and serves millions.</cite>"

### Strengths (minimum 3)
- Each strength needs citation support with cite tags
- Be specific, not generic ("Strong 4-star rating" not just "Well-rated")
- Include quantitative evidence where available

### Dimension Explanations
- Three narrative dimensions: credibility, impact, alignment
- **impact**: How efficiently and effectively the charity delivers programs.
  Use the component details (Cost Per Beneficiary, Directness, Financial Health, etc.)
  and their improvement suggestions to ground your explanation.
- **alignment**: How well the charity fits Muslim donor priorities.
  Use component details (Muslim Donor Fit, Cause Urgency, Underserved Space, etc.).
- **credibility**: Why donors can trust this charity.
  Draw from Data Confidence signals (verification tier, transparency, data quality),
  governance data, and external evaluations.
- Explain the SUBSTANCE behind each dimension — what's strong, weak, and why
- Do NOT reveal internal assessment scores (e.g., "impact score 37/50", "AMAL score 81/100")
  — describe the assessment qualitatively instead
- Real financial data IS encouraged: "$907 cost per beneficiary", "80% program ratio",
  "100/100 Charity Navigator rating" — these are public data, not internal scores
- Wrap specific evidence in cite tags
- Each dimension needs explanation + inline citations
- **improvement**: For each dimension, include a constructive 1-2 sentence
  suggestion for what the charity can do to improve that dimension.
  Use cite tags to reference specific data points. Leave empty string "" if
  the dimension is already near-perfect (score >= 90% of max).
  Use the component-level improvement suggestions provided in Score Details.

### Case Against (Constructive Limitations)
- Focus ONLY on verifiable concerns from authoritative data:
  - Financial efficiency (if program expense ratio < 70%)
  - Transparency gaps (missing audits, limited reporting)
  - Scale limitations (geographic reach, capacity constraints)
  - Data gaps (missing outcomes, limited evidence base)
- DO NOT cite news articles, Wikipedia controversies, or political allegations
- Many Muslim charities face politically-motivated attacks from hostile sources
  - These are NOT valid concerns for donor guidance
- If no verifiable concerns exist, the case_against can be minimal
- Include mitigation where charity is addressing concerns

### Citation Quality
- ONLY cite authoritative, verifiable sources:
  - form990: IRS Form 990 data (ProPublica)
  - rating: Charity Navigator, BBB, Candid, GiveWell
  - evaluation: GiveWell, academic studies, third-party evaluations
  - website: Charity's own website and annual reports
- DO NOT cite:
  - News articles (too often biased against Muslim charities)
  - Wikipedia (editable, often contains hostile content)
  - Advocacy/watchdog sites (often politically motivated)

### CRITICAL: Claim-Quote Accuracy

Every citation must have a CLAIM that is DIRECTLY SUPPORTED by the QUOTE:

**Rule 1: No Subjective Qualifiers**
- DO NOT add adjectives like "rigorous", "exceptional", "perfect" unless they appear in the source
- BAD: claim="rigorous monitoring system" when quote="We track net usage"
- GOOD: claim="monitoring system that tracks net usage" matches quote="We track net usage"

**Rule 2: No Fabricated Sub-Scores**
- ONLY cite scores that appear in the provided data
- If you only have `overall_score: 94`, do NOT claim "100/100 Impact score" or any beacon sub-score
- BAD: "perfect 100/100 Impact & Measurement score" when data only has overall_score=94
- GOOD: "overall score of 94/100 from Charity Navigator"

**Rule 3: Quote Must Support Claim**
- The quote field must contain text that DIRECTLY supports the claim
- If you cannot find a quote that supports the claim, DO NOT make that claim
- BAD: claim="100% of donations go to programs" with quote="We are committed to efficiency"
- GOOD: claim="100% of donations go to programs" with quote="100% of public donations buy nets"

**Rule 4: Use Exact Numbers from Data**
- For scores, ratios, and metrics, use the EXACT values from the provided data
- DO NOT round, estimate, or "improve" numbers
- If overall_score=94.0, say "94/100", not "perfect score" or "nearly 100"

**Rule 5: Low/Zero/Missing Program Expense Ratio - Never Hallucinate**
- If program_expense_ratio is null, missing, 0%, or very low (<10%), this indicates INCOMPLETE DATA
- DO NOT invent or hallucinate any percentage (like 85%, 90%, etc.)
- DO NOT describe this as "efficient operations" or "lean" when you have no data
- DO NOT claim ANY specific program expense percentage when the data is null
- Instead, note that "expense breakdown data is unavailable" or "financial breakdown is incomplete"
- The effectiveness_score will be LOW (5 or less) in these cases - your narrative must be consistent
- BAD: "spending 85% of its budget on programs" when program_expense_ratio is null
- BAD: "dedicates 90% to program expenses" when program_expense_ratio is null
- GOOD: "Financial data for expense breakdown is limited" when program ratio is null

**Rule 6: Zakat Eligibility Must Match Source Data**
- Only claim a charity is "zakat-eligible" if `charity_claims_zakat: true` in the baseline
- If wallet_tag is "SADAQAH-ELIGIBLE", do NOT describe programs as "zakat-eligible"
- Do NOT include "Zakat-compliant" in donor motivations unless source data confirms it

## Investment Memo Sections (NEW in V3)

These sections help sophisticated donors make informed giving decisions.

### Using New Data Sources

The Investment Memo Data now includes additional sources. Use them as follows:

**BBB Wise Giving Alliance Standards** (if available):
- Include in `organizational_capacity` section: standards met, audit status, governance policies
- If charity fails standards, include in `case_against` with specific failed standards
- Use BBB Review URL in citations when referencing BBB data
- "Meets all 20 BBB standards" is a strong trust signal worth highlighting in strengths

**Grantmaking Activity** (if charity makes grants):
- If `is_grantmaker` is true, include grant activity in the summary narrative
- For charities where >30% of expenses go to grants, this is a key differentiator
- List top grant recipients in `similar_organizations` context when relevant
- Foreign grants indicate international reach

**Charity Navigator Ratings** (if available):
- Use CN beacon scores (Impact, Accountability, Culture, Leadership) in dimension explanations
- Include `overall_score` and `financial_score` in `financial_deep_dive`
- `fundraising_efficiency` (<$0.25 per $1 raised is excellent) belongs in effectiveness discussion
- `working_capital_ratio` (months of expenses in reserves) belongs in long_term_outlook
- CN CEO compensation can be used in organizational_capacity

**CRITICAL: CN Score Distinction**
- `overall_score` = the aggregate CN rating (average of all beacons) - use this when citing "CN rating"
- `cn_financial_score` = just the Accountability & Finance beacon score - NOT the overall rating
- NEVER cite `cn_financial_score` as the "overall" or "total" CN rating
- If `overall_score` is null/missing, do NOT claim the charity has a CN rating

**Candid Outcome Metrics** (if available):
- `metrics_count` and `max_years_tracked` indicate evidence quality
- Include in `impact_evidence.outcome_tracking_years`
- `evaluation_documents` count goes in `external_evaluations`
- Charting Impact data shows strategic clarity - mention if present

**Revenue Breakdown** (if available):
- Include in financial_deep_dive narrative
- Diversified revenue (not 100% donations) indicates sustainability
- Program service revenue indicates earned income capability

**Beneficiary Data** (if available):
- Include beneficiaries_served in summary and impact_evidence
- Cost metrics (e.g., "$3.50 per meal") go in effectiveness discussion

### Donor Fit Matrix
- Help donors quickly assess if this charity matches their priorities
- Use the provided wallet_tag for zakat_status
- List actual geographic coverage from website data
- Explain evidence_rigor based on the evidence grade

### Ideal Donor Profile (CRITICAL - NEW)

This is the MOST IMPORTANT section for differentiation. Generate a DONOR-CENTRIC
"best for" profile. This is NOT about what the charity does - it's about what
KIND of donor would be most aligned with this charity.

**Key principle**: Describe the DONOR, not the charity.

Consider:
- **Donor motivations**: Why would someone CHOOSE this charity over alternatives?
- **Giving style**: Long-term commitment vs one-time? Monthly vs annual?
- **Risk tolerance**: Does this charity suit cautious donors or those comfortable with uncertainty?
- **Values alignment**: Evidence-focused? Faith-driven? Community-oriented?
- **Not ideal for**: Be HONEST about who might prefer different options

**GOOD best_for_summary examples** (describes DONOR):
- "Best for donors who prioritize rigorous evidence of impact and want to save the most lives per dollar"
- "Best for donors who value Islamic institutional development and want to invest in long-term community infrastructure"
- "Best for donors who care about Palestinian humanitarian needs and want zakat-eligible direct aid"
- "Best for donors who want to fulfill zakat through established, large-scale humanitarian operations"

**BAD best_for_summary examples** (describes CHARITY - DO NOT USE):
- "Best for international humanitarian relief" ❌ (describes work, not donor)
- "Best for education in South Asia" ❌ (describes programs, not donor persona)
- "Best for poverty alleviation" ❌ (describes cause, not donor profile)

**donor_motivations** should answer: "I want to donate here because I..."
**giving_considerations** should be honest about: scale, evidence gaps, access challenges, etc.
**not_ideal_for** should be constructive: "Donors seeking X may prefer alternatives"

### Impact Evidence
- Explain what the evidence grade means in plain language
- theory_of_change: PUBLISHED (has explicit TOC), DOCUMENTED (implied in materials), IMPLICIT (unclear)
- If no RCTs exist, say so honestly
- Include educational note about why evidence matters

### Financial Deep-Dive
- Use actual Form 990 data for yearly_financials
- Compare program_expense_ratio to peer_program_ratio_median
- Calculate revenue_cagr_3yr from the 3-year trend
- Show how this charity compares to peers, not just industry standards

### Long-Term Outlook
- room_for_funding: HIGH (can absorb more), MEDIUM (some capacity), LOW (near capacity), UNKNOWN
- maturity_stage: Startup (<5 yrs), Growth (5-15 yrs), Established (15-30 yrs), Institution (30+ yrs)
- Include strategic priorities if available from website

### Organizational Capacity
- Use Form 990 data for employees, compensation
- Calculate ceo_compensation_pct_revenue = (ceo_compensation / total_revenue) * 100
- Use Candid data for board size and governance policies

### Similar Organizations
- Use the provided similar_orgs from Investment Memo Data
- Each similar org should have a clear differentiator

### Data Confidence
- Be honest about data gaps
- confidence_score: 90+ (excellent), 70-89 (good), 50-69 (moderate), <50 (limited)

## Consistency Rules

- DO NOT contradict baseline scores or facts
- DO NOT recalculate AMAL scores
- If agent discoveries reveal conflicting information, present as nuance, not contradiction
- Reference baseline wallet tag correctly (ZAKAT-ELIGIBLE or SADAQAH-ELIGIBLE)

## Data Provided

{charity_data}

## Available Citation Sources

{citation_sources}

## Baseline Context

{baseline_context}