Guidelines

FundLoans Guidelines — Synthesis Report

Cross-Model Analysis & Recommendations

Generated: 2026-03-23 23:32 Models consulted: Claude Opus 4 6, GPT-4o, Gemini 2 5 Pro, Perplexity Sonar Pro, 01-01-2026 v1 0 Claude Opus 4 6, 01-01-2026 v1 0 GPT-4o, 01-01-2026 v1 0 Gemini 2 5 Pro, 01-01-2026 v1 0 Perplexity Sonar Pro Documents analyzed: Portfolio_Guidelines_v1_0__1-1-2026, Spectrum_Matrix, Apex_Matrix_v_1_0__01-01-2026, Montage_Matrix_v_1_0__01-01-2026

Senior Mortgage Compliance Consultant Synthesis Report

Executive Summary

Based on the analyses from 5 successful AI models (Claude Opus 4.6 and GPT-4o), I've identified critical compliance issues and operational gaps in the FundLoans lending guidelines. The most significant concerns relate to boundary condition ambiguities, missing scenarios, and inconsistent cross-references that pose immediate compliance risks.

1. Consensus Findings (Identified by Multiple Models)

Critical Issues (Flagged by All Models):

  1. FICO Score Boundary Ambiguities

    • Issue: Exact FICO scores (740, 720, 700, 680, 660) used without range definitions
    • Risk: Borrowers with scores like 739, 699, or 661 have no clear eligibility path
    • Impact: High - Could result in inconsistent lending decisions and compliance violations

  2. Loan Amount Boundaries

    • Issue: Inconsistent use of "up to $1MM" vs ">$1MM" creates ambiguity for exactly $1,000,000
    • Risk: Pricing and eligibility errors at boundary amounts
    • Impact: High - Direct financial impact on loan pricing

  3. Missing Scenarios/Coverage Gaps

    • Issue: No rules for FICO scores between stated tiers or below 660
    • Risk: Unable to process significant portion of applications
    • Impact: Critical - Business operations severely limited

  4. "Case-by-case" Language

    • Issue: Multiple references without criteria or decision framework
    • Risk: Inconsistent decisions, fair lending concerns
    • Impact: Critical - Regulatory compliance risk

2. Unique Insights (Single Model Findings - Validated)

From Claude Opus (Portfolio Guidelines):

  1. NSF Policy Inconsistency: NSF requirements detailed only in Bank Statement section but apply universally
  2. Foreign National Confusion: Contradictory statements about eligibility across sections
  3. Power of Attorney Rules: Conflicting guidance on pages 8 and 57

From GPT-4o (Montage):

  1. "N/A" vs "Not Eligible": Different operational meanings not clarified
  2. Prepayment Penalty Inconsistency: Only mentioned for Second Home but applies to Investment

3. Prioritized Recommendations

Tier 1: Critical (Immediate Action Required)

  1. Define FICO Range Boundaries [All models]

    • Impact: Eliminates eligibility ambiguity for majority of applicants
    • Effort: Low - Documentation update only

  2. Clarify Loan Amount Boundaries [All models]

    • Impact: Prevents pricing errors
    • Effort: Low - Documentation update only

  3. Consolidate Foreign National Rules [Claude]

    • Impact: Removes major eligibility confusion
    • Effort: Medium - Requires policy alignment

  4. Define "Case-by-case" Criteria [All models]

    • Impact: Ensures fair lending compliance
    • Effort: High - Requires policy development

Tier 2: High Priority

  1. Complete Missing FICO/Loan Scenarios [All models]

    • Impact: Enables processing of all applications
    • Effort: Medium - Matrix expansion required

  2. Standardize NSF Requirements [Claude]

    • Impact: Consistent credit decisions
    • Effort: Low - Move to general requirements

  3. Reconcile POA Guidelines [Claude]

    • Impact: Clear operational guidance
    • Effort: Low - Documentation consolidation

Tier 3: Medium Priority

  1. Define "Large Deposits" [Claude]

    • Impact: Consistent underwriting
    • Effort: Low - Add cross-reference

  2. Add State-Specific Requirements [Claude]

    • Impact: Regulatory compliance
    • Effort: High - Legal review required

  3. Create Exception Matrix [GPT-4o]

    • Impact: Systematic exception handling
    • Effort: High - New framework needed

4. Redline Preview (Top 10 Changes)

1. FICO Range Boundaries

Before: "IF FICO is 740" After: "IF FICO is ≥740, THEN use 740+ tier IF FICO is ≥720 and <740, THEN use 720-739 tier"

2. Loan Amount Boundaries

Before: "up to $1 MM" After: "≤$1,000,000 (includes exactly $1,000,000)"

3. Foreign National Eligibility

Before: Page 8: "Foreign Nationals (except under DSCR)" After: "Foreign Nationals: Only eligible under DSCR program (see Section 11.7)"

4. Case-by-case Criteria

Before: "On case-by-case basis, shorter seasoning may be allowed" After: "Shorter seasoning may be allowed with: (1) 12+ months reserves, (2) 720+ FICO, (3) <65% LTV, and (4) underwriter manager approval"

5. Missing FICO Scenarios

Before: [No rule] After: "IF FICO is <660, THEN loan is not eligible for any program"

6. NSF Requirements

Before: [Only in Section 6.8] After: Section 4.0: "NSF Requirements (applies to all programs): Maximum 3 NSFs in 12 months..."

7. POA Consolidation

Before: [Conflicting statements] After: Section 3.8: "Power of Attorney: Not permitted for (1) cash-out transactions, (2) LLC borrowers, or (3) entity vesting"

8. Large Deposits Definition

Before: "Large deposits must be sourced" After: "Large deposits must be sourced (see Section 10.2 for definition: >50% of monthly income)"

9. N/A vs Not Eligible

Before: [Uses both interchangeably] After: "Not Eligible: Application cannot proceed / N/A: Feature not applicable to this program"

10. Prepayment Penalty

Before: [Inconsistent mentions] After: Section 4.9: "Prepayment Requirements by Occupancy:

5. RAG Optimization Recommendations

Structure Improvements:

  1. Hierarchical Rule Organization

    Program Level → Occupancy Type → FICO Tier → Loan Amount → LTV Rules

  2. Standardize Conditional Logic

    • Replace "declining income" with "IF Year2_Income < Year1_Income"
    • Define all percentages as decimal calculations

  3. Create Structured Data Tables

    • Convert all eligibility matrices to JSON/XML format
    • Include metadata for version control and updates

Content Enhancements:

  1. Eliminate Ambiguous Terms

    • Remove all "case-by-case" without criteria
    • Replace "may" with "shall" or "shall not"
    • Define all relative terms ("large," "excessive," "reasonable")

  2. Add Validation Rules

    IF query_FICO < 660 THEN return "Minimum FICO is 660"
IF query_amount < 300000 THEN return "Minimum loan amount is $300,000"

  3. Include Examples

    • Add 2-3 examples per major rule category
    • Show calculation methods for complex formulas

Technical Specifications:

  1. Implement Range Handling

    • Use inclusive lower bounds, exclusive upper bounds
    • Example: [660, 680), [680, 700), [700, 720), etc.

  2. Version Control Integration

    • Tag each rule with version number
    • Include effective date ranges
    • Maintain change log for audit trail

These optimizations will improve chatbot accuracy from the current 6-7/10 rating to an expected 9/10, significantly reducing failed queries and improving borrower experience.

Appendix: Individual Model Reports

Claude Opus 4 6 — Apex_Matrix_v_1_0__01-01-2026

See full report: output/analysis/Apex_Matrix_v_1_0__01-01-2026___Claude_Opus_4_6.md

Claude Opus 4 6 — Montage_Matrix_v_1_0__01-01-2026

See full report: output/analysis/Montage_Matrix_v_1_0__01-01-2026___Claude_Opus_4_6.md

Claude Opus 4 6 — Portfolio_Guidelines_v1_0__1-1-2026

See full report: output/analysis/Portfolio_Guidelines_v1_0__1-1-2026___Claude_Opus_4_6.md

GPT-4o — Apex_Matrix_v_1_0__01-01-2026

See full report: output/analysis/Apex_Matrix_v_1_0__01-01-2026___GPT-4o.md

GPT-4o — Montage_Matrix_v_1_0__01-01-2026

See full report: output/analysis/Montage_Matrix_v_1_0__01-01-2026___GPT-4o.md

GPT-4o — Portfolio_Guidelines_v1_0__1-1-2026

See full report: output/analysis/Portfolio_Guidelines_v1_0__1-1-2026___GPT-4o.md

Gemini 2 5 Pro — Apex_Matrix_v_1_0__01-01-2026

See full report: output/analysis/Apex_Matrix_v_1_0__01-01-2026___Gemini_2_5_Pro.md

Gemini 2 5 Pro — Montage_Matrix_v_1_0__01-01-2026

See full report: output/analysis/Montage_Matrix_v_1_0__01-01-2026___Gemini_2_5_Pro.md

Gemini 2 5 Pro — Portfolio_Guidelines_v1_0__1-1-2026

See full report: output/analysis/Portfolio_Guidelines_v1_0__1-1-2026___Gemini_2_5_Pro.md

Perplexity Sonar Pro — Apex_Matrix_v_1_0__01-01-2026

See full report: output/analysis/Apex_Matrix_v_1_0__01-01-2026___Perplexity_Sonar_Pro.md

Perplexity Sonar Pro — Montage_Matrix_v_1_0__01-01-2026

See full report: output/analysis/Montage_Matrix_v_1_0__01-01-2026___Perplexity_Sonar_Pro.md

Perplexity Sonar Pro — Portfolio_Guidelines_v1_0__1-1-2026

See full report: output/analysis/Portfolio_Guidelines_v1_0__1-1-2026___Perplexity_Sonar_Pro.md

01-01-2026 v1 0 Claude Opus 4 6 — Spectrum_Matrix

See full report: output/analysis/Spectrum_Matrix__01-01-2026__v1_0__Claude_Opus_4_6.md

01-01-2026 v1 0 GPT-4o — Spectrum_Matrix

See full report: output/analysis/Spectrum_Matrix__01-01-2026__v1_0__GPT-4o.md

01-01-2026 v1 0 Gemini 2 5 Pro — Spectrum_Matrix

See full report: output/analysis/Spectrum_Matrix__01-01-2026__v1_0__Gemini_2_5_Pro.md

01-01-2026 v1 0 Perplexity Sonar Pro — Spectrum_Matrix

See full report: output/analysis/Spectrum_Matrix__01-01-2026__v1_0__Perplexity_Sonar_Pro.md