Real Estate Issuer Credit Analysis Pipeline

Multi-phase credit analysis system for real estate investment trusts (REITs) using Claude Code agents.

Why This Exists

Credit analysts shouldn't be human copy machines.

If you've ever spent 6 hours copying numbers from PDFs, fighting with Excel, hunting for footnotes on page 63—only to be too exhausted to think clearly about the actual credit decision—this tool is for you.

This isn't about making credit analysis faster. It's about getting your brain back.

The tool runs automatically. You hit go, get coffee, and come back 30 minutes later to a complete professional credit opinion. Not a draft. Not a summary. Complete.

Because life's too short to copy numbers from PDFs.

Read the full story: Why I Built This

What's New in v1.0.15 🎉

Distribution Cut Prediction Model v2.2 Deployed (October 29, 2025)

✅ Model v2.2 fixes critical underestimation issue - now accurately predicts distribution cut risk for distressed REITs

The Problem with v2.1:

REIT A: Predicted 2.1% (Very Low) when actual risk was 67.1% (High)
Underestimation: 65 percentage points off
Root cause: Feature distribution mismatch (trained on total AFCF, but Phase 3 calculates sustainable AFCF)

Model v2.2 Improvements:

✅ 67.1% High risk prediction for REIT A (was 2.1% Very Low) - aligns with critical distress
✅ Sustainable AFCF methodology - matches Phase 3 calculations (Issue #40)
✅ 28 Phase 3 features (was 54 with market/macro) - more focused feature set
✅ Validated on 3 REITs - improvements of +27 to +65 percentage points
✅ Production deployment - default model path updated, v2.1 archived

Performance (5-fold CV):

F1 Score: 0.870, ROC AUC: 0.930, Accuracy: 87.5%
Top drivers: monthly_burn_rate, acfo_calculated, available_cash

See: Model v2.2 Documentation | Deployment Summary | Analysis

Previous Updates (v1.0.13)

Structural Considerations Content Extraction (October 21, 2025)

✅ Debt Structure, Security & Collateral, Perpetual Securities sections now auto-populated from Phase 4 analysis

Debt Structure: Credit facilities, covenant compliance, debt profile
Security & Collateral: Unencumbered asset pool, LTV ratios, recovery estimates
Perpetual Securities: Automatically detected or marked "Not applicable"

Impact: +15% report completeness, $0 token cost. See CHANGELOG.md.

Overview

This system performs comprehensive credit analysis on real estate issuers (REITs, real estate companies) using a multi-phase pipeline that achieves 99.2% token reduction while generating professional Moody's-style credit opinion reports.

Key Features

5-Phase Sequential Pipeline: Proven architecture (PDF→Markdown→JSON→Metrics→Analysis→Report)
Distribution Cut Prediction: ML model v2.2 predicts 12-month distribution cut risk (High accuracy: F1=0.87, ROC AUC=0.93)
99.2% Token Reduction: File reference patterns reduce Phase 2 from ~140K to ~1K tokens
Dual PDF Conversion Methods: Choose between speed (PyMuPDF4LLM+Camelot, ~30s) or quality (Docling, ~20min)
Market Data Integration: Automated price stress, volatility, and momentum analysis via OpenBB Platform
Macro Environment Tracking: Bank of Canada and Federal Reserve rate monitoring with credit stress scoring
Context-Efficient Phase 2: File references preserve ~199K tokens for extraction logic
Absolute Path Implementation: Reliable execution from any working directory
Organized Output: Issuer-specific folders with separate temp and reports directories
Claude Code Integration: Uses Claude Code agents for intelligent extraction and analysis
Zero-API Dependency: Core pipeline works entirely within Claude Code (OpenBB optional, $0 cost)
Test-Driven Development: Comprehensive test suite for all phases
Production Ready: Generates professional credit opinion reports with 5-factor scorecard analysis
100% Success Rate: Sequential markdown-first approach prevents context window exhaustion

Architecture (v1.0.15 - Sequential Markdown-First with ML Prediction)

Phase 1        Phase 2          Phase 3         Phase 3.5        Phase 4         Phase 5
PDF→MD    →   MD→JSON     →    Calculations  →  Enrichment  →   Agent      →    Report
PyMuPDF/      File refs        Pure Python      ML Model v2.2    Slim Agent      Template
Docling       (~1K tok)        0 tokens         0 tokens         (12K tok)       (0 tok)
30s-20min     Efficient        FFO/AFFO/ACFO    Cut risk:        Credit          Final
                               AFCF metrics     67% High         analysis        report

Phase 3.5 (Optional - Enrichment):

Market risk data (OpenBB Platform): Price stress, volatility, momentum
Macro environment (Bank of Canada, Federal Reserve): Rate cycles, credit stress
Distribution history: 10-year dividend history, cut detection, recovery analysis
Distribution cut prediction (Model v2.2): 12-month cut probability with risk drivers

Two PDF Conversion Methods (Phase 1)

Method 1: PyMuPDF4LLM + Camelot (Default - Fast)

Command: /analyzeREissuer @statements.pdf @mda.pdf "Issuer Name"
Speed: ~30 seconds for 2 PDFs (75 pages total)
Table Format: Enhanced 14-column tables with metadata
Use Case: Interactive analysis, fast iteration, production workloads
Extraction: 113 tables from 75 pages, superior to pure PyMuPDF4LLM
Output: 545KB markdown with rich table formatting

Method 2: Docling (Alternative - Cleaner)

Command: /analyzeREissuer-docling @statements.pdf @mda.pdf "Issuer Name"
Speed: ~20 minutes for 2 PDFs (Docling FAST mode)
Table Format: Compact 4-column tables, cleaner structure
Use Case: Overnight batch processing, cleaner extraction testing
Extraction: Same table coverage, more compact markdown
Output: More concise markdown, easier to parse manually

PDF Conversion Comparison

Method	Phase 1 Time	Table Format	Output Size	Best For
PyMuPDF4LLM + Camelot	~30s	Enhanced (14 cols)	545KB	Interactive, production
Docling FAST	~20min	Compact (4 cols)	Smaller	Batch, testing

Both methods produce identical Phase 2-5 outputs - the choice only affects Phase 1 processing time and markdown structure.

Why Sequential Markdown-First?

Approach	Phase 2 Token Cost	Context Available	Result
Direct PDF Reading	~136K tokens	~64K remaining	❌ Context exhausted
Markdown-First (v1.0.4+)	~1K tokens (file refs)	~199K remaining	✅ Reliable extraction

Key Benefits:

✅ 99.2% token reduction: File references (~1K) vs embedding content (~140K tokens)
✅ Enhanced table extraction: Both methods capture 113 tables from 75 pages
✅ Context preservation: Leaves ~199K tokens for extraction logic and validation
✅ Flexible conversion: Choose speed (PyMuPDF+Camelot) or quality (Docling) per use case
✅ Absolute paths: Reliable execution from any working directory using Path.cwd()
✅ Proven reliability: 100% success rate on production workloads

Distribution Cut Prediction (Model v2.2)

Predicts 12-month distribution cut probability for Canadian REITs using logistic regression trained on 24 observations (11 cuts, 13 controls).

Model Performance:

F1 Score: 0.870 (5-fold CV) - Excellent balance between precision and recall
ROC AUC: 0.930 - Strong discrimination between cut vs. no-cut
Accuracy: 87.5% - High overall prediction accuracy

Key Features (Top 5):

Monthly burn rate - Cash depletion speed (most predictive)
ACFO calculated - Sustainable operating cash flow
Available cash - Immediate liquidity
Total available liquidity - Cash + undrawn facilities
Self-funding ratio - AFCF / Total obligations

Example Predictions:

REIT	Cut Probability	Risk Level	Financial Context
REIT A	67.1%	🔴 High	Cash runway: 1.6 months, Self-funding: -0.61x
REIT B	48.5%	🔴 High	Sustainable AFCF negative
REIT C	29.3%	🟠 Moderate	AFFO payout: 115%

What Changed in v2.2:

✅ Fixed 65-point underestimation in severe distress cases
✅ Uses sustainable AFCF methodology (aligns with Phase 3)
✅ 28 Phase 3 features only (removed market/macro for simplicity)
✅ Deployed to production (Oct 29, 2025)

Risk Classification:

🟢 Very Low: 0-10% probability
🟡 Low: 10-25%
🟠 Moderate: 25-50%
🔴 High: 50-75%
🚨 Very High: 75-100%

See: Model Documentation | Deployment Analysis

Output Structure

Issuer_Reports/
  {Issuer_Name}/
    temp/                                    # Intermediate files (can delete)
      phase1_markdown/*.md                   # Converted PDFs (markdown format)
      phase2_extracted_data.json             # Extracted financial data
      phase2_extraction_prompt.txt           # Extraction prompt for debugging
      phase3_calculated_metrics.json         # Calculated metrics (FFO/AFFO/ACFO/AFCF)
      phase4_enriched_data.json              # Phase 3 + market/macro/prediction (optional)
      phase4_agent_prompt.txt                # Agent prompt for debugging
      phase4_credit_analysis.md              # Qualitative credit assessment
    reports/                                 # Final reports (permanent)
      2025-10-29_031335_Credit_Opinion_{issuer}.md  # Timestamped final report

Note: If Phase 3.5 enrichment runs successfully, phase4_enriched_data.json includes:

Phase 3 calculated metrics
Market risk assessment (price stress, volatility, momentum)
Macro environment (BoC/Fed rates, credit stress score)
Distribution history (10-year dividend data, cut detection)
Distribution cut prediction (Model v2.2: probability, risk level, top drivers)

Installation

Prerequisites

Python 3.10+
Node.js 18+ (for Claude Code)
Git

Setup

1. Install Claude Code:

# Install Claude Code CLI globally
npm install -g @anthropic-ai/claude-code

# Verify installation
claude-code --version

2. Set up the project:

# Clone repository
git clone https://github.com/reggiechan74/issuer-credit-analysis.git
cd issuer-credit-analysis

# Install Python dependencies
pip install -r requirements.txt

# Run tests to verify installation
pytest tests/

3. Start Claude Code:

# Open Claude Code in the project directory
claude-code

# Or use the web interface at:
# https://claude.com/claude-code

Configuration (v1.0.4)

Configure the extraction pipeline via config/extraction_config.yaml.

Recommended Settings (Default in v1.0.4)

phase1_extraction:
  method: "enhanced"  # PyMuPDF4LLM + Camelot (best table quality)

phase2_extraction:
  method: "manual"  # Markdown→JSON with file references
  manual:
    prompt_strategy: "reference"  # ~1K tokens vs ~140K embedded

Why These Defaults?

Setting	Token Cost	Context Available	Reliability
manual + reference	~1K tokens	~199K remaining	✅ 100% success
~~pdf_direct~~ (deprecated)	~136K tokens	~64K remaining	❌ Context exhaustion

Alternative: Agent-Based Extraction

For very large files (>10MB), use the agent-based approach:

phase2_extraction:
  method: "agent"  # Uses financial_data_extractor agent
  # Cost: ~$0.30 per extraction

Quick Start

Using Slash Commands (Recommended)

Method 1: Fast Analysis (Default) - /analyzeREissuer

Best for interactive analysis and production workloads:

# With Claude Code open in this directory:
/analyzeREissuer @statements.pdf @mda.pdf "REIT Name"

# The slash command automatically executes all phases:
# 1. Phase 1: PDF → Markdown (PyMuPDF4LLM + Camelot, 113 tables extracted)
# 2. Phase 2: Markdown → JSON (file references, ~1K tokens)
# 3. Phase 3: Calculate metrics (0 tokens, pure Python)
# 3.5. Enrichment: Market/macro data + distribution cut prediction (Model v2.2)
# 4. Phase 4: Credit analysis (slim agent, ~12K tokens)
# 5. Phase 5: Generate report (0 tokens, templating)

# Total time: ~60 seconds | Total cost: ~$0.30
# Output includes: Distribution cut risk prediction with risk level

Method 2: Cleaner Extraction - /analyzeREissuer-docling

Alternative for batch processing with cleaner markdown output:

# Same usage, different PDF conversion method:
/analyzeREissuer-docling @statements.pdf @mda.pdf "REIT Name"

# Uses Docling for Phase 1 (slower but more compact markdown)
# Phases 2-5 are identical to Method 1

# Total time: ~20 minutes | Total cost: ~$0.30
# Best for: Overnight batch jobs, testing cleaner extraction

When to use Docling:

Overnight/batch processing (time not critical)
Testing if cleaner markdown improves extraction quality
Fallback if PyMuPDF4LLM has issues with specific PDFs

Burn Rate Analysis: /burnrate (New in v1.0.7)

Generate comprehensive cash burn rate and liquidity runway analysis:

# Using issuer name (searches Issuer_Reports/)
/burnrate "REIT Name"

# Using issuer abbreviation
/burnrate REIT

# Using direct path to Phase 2 JSON
/burnrate Issuer_Reports/REIT_Name/temp/phase2_extracted_data.json

# Report includes:
# - Cash burn rate (monthly & annualized)
# - Cash runway (months until depletion)
# - Liquidity risk assessment (CRITICAL/HIGH/MODERATE/LOW)
# - Self-funding ratio and sustainable burn analysis
# - Credit implications and recommended actions

Manual Execution (Individual Phases)

If you prefer to run phases individually:

Phase 1: PDF → Markdown (MUST run first)

Choose your PDF conversion method:

# Method 1: PyMuPDF4LLM + Camelot (Fast - 30 seconds)
python scripts/preprocess_pdfs_enhanced.py \
  --issuer-name "REIT Name" \
  statements.pdf mda.pdf

# Method 2: Docling (Cleaner - 20 minutes)
python scripts/preprocess_pdfs_docling.py \
  --issuer-name "REIT Name" \
  statements.pdf mda.pdf

# Both create: Issuer_Reports/REIT_Name/temp/phase1_markdown/*.md

Phase 2: Markdown → JSON (after Phase 1 completes)

# Extract financial data using file references (~1K tokens)
python scripts/extract_key_metrics_efficient.py \
  --issuer-name "REIT Name" \
  Issuer_Reports/REIT_Name/temp/phase1_markdown/*.md

# Then Claude Code reads the prompt and extracts data

Phase 3: Metric Calculations

python scripts/calculate_credit_metrics.py \
  Issuer_Reports/REIT_Name/temp/phase2_extracted_data.json

Phase 4: Credit Analysis (requires Claude Code agent)

# Within Claude Code:
# Invoke issuer_due_diligence_expert_slim agent with metrics from Phase 3

Phase 5: Final Report Generation

python scripts/generate_final_report.py \
  Issuer_Reports/REIT_Name/temp/phase3_calculated_metrics.json \
  Issuer_Reports/REIT_Name/temp/phase4_credit_analysis.md

Usage Examples

Complete Analysis via Slash Command

# Example 1: Single REIT with multiple PDFs (recommended)
/analyzeREissuer @financial_statements.pdf @mda.pdf "REIT Name"

# Example 2: Quarterly analysis
/analyzeREissuer @Q2_2025_statements.pdf "REIT Name"

Manual Phase 1 Execution (if needed)

# Using PyMuPDF4LLM + Camelot (current default)
python scripts/preprocess_pdfs_enhanced.py \
  --issuer-name "REIT Name" \
  financial_statements.pdf mda.pdf

# Results: 113 tables extracted from 75 pages
# Output: Issuer_Reports/REIT_Name/temp/phase1_markdown/

Cleanup

# Remove temporary files after analysis (keeps reports)
rm -rf Issuer_Reports/*/temp/

# Keep everything organized by issuer
ls Issuer_Reports/REIT_Name/reports/
# Output:
# 2025-10-17_153045_Credit_Opinion_REIT_Name.md
# 2025-10-18_091230_Credit_Opinion_REIT_Name.md

Testing

Run the complete test suite:

# All tests
pytest tests/

# Specific phase
pytest tests/test_phase3_calculations.py -v

# With coverage
pytest tests/ --cov=scripts --cov-report=html

Current Test Status:

Phase 1: ✅ All tests passing
Phase 2: ✅ All tests passing
Phase 3: ✅ All tests passing
Phase 4: ✅ All tests passing
Phase 5: ✅ All tests passing (13/19 active, 6 skipped)

Agent Profiles

issuer_due_diligence_expert_slim (Recommended)

Size: 7.7KB (85% reduction vs full agent) Focus: Qualitative credit assessment from pre-calculated metrics Token Usage: ~12,000 tokens per analysis Version: 1.0.1 (with parallel peer research)

Strengths:

Fast execution (30-60 seconds)
Parallel web research for peer comparisons (new in v1.0.1)
Consistent output format
Comprehensive 5-factor scorecard including 12 detailed sections
Evidence-based assessments with proper citations

issuer_due_diligence_expert (Full Version)

Size: 60KB Use Cases: Complex scenarios requiring deep domain knowledge

Features

Credit Metrics Calculated

Leverage Metrics:

Debt/Gross Assets (%)
Net Debt Ratio (%)
Total Debt, Net Debt, Gross Assets

REIT Metrics:

FFO (Funds From Operations)
AFFO (Adjusted FFO)
ACFO (Adjusted Cash Flow from Operations)
AFCF (Adjusted Free Cash Flow)
FFO/AFFO per unit
FFO/AFFO payout ratios
Distribution coverage

Coverage Ratios:

NOI/Interest Coverage
EBITDA/Interest Coverage
Debt Service Coverage
AFCF Debt Service Coverage
AFCF Self-Funding Ratio

Liquidity & Burn Rate Analysis (v1.0.7):

Cash burn rate (monthly & annualized)
Cash runway (months until depletion)
Liquidity risk assessment (CRITICAL/HIGH/MODERATE/LOW)
Sustainable burn rate analysis
Self-funding ratio (AFCF / Net Financing Needs)
Key Insight: Identifies REITs with positive AFCF that still burn cash

Credit Analysis Sections

The system generates comprehensive reports with 15+ sections:

Executive Summary - Rating and credit story
Credit Strengths - Quantified positive factors
Credit Challenges - Risk factors with mitigants
Rating Outlook - Stable/Positive/Negative with timeframe
Upgrade Factors - Specific thresholds for improvement
Downgrade Factors - Quantified triggers
5-Factor Scorecard - Detailed rating methodology
Key Observations - Portfolio quality, unusual metrics
Peer Comparison - Parallel web research of 3-4 comparable REITs with citations (v1.0.1)
Scenario Analysis - Base/Upside/Downside/Stress cases with pro forma metrics
Structural Considerations - NEW in v1.0.13 - Auto-extracted from Phase 4 analysis:
- Debt Structure: Credit facilities, covenant compliance, debt profile
- Security & Collateral: Unencumbered assets, LTV ratios, recovery estimates
- Perpetual Securities: Hybrid capital instruments or "Not applicable"
ESG Considerations - Environmental, Social, Governance factors with CIS scoring
Company Background - Corporate structure, history, portfolio composition
Business Strategy - Strategic priorities and capital allocation
Detailed Financial Analysis - FFO/AFFO/ACFO/AFCF reconciliations and bridge analysis

Safety Features

No Hardcoded Data: All calculations use explicit inputs
Loud Failures: Invalid data triggers clear error messages
Validation Checks: Balance sheet balancing, NOI margins, occupancy ranges
Evidence Quality: Strong/moderate/limited evidence labels
Professional Caveats: Clear disclaimers on limitations

Token Usage & Cost

Phase	Tokens	Cost (approx)	Time	Details
Phase 1	0	$0.00	10-15s	PyMuPDF4LLM + Camelot (113 tables from 75 pages)
Phase 2	~1,000	$0.00	5-10s	File references (not embedded content)
Phase 3	0	$0.00	<1s	Pure Python calculations
Phase 4	~12,000	~$0.30	30-60s	Slim agent credit analysis
Phase 5	0	$0.00	<1s	Template-based report generation
Total	~13,000	~$0.30	~60s	99.2% token reduction

Performance Comparison

Approach	Total Tokens	Cost	Success Rate	Notes
v1.0.4 (Current)	~13,000	$0.30	100%	File reference patterns
~~v1.1.0 PDF Direct~~	~148,000	$3.70	0%	Context exhaustion (reverted)
Original Single-Pass	~121,500	$3.04	~30%	Frequent context errors

Key Achievement: 99.2% token reduction in Phase 2 alone (~140K → ~1K tokens)

Project Origin

This project was extracted from the geniusstrategies repository, which explores cognitive strategy coaching through AI agents embodying historical geniuses' thinking patterns.

The credit analysis pipeline was developed as a domain expert implementation demonstrating:

Multi-phase pipeline architecture
Claude Code agent integration
Test-driven development practices
Production-ready financial analysis

Recent Improvements

v1.0.7 Updates (Latest)

Issue #7 - Cash Burn Rate and Liquidity Runway Analysis (Implemented)

✅ 4 new calculation functions: burn rate, cash runway, liquidity risk, sustainable burn
✅ New /burnrate slash command for comprehensive liquidity analysis
✅ 36 tests passing (25 unit + 11 integration)
✅ Critical Discovery: REITs can have positive AFCF but still burn cash when financing needs exceed free cash flow
✅ Production-ready: Tested with Dream Industrial REIT and Artis REIT

Issue #6 - AFCF (Adjusted Free Cash Flow) Calculations (Implemented)

✅ New metric: AFCF = ACFO + Net Cash Flow from Investing
✅ More conservative than ACFO - includes ALL investment activities
✅ AFCF coverage ratios: debt service, distributions, self-funding
✅ 17 tests passing with comprehensive validation

Issue #5 - ACFO Implementation (Implemented)

✅ Automated ACFO calculation using REALPAC methodology
✅ 17 adjustments to IFRS CFO for normalized operating cash flow
✅ Prevents double-counting with AFCF calculations

v1.0.4 Updates

Issue #1 - PDF Markdown Conversion (Resolved)

✅ Implemented PyMuPDF4LLM + Camelot hybrid approach
✅ Extracts 113 tables from 75-page documents with high fidelity
✅ Superior table structure preservation vs MarkItDown

Issue #2 - Context Length Optimization (Research Complete)

✅ Comprehensive semantic chunking research completed (docs/SEMANTIC_CHUNKING_RESEARCH.md)
✅ Current file reference architecture already achieves 99.2% token reduction
📋 Optional semantic chunking planned for v1.1.0 (documents >256KB)
🔮 RAG-based approach for v2.0.0 (documents >1MB, future consideration)

Code Quality Improvements

✅ Absolute path implementation using Path.cwd() for reliable execution
✅ Fixed property_count field mapping bug (Phase 3 calculations)
✅ Enhanced error handling and validation across all phases
✅ 100% test passing rate (Phase 1-5)

Contributing

Contributions welcome! Areas of interest:

Additional asset classes (corporate bonds, structured finance)
Enhanced portfolio quality metrics
Integration with financial data APIs
Visualization dashboards
Semantic chunking implementation (v1.1.0 roadmap available in Issue #2)
Enhanced peer comparison analytics (parallel research implemented in v1.0.1)

License

Licensed under the Apache License, Version 2.0 (the "License").

This project is licensed under Apache 2.0, which allows:

✅ Commercial use
✅ Modification and distribution
✅ Patent use (explicit patent grant included)
✅ Private use
✅ Integration into proprietary systems

Requirements:

Attribution to the original author (Reggie Chan)
Include copy of the Apache 2.0 license
State any significant changes made

For the complete license text, see the LICENSE file or visit http://www.apache.org/licenses/LICENSE-2.0

Disclaimer

IMPORTANT NOTICE: PLEASE READ CAREFULLY BEFORE USING THIS SOFTWARE

Purpose and Scope

This software tool (the "Tool") is designed solely for informational and analytical purposes to assist qualified credit professionals in evaluating real estate investment trusts ("REITs") and related issuers. The Tool generates automated credit assessments based on financial data extraction and machine learning models.

Not Investment Advice or Credit Rating

THIS TOOL DOES NOT PROVIDE INVESTMENT ADVICE, CREDIT RATINGS, OR RECOMMENDATIONS. The Tool is not:

Investment advice, investment recommendations, or investment research
A credit rating or credit opinion as defined by securities regulators or rating agencies
A substitute for independent professional credit analysis or due diligence
A guarantee, warranty, or assurance regarding credit quality, investment returns, or future performance
Approved, endorsed, or validated by any credit rating agency, securities regulator, or financial authority
Intended for retail investors or non-professional users

No Fiduciary Relationship

Use of this Tool does not create any fiduciary, advisory, or agency relationship between the user and the Tool's creators, maintainers, or contributors. Users retain full responsibility for their own investment and credit decisions.

Limitations and Risks

Machine Learning Model Limitations:

The distribution cut prediction model (v2.2) is trained on limited historical data (24 observations) and may not accurately predict future events
Model performance metrics represent historical backtesting and do not guarantee future accuracy
The model may produce false positives (incorrectly predicting distribution cuts) or false negatives (failing to predict actual cuts)
Model predictions should be validated against independent analysis and not relied upon exclusively

Data and Extraction Risks:

The Tool relies on automated PDF extraction which may introduce errors, omissions, or misinterpretations
Financial data accuracy depends on the quality and completeness of source documents
The Tool cannot verify the accuracy of issuer-reported financial statements
Users must independently verify all extracted data against original source documents

Third-Party Data:

Market data, macroeconomic data, and dividend history are sourced from third-party providers (OpenBB Platform, TMX, YFinance, Bank of Canada, Federal Reserve) and may contain errors or delays
The Tool's creators make no representations regarding the accuracy, completeness, or timeliness of third-party data

Analytical Limitations:

The Tool applies standardized REALPAC methodologies which may not be appropriate for all issuers or circumstances
Automated analysis cannot replace human judgment regarding qualitative factors, management quality, or strategic considerations
The Tool does not consider all factors relevant to credit analysis, including but not limited to: litigation risks, environmental liabilities, regulatory changes, or market conditions

Professional Review Required

ALL OUTPUTS FROM THIS TOOL MUST BE REVIEWED, VALIDATED, AND APPROVED BY QUALIFIED CREDIT PROFESSIONALS before being used for any credit decision, investment decision, or publication. Users should:

Independently verify all extracted financial data
Review model predictions against independent credit analysis
Consider qualitative factors not captured by the Tool
Obtain appropriate credit committee or investment committee approval
Ensure compliance with applicable investment policies, credit policies, and regulatory requirements

Regulatory Compliance

Users are solely responsible for ensuring their use of this Tool complies with all applicable laws, regulations, and professional standards, including but not limited to:

Securities laws and regulations
Investment adviser regulations (if applicable)
FINRA, SEC, IIROC, or other regulatory requirements
Internal compliance policies and procedures
Professional conduct standards and fiduciary duties

No Warranties

THIS TOOL IS PROVIDED "AS IS" WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED. The creators disclaim all warranties including, but not limited to, warranties of accuracy, completeness, merchantability, fitness for a particular purpose, and non-infringement. The creators do not warrant that the Tool will be error-free, uninterrupted, or free from defects.

Limitation of Liability

TO THE MAXIMUM EXTENT PERMITTED BY APPLICABLE LAW, THE CREATORS, MAINTAINERS, AND CONTRIBUTORS SHALL NOT BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, CONSEQUENTIAL, SPECIAL, OR PUNITIVE DAMAGES arising out of or related to the use of this Tool, including but not limited to: investment losses, trading losses, lost profits, data inaccuracies, or business interruption, even if advised of the possibility of such damages.

Intellectual Property and Proper Use

This Tool is licensed under GNU General Public License v3.0. Users must comply with all license terms. Outputs generated by this Tool:

Should not be represented as official credit ratings or third-party research
Must include appropriate disclaimers when shared with third parties
Must attribute automated/machine-generated analysis appropriately
Should not be used to circumvent regulatory requirements for independent credit analysis

Acknowledgment

BY USING THIS TOOL, YOU ACKNOWLEDGE THAT YOU HAVE READ, UNDERSTOOD, AND AGREE TO BE BOUND BY THIS DISCLAIMER. If you do not agree to these terms, do not use this Tool.

Questions or Concerns

For questions regarding proper use of this Tool, consult with your legal, compliance, or risk management teams before proceeding.

Last Updated: January 2025 Version: 1.0.15

Support

Issues: GitHub Issues
Discussions: GitHub Discussions
Original Project: geniusstrategies

Built with Claude Code | Documentation | Examples

Name		Name	Last commit message	Last commit date
Latest commit History 195 Commits
.claude		.claude
Issuer_Reports		Issuer_Reports
analysis		analysis
config		config
data		data
docs		docs
models		models
predictions		predictions
research		research
scripts		scripts
templates		templates
tests		tests
.gitignore		.gitignore
CHANGELOG.md		CHANGELOG.md
CLAUDE.md		CLAUDE.md
LICENSE		LICENSE
PIPELINE_ARCHITECTURE.md		PIPELINE_ARCHITECTURE.md
README.md		README.md
requirements.txt		requirements.txt
why_I_built_this.md		why_I_built_this.md

License

reggiechan74/issuer-credit-analysis

Folders and files

Latest commit

History

Repository files navigation

Real Estate Issuer Credit Analysis Pipeline

Why This Exists

What's New in v1.0.15 🎉

Distribution Cut Prediction Model v2.2 Deployed (October 29, 2025)

Previous Updates (v1.0.13)

Overview

Key Features

Architecture (v1.0.15 - Sequential Markdown-First with ML Prediction)

Two PDF Conversion Methods (Phase 1)

PDF Conversion Comparison

Why Sequential Markdown-First?

Distribution Cut Prediction (Model v2.2)

Output Structure

Installation

Prerequisites

Setup

Configuration (v1.0.4)

Recommended Settings (Default in v1.0.4)

Why These Defaults?

Alternative: Agent-Based Extraction

Quick Start

Using Slash Commands (Recommended)

Manual Execution (Individual Phases)

Usage Examples

Complete Analysis via Slash Command

Manual Phase 1 Execution (if needed)

Cleanup

Testing

Agent Profiles

issuer_due_diligence_expert_slim (Recommended)

issuer_due_diligence_expert (Full Version)

Features

Credit Metrics Calculated

Credit Analysis Sections

Safety Features

Token Usage & Cost

Performance Comparison

Project Origin

Recent Improvements

v1.0.7 Updates (Latest)

v1.0.4 Updates

Contributing

License

Disclaimer

Purpose and Scope

Not Investment Advice or Credit Rating

No Fiduciary Relationship

Limitations and Risks

Professional Review Required

Regulatory Compliance

No Warranties

Limitation of Liability

Intellectual Property and Proper Use

Acknowledgment

Questions or Concerns

Support

About

Topics

Resources

License

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Contributors 2

Uh oh!

Languages

Packages