Comprehensive bug analysis: 14 issues identified across API and notebooks

[Copilot]

Conducted complete codebase audit to identify bugs and inconsistencies without applying fixes. Found 14 issues ranging from critical runtime failures to code quality concerns.

Critical Issues (5)

API non-functional due to missing feature transformations:

• log_annual_inc and log_revol_bal created in training (notebook 03, cell 9) but absent in risk_scorer.py
• dti_risk and income_category expected in categorical_cols but never generated from raw dti and annual_inc
• Model will fail on prediction due to feature mismatch

Division by zero:

• loan_to_income = loan_amnt / annual_inc (line 61)
• payment_to_income = (installment * 12) / annual_inc (line 64)
• No validation that annual_inc > 0

Unhandled FileNotFoundError:

• joblib.load() calls (lines 47-49) crash without helpful error if model files missing

High Severity (1)

Train-serve skew:

• Training clips loan_to_income and payment_to_income to 99th percentile
• API applies no clipping
• Extreme values handled inconsistently between training and inference

Medium Severity (3)

• Division by zero in training notebooks (same as API issues)
• Employment length parsing: str.extract(r"(\d+)") treats "< 1 year" as 0.0
• Term field requires exact whitespace match: " 36 months" vs "36 months"

Low Severity (5)

• Global warnings.filterwarnings("ignore") in all notebooks
• Hardcoded relative paths: Path("../data/raw")
• README example incomplete (missing term, installment, sub_grade, etc.)
• No input validation in predict() method
• No required field checking

Deliverable

Complete findings documented in BUG_REPORT.md with:

• Reproduction steps
• Code snippets
• Impact analysis
• Fix recommendations

No code changes applied per requirements.

Original prompt

go through the entire coat base and find me any bugs or inconsistencies, but don't do anything just find them. That's what you're doing. You're not fixing you're not touching the code. You're just finding them for me.

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.