data-verification-review

Data Verification (Review Session) §

Date: 2026-03-25 Category: Parsing / Data Processing Parts Completed: 5/5 (review) Language: Python

Problem Summary §

Business validation system that checks CSV records against multiple criteria: blocked terms, business name consistency across descriptors, and length requirements.

Key Bugs Found on Review §

1. Missing continue after short array check §

if len(business_arr) < 6:
    result.append(f"NOT VERIFIED: {business_name}")
    # Was missing continue — fell through to next if, causing double entries
    continue

2. Case sensitivity in blocked terms check §

# Bug: checked uppercase term against original-case descriptor
any(term.upper() in full_statement_descriptor for term in self.blocked_terms)
# Fix: uppercase both sides
any(term.upper() in full_statement_descriptor.upper() for term in self.blocked_terms)

3. Union vs separate check for col4/col5 matching §

• Spec says “50% of col2 words must appear in either col4 or col5”
• Original code checked col4 and col5 independently with OR
• Fix: union the word sets, then check business words against the union

total_words = full_statement_descriptor_words | short_statement_descriptor_words
match_count = sum(word in total_words for word in business_words)

4. Matching direction — semantic clarity §

• sum(word in total_words for word in business_words) — iterate business words, check against descriptors
• Not sum(word in business_words for word in total_words) — same result for unique words but semantically backwards
• In an interview, loop direction should match the spec language

5. Division by zero guard §

• If business_words is empty (name is all ignore words like “LLC INC”), divides by zero
• Fix: if not business_words: return False

Key Learnings §

• Always trace through code with concrete examples before declaring it correct
• Loop direction should match the English spec — shows precise translation of requirements
• Union vs separate check is a common spec interpretation gotcha
• Guard against empty collections before division