Intelligence Capabilities (New)

When working with imperfect input files (e.g., user uploads, legacy system exports), apply the following intelligent processing strategies.

1. Intelligent Header Detection

Full Guide: header-detection.md

Never assume the first row is the header. Use the following heuristics:

• Semantic Analysis: Look for row content matching known keywords using Levenshtein distance (e.g., "Telefome" ≈ "Telefone").
• Data Pattern Analysis: The header is typically the last text-heavy row before data-heavy rows (dates, emails, CPFs).
• Context: Headers have high density (few empty cells) and are often below metadata rows (titles, dates).

2. Semantic Column Mapping

Full Guide: column-mapping.md

Map source columns to target fields using specific scoring:

• Semantic (40%): Normalization + Synonyms (e.g., "Zap" -> "Phone").
• Pattern (30%): Data validation samples (e.g., column has \d{11} -> "CPF").
• Context (30%): Position and frequency of occurrence.