Legal Corpus Quality

What Is a Legal Corpus?

A legal corpus is the database of legal documents that feeds a RAG system. Its quality directly determines the quality of responses.

Sources of a Legal Corpus

1. Legislation: BOE, regional official gazettes, EUR-Lex
2. Case law: CENDOJ, constitutional court databases
3. Administrative doctrine: DGT resolutions, DGRN, circulars
4. Sectoral regulations: regulators (CNMC, Bank of Spain, CNMV)

Construction Process

1. Ingestion: automated download from official sources
2. Parsing: extraction of structured text (articles, grounds, ruling)
3. Chunking: division into coherent semantic fragments
4. Embedding: vectorization of each fragment
5. Indexing: storage in vector database (pgvector)

Quality Problems

• Duplicates: the same regulation appears multiple times with slightly different wording
• Obsolete versions: repealed regulations still indexed
• Encoding: special characters poorly encoded (mojibake)
• Stubs: empty or incomplete records that contaminate results

Quality Metrics

Metric	Description	Target
Coverage	% of key regulations indexed	>95%
Freshness	Average update time	<48h
Deduplication	Ratio of duplicates removed	>99%
Encoding	% of texts without encoding errors	100%

Continuous Maintenance

A legal corpus is not static. It requires:

• Daily monitoring of BOE and official gazettes
• Periodic deduplication
• Re-indexing after corrections
• Quality benchmarks against exam-type questions