Data Quality & Contributing

This data is extracted automatically from thousands of PDFs and spreadsheets published by hundreds of individual public bodies, each with their own formatting. Extraction errors are expected. Rather than hide them, this pipeline publishes exactly what it could and could not process, so problems are inspectable instead of silent.

The two levels of quality signal

Dataset-level: did a body produce data at all?

pipeline-data.json records, per public body, whether each processing stage succeeded or failed: finding its website, its FOI page, its FOI contact email, its disclosures page, its disclosure files, and its FOI requests. This tells you coverage — which bodies produced data and where extraction stopped for the ones that didn't.

disclosure-files.json lists the source documents the pipeline discovered and attempted to process, so you can see exactly what went in.

Per-record: how confident is this one row?

Individual records in the FOI Disclosures dataset carry known_issues and missing_columns fields. These are per-record extraction-confidence signals, not a global quality score — a flagged record is not necessarily wrong, only less certain in the specific way flagged. See the FOI Disclosures README's Known Limitations section for the authoritative explanation.

These two levels answer different questions and should never be combined into one number: knowing a body's disclosure files failed to process tells you nothing about the confidence of any individual record, and vice versa.

Report a problem or contribute an improvement

Spotted a wrong record, a missing body, or anything else that looks off? Report it or suggest a fix at www.publicinformation.ie — no technical knowledge or git required.

Want to change how the pipeline itself works? The source code is at the Codeberg repository.