How the Data Quality Scorecard Works
AnyTool profiles and scores a dataset entirely in the browser. The CSV is parsed with PapaParse (RFC-4180, so quoted fields are safe), then every column is profiled: each non-empty cell is classified (boolean, integer, number, email, date, or text), giving the column a dominant inferred type, a fill rate, a distinct-value count, and a type-purity percentage. Numbers are classified before dates so an ID column of 5-digit numbers is never mistaken for an Excel-serial date, and date validity reuses the project's UTC date parser. Four dimensions are scored 0–100: Completeness (share of non-empty cells), Validity (cells matching their column's inferred type; free-text columns are not penalised), Consistency (average per-column type purity), and Uniqueness (share of non-duplicate rows). These combine into a weighted overall grade (completeness and validity weighted most). The tool lists concrete findings — low-fill columns, mixed-type columns, constant columns, duplicate rows — and exports a per-column report as CSV. Nothing is uploaded.
- Runs client-side — the dataset is never uploaded
- RFC-4180 parsing + per-column type inference
- Fill rate, distinct count and type purity per column
- Four scores: Completeness, Validity, Consistency, Uniqueness
- Numbers classified before dates (no ID-as-date false positives)
- Concrete findings + downloadable CSV report; no CDN, no backend
