Audit a Vendor CSV for Gaps, Banned Terms and Name Mismatches

A payments platform onboards new vendors from a CSV export: one header row naming the columns, then one row per vendor. Write a validator that checks every row against a set of rules supplied as configuration, and reports which rules each row fails. Nothing about field names, required fields, length limits, banned words or how "enough" name overlap is defined should be hardcoded — all of it is configuration, and you should state the shape you chose.

vendor_id,vendor_name,legal_name,notes
1,Acme Robotics,Acme Robotics LLC,Primary supplier
2,,Bright Labs LLC,Missing display name
3,  ACME   Robotics ,acme robotics llc,Spacing and case only
4,Zed,Zed Logistics Inc,Display name too short

Function Signature

def validate_vendor_records(csv_text: str, config: dict) -> list[dict]:

Rules

• The first line is a header; every following line is one vendor record. A field value may itself contain a comma if quoted, so parse with a real CSV reader rather than splitting on commas.
• The validator only reports problems — it never edits, reorders, drops, or deduplicates records.
• config supplies: required_fields, length_limits (per-field (min, max) after trimming), forbidden_words (matched whole-word, case-insensitive, on every field), overlap_fields (a pair of field names to compare), ignored_words (stripped out before comparing, e.g. llc, inc), and min_overlap_ratio.
• A field counts as missing if it is empty or only whitespace after trimming.
• For the overlap check: lowercase both overlap_fields values, strip punctuation, split into words, drop ignored_words, then compute the Jaccard ratio |intersection| / |union| of the two word sets. If both fields are non-empty and the ratio is below min_overlap_ratio, flag "name_mismatch".
• Every row appears in the output, in input order, even with an empty issue list.

Part 1 — Per-field checks

For each record, check every required_fields entry for presence and every length_limits entry for range, and scan every field for forbidden_words.

Part 2 — Cross-field overlap

Using overlap_fields and ignored_words, compute the Jaccard ratio described above and flag records whose two name fields don't look related enough.

Example

config = {
    "required_fields": ["vendor_id", "vendor_name", "legal_name"],
    "length_limits": {"vendor_name": (4, 80), "legal_name": (4, 120)},
    "forbidden_words": ["test", "dummy"],
    "overlap_fields": ("vendor_name", "legal_name"),
    "ignored_words": {"inc", "llc", "ltd", "co", "corp", "company", "the"},
    "min_overlap_ratio": 0.5,
}
# Row 2 (missing vendor_name) -> {"row": 2, "issues": ["missing:vendor_name"]}
# Row 4 (vendor_name "Zed", length 3 < min 4) -> {"row": 4, "issues": ["length:vendor_name"]}

Constraints

• Up to 10^4 rows and 50 columns.
• Field values are printable text; forbidden words and ignored words are case-insensitive, whole-word matches.