additional_features
["Structured taxonomy by rule_type/subtype (part-whole, conditional presence, range/order, hierarchy, ratio, contradiction patterns, repeated consistency, single-value, uniqueness, existence)", "Conforms to the CSV/JSON schema (fields: rule_id, rule_name, rule_type, subtype, scope, condition, expression, variables, group_by, human_desc_ko, severity, enabled, comment)"]
example_commands
["Given my survey JSON and qcmeta.json, propose LGMeta rule drafts in the lgmeta_draft.csv structure.", "Convert required_if/skip_if into conditional_presence rules, and mark uncertain ones with TODO comments.", "Find variables labeled with 'male/female/total' patterns and propose part_whole candidate rules.", "With household_id and person_id, suggest existence and repeated_consistency rules."]
gpt_id
g-692c1c0e69388191af46f23674c5a103-lgmeta
ideal_use_cases
["Convert survey routing logic (required_if/skip_if) into multi-variable validation rule drafts", "Quickly derive sum/parts checks (e.g., male+female=total) and percentage-sum (100%) candidates", "Draft anti-duplication rules for identifiers (phone/ID/business numbers)", "Suggest existence and cross-check rules for household-person repeated data structures", "Generate candidate rules for ordering/ranges (dates/ages/amounts) and hierarchy constraints (e.g., profit layers)"]
limitations
["All auto-generated rules are disabled by default (enabled=false) and must be reviewed by a human", "Without clear semantics/codebooks (e.g., missing/special codes), suggestions may be over/under-inclusive", "If key context (questionnaire/codebook/sampling design) is missing, some rules may remain TODO-only"]
target_users
["Data quality (QC) practitioners", "Survey/panel designers and analysts", "ETL and data pipeline engineers"]