data-clean

Data Clean

Clean the given tabular data file by fixing common data quality issues.

Cowork note: If relative paths don't resolve, call mcp__qsv__qsv_get_working_dir and mcp__qsv__qsv_set_working_dir to sync the working directory.

Steps

1. Index: Run mcp__qsv__qsv_index on the file for fast random access in subsequent steps.

2. Assess current state: Run mcp__qsv__qsv_sniff and mcp__qsv__qsv_count to understand the file format and size.

3. Profile for cleaning decisions: Run mcp__qsv__qsv_stats with cardinality: true, stats_jsonl: true. Read .stats.csv to decide which cleaning steps are needed:

   Stats Column	What It Reveals	Cleaning Action
   nullcount, sparsity	Missing values per column	If sparsity > 0.5, decide: impute, drop column, or flag
   cardinality vs row count	Duplicate rows exist if any key column has cardinality < row count	Run dedup
   min_length, max_length	String length variation	Large gap suggests ragged data or embedded whitespace
   sort_order	Whether data is pre-sorted	Use dedup --sorted for streaming mode if sorted
   mode, mode_count	Dominant values	If mode_count > 80% of rows, investigate data entry defaults
   type	Inferred types	String columns that should be numeric indicate format issues

4. Check headers: Run mcp__qsv__qsv_headers to inspect column names. If names contain spaces, special characters, or are duplicated, plan to use safenames.

5. Build cleaning steps: Apply these operations in order (skip any that aren't needed based on assessment):

   a. safenames - Normalize column names to safe, ASCII-only identifiers (removes spaces, special chars, ensures uniqueness)

   b. fixlengths - Ensure all rows have the same number of fields (pads short rows, truncates long rows)

   c. sqlp - Remove leading/trailing whitespace from columns using TRIM(). Example: SELECT TRIM(col1) AS col1, TRIM(col2) AS col2 FROM _t_1.

   d. dedup - Remove exact duplicate rows. Loads all data into memory and sorts internally. Use --sorted if input is already sorted to enable streaming mode with constant memory.

   e. validate - If a JSON Schema is available, validate against it and report violations.

6. Verify results: Run mcp__qsv__qsv_count on the output to confirm row count. Run mcp__qsv__qsv_stats with cardinality: true to verify improvements.

7. Report changes: Summarize what was cleaned:

   • Headers renamed (before -> after)
   • Rows with wrong field count (fixed by fixlengths)
   • Duplicate rows removed
   • Whitespace trimmed

Cleaning Steps

Call each tool sequentially, passing the output of one step as input to the next:

1. mcp__qsv__qsv_command with command: "safenames", input_file: "<file>", output_file: "step1.csv"
2. mcp__qsv__qsv_command with command: "fixlengths", input_file: "step1.csv", output_file: "step2.csv"
3. mcp__qsv__qsv_sqlp with input_file: "step2.csv", sql: "SELECT TRIM(col1) AS col1, TRIM(col2) AS col2, ... FROM _t_1", output_file: "step3.csv" (list all columns with TRIM)
4. mcp__qsv__qsv_command with command: "dedup", input_file: "step3.csv", output_file: "<output>"

Notes

• Always preserve the original file - write output to a new file
• For large files (> 100MB), dedup loads entire file into memory to sort and deduplicate; consider using sqlp with SELECT DISTINCT instead
• safenames uses --mode conditional by default (only renames if needed)
• If the user specifies particular columns to clean, use column selection syntax instead of cleaning all columns
• dedup loads all data into memory and sorts internally; if input is already sorted, use --sorted for streaming mode
• Use mcp__qsv__qsv_search_tools to find additional cleaning tools if needed (e.g., replace for regex substitution)