CSV Data Cleaner

Upload a messy CSV and get back a clean one — duplicates removed, whitespace trimmed, null-like values (null, n/a, none, -, …) normalized to empty cells, and empty rows dropped. You also get a STATS.json report showing exactly what changed.

Input

Field	Description
CSV file to clean	Upload a .csv file from your computer
Or: CSV file URL	Public URL of a CSV file
Or: paste CSV text	Paste raw CSV content directly
Delimiter	, ; tab or `
Remove duplicate rows	On by default
Trim whitespace	On by default
Normalize null-like values	On by default
Remove empty rows	On by default

Only one source is needed; priority is uploaded file → URL → pasted text.

Output

• Dataset: one item per cleaned row, with a stable recordId. Duplicate headers receive deterministic __2, __3, … suffixes and cells beyond the source header use extraColumn_1, extraColumn_2, … so no cell is lost.
• Key-value store CLEANED: the cleaned CSV file (text/csv)
• Key-value store STATS: JSON report, e.g.

{
  "totalInputRows": 1000,
  "chargedInputRows": 1000,
  "outputRows": 972,
  "duplicatesRemoved": 21,
  "nullsNormalized": 45,
  "emptyRowsRemoved": 7,
  "columns": 8,
  "sourceColumnNames": ["id", "name", "email", ...],
  "datasetFieldNames": ["id", "name", "email", ...],
  "delimiterUsed": ","
}

Pricing (pay-per-event)

• $0.05 per CSV file cleaned
• $0.0001 per input row processed

Charges are journaled before paid work, and each intent links to a partitioned source snapshot. If the normal post-charge journal update fails, an immutable acceptance receipt retains the accepted count and snapshot link. If a charge is only partly accepted, only its charged prefix is delivered and the raw suffix is preserved in byte-bounded RECOVERY-* records. Recovery carries actual fitted cleaned values as well as raw rows. Terminal dataset or key-value-store failures also create a PRESERVATION-REPORT for record-ID reconciliation; a partial preservation report lists every successful key and record ID rather than claiming all-or-nothing success.

Runs admit at most 8 MiB of CSV input, 100,000 parsed rows, and 500,000 conservatively estimated CSV cells. Larger inputs fail before parser allocation, cleaning, or row billing instead of risking unbounded memory use.

Run via API

curl -H "Authorization: Bearer $APIFY_TOKEN" \
     -H "Content-Type: application/json" \
     -d '{"csvUrl": "https://example.com/messy.csv"}' \
     "https://api.apify.com/v2/acts/<username>~csv-data-cleaner/runs?waitForFinish=60"

Local development

mkdir -p storage/key_value_stores/default
cp test/INPUT.json storage/key_value_stores/default/INPUT.json
APIFY_LOCAL_STORAGE_DIR=./storage python3 -m src.main
# results: storage/datasets/default/ + storage/key_value_stores/default/CLEANED + STATS