Date Formats

Columns detected as date type must contain values that the prediction model can parse. This page explains which formats are accepted, which are auto-normalized, and what to do if you encounter a PRED_DATE_FORMAT error.

Recommended Format

Use ISO 8601 for all date values:

These formats are unambiguous and always accepted without transformation.

Auto-Normalized Formats

The following formats are automatically converted to ISO 8601 before the request reaches the prediction model:

Ambiguous Dates

When both day and month are <= 12 (e.g. 01/02/2025), the system assumes day-first (European convention):

• 01/02/2025 → February 1, 2025 (not January 2)
• 05-03-2025 → March 5, 2025

Empty Values in Date Columns

Empty cells or whitespace-only values in date columns are converted to null before sending to the model. This is accepted by the prediction service.

Formats That Do Not Work

The following will not be recognized as dates and may cause a PRED_DATE_FORMAT error:

Error Response

When date normalization fails and the model cannot parse a date column, you will receive:

{
  "code": "ERR_PRED_DATE_FORMAT",
  "what": "Date format not recognized",
  "why": "Column \"ORDERDATE\" contains date values in an unsupported format (e.g. \"28-11-2025\", \"15-11-2025\")",
  "action": "Use ISO 8601 format (YYYY-MM-DD) for date values",
  "docsLink": "/docs/api/date-formats"
}

HTTP status: 400

Best Practices

1. Always use YYYY-MM-DD when generating data programmatically
2. Avoid locale-dependent formats like MM/DD/YYYY (US) — they conflict with the day-first assumption
3. Include time as THH:MM:SS if your data has timestamps (use T separator, not space)
4. Leave cells empty rather than using placeholder text like N/A or TBD for missing dates