> ## Documentation Index
> Fetch the complete documentation index at: https://docs.drpn.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# File formats

> Format requirements for CSV and JSON source files used in Darpan.

These requirements apply to CSV and JSON source files used in Darpan reconciliation runs.

## General requirements

* Use stable headers or field names.
* Keep identifiers as strings when leading zeros or formatting matter.
* Use consistent date formats within a file.
* Avoid mixing summary rows with detail rows.
* Keep one logical record per row when using CSV-like formats.

## CSV guidance

Use UTF-8 encoding and a consistent delimiter. If values can contain commas, quote them consistently.

The header row names the fields. Darpan can build a flat schema directly from it — see [Create a schema](/guides/create-schema).

### Blank cells

A blank CSV cell is read as an **empty value**, not as a missing field. The distinction matters when a run compares the two sides:

| Source state                | How Darpan reports it                     |
| --------------------------- | ----------------------------------------- |
| Column present, cell blank  | The field is present with an empty value. |
| Column absent from the file | The field is absent.                      |

A CSV cell holds text or holds nothing; the format has no way to express a null. Reporting a blank cell as an empty value keeps "this column was blank" distinguishable from "this column was not in the file", which would otherwise produce a difference whose one side simply looked absent.

The compare path and the schema-inference header parser read CSV with the same options, so a file that parses one way in the wizard parses the same way in a run.

## JSON guidance

Use predictable object keys and arrays. Avoid changing object nesting between runs unless the schema is updated with the change.

## Troubleshooting

If upload or parse behavior is unexpected, check encoding, delimiters, header names, blank lines, duplicate columns, and field type assumptions first.

***

To define how source file fields map into Darpan, see [Schema model](/reference/schema-model).