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

# Document intake best practices

This guide explains how to collect and submit documents in a way that preserves **forensic value** and improves detection reliability.

## The main rule: submit the original file bytes

Submit the **raw original file** as received from the end user (or downstream customer). Avoid intake paths that **re-save, re-encode, or reconstruct** the document before analysis.

<Warning>
  Do not submit files that were pre-processed by third-party tools unless you have no alternative.
</Warning>

### Why this matters

Many systems silently modify documents (resizing, JPEG re-compression, stripping metadata, PDF reconstruction). These changes can reduce or eliminate forensic signals.

Common culprits:

* email gateways
* document management systems
* file sharing tools
* messaging apps (WhatsApp/Slack)
* “optimize” or “compress” functions

### Practical guidance

| Do                                                   | Don’t                                              |
| ---------------------------------------------------- | -------------------------------------------------- |
| Collect the file via direct upload in your user flow | Route via messaging apps (often re-encodes images) |
| Store and forward the original bytes unchanged       | Re-save documents in editors “to standardize” them |
| Upload binary bytes to the presigned URL             | Base64-encode and re-wrap the file payload         |

***

## Format considerations

### PDFs (best for automation and accuracy)

Digitally issued PDFs (e.g., bank statements, utility bills, forms) generally enable stronger detection and automation and are less sensitive to quality loss from re-encoding.

### Images (more sensitive to quality)

Images require more care:

* authenticity is harder to determine
* pre-digital tampering (before scanning) may not be detectable without content analysis
* lower image quality increases false positives

***

## Redaction guidance

Redaction and annotation creates a **new modified document**, which can reduce forensic value.

Key impacts:

* may change document structure/metadata and obscure evidence
* invalidates digital signatures/hashes used for authenticity verification
* some tools reconstruct PDFs in ways that increase false positives

<Tip>
  If possible, analyze originals first and apply redaction only after analysis. If you must redact before analysis, keep the original stored separately for audit/debug (if allowed by your policy).
</Tip>

***

## Supported file formats

See [Supported formats](/supported-formats).

***

## Go-live checklist

Before you go live, validate:

* You can capture and store **original bytes** end-to-end.
* Your pipeline does not re-encode, resize, reconstruct, or “optimize” documents.
* You have a strategy for low-quality image handling (re-request vs manual review).
* Your logs contain only non-PII identifiers (e.g., `submission_id`, safe `query_id`, timestamps, stage/cell).