# Prepare data

> Upload original files, join measurements to sample sheets by ID, convert units with a recorded factor, and keep every sample accounted for—without altering the originals.

## Prepare the data. Keep the originals.

Connect measurements to the right samples, review joins and unit conversions, and approve the analysis input. Original files and the steps that produced each version remain traceable.

## Original files to approved input.

1. **Receive the original.** Upload the source file directly. Confirm the received bytes and keep them unchanged.

2. **Match by sample ID.** Connect measurements and sample information. Surface duplicate, missing and unexpected IDs.

3. **Review the proposed change.** Approve a supported one-to-one join or fixed-factor unit conversion before it is used.

4. **Use the derived version.** The analysis refers to the approved input and its source mapping.

## A join should explain every row.

| Input issue | What SciGround returns |
| --- | --- |
| The same sample ID occurs twice | An explicit duplicate-ID issue to resolve. |
| A measurement has no sample-sheet entry | The unmatched sample remains visible. |
| Measurement units differ | A proposed supported conversion, with its factor and rounding recorded. |

This is a preparation example, not an issued result. Missing values are not silently replaced with zeros; equipment-failure workflows have their own conditions.

## An input with a traceable history

- A quality report listing every issue by sample
- A map from each original row and column to the analysis input
- An approved, versioned derived input
- A record of planned, observed, and missing samples with reasons

## Preparation rules and source handling.

### Originals stay original.

In ChatGPT or Claude, choose files from the upload card in your chat; command-line tools use a companion helper. Files go straight to SciGround and are stored unchanged once their size and digital fingerprint (SHA-256) are confirmed. SciGround accepts originals only as uploaded files, never as values pasted into a request.

### Every sample accounted for.

Each planned sample is matched exactly once to its group and measurement. Duplicate, missing, or extra IDs, unknown groups, unit mismatches, missing values, and non-numeric entries are flagged instead of silently dropped.

### Join measurements to sample sheets.

Combine a measurement table with a sample-information table through an explicit one-to-one ID join. SciGround proposes the join, you approve it, and the derived input becomes a new version—no hidden row loss, duplication, or averaging.

### Unit conversion you can audit.

Convert between units of the same dimension with a fixed factor. The factor, its source, and the rounding are recorded, and the converted values are rechecked before analysis.

### Lost a measurement? Keep the record honest.

When equipment failure costs you some values, the planned samples and the failure reasons stay in the record. The analysis uses only the values actually measured, with the sample count adjusted correctly—missing values are never filled with zeros.

## Limits and conditions

SciGround protects your data rather than reshaping it. See which preparation steps it deliberately leaves out. [Read the limits](https://www.licklider.ai/sciground/data/limits/) ([Markdown](https://www.licklider.ai/sciground/data/limits.md)).

## Related

- [Supported analyses](https://www.licklider.ai/sciground/analyses/)
- [Data and security](https://www.licklider.ai/sciground/security/)
- [Start with data](https://www.licklider.ai/docs/sciground-start/#data)