# Prepare data: limits and conditions

> Which preparation steps are supported, which are deliberately excluded, and what the checks cannot establish.

Data preparation: limits and conditions.

## Steps it does not perform

- **No imputation or automatic exclusion.** Missing values are not filled in, and outliers are not removed automatically.
- **No normalization or batch correction.** General normalization and batch correction are not performed. Bulk RNA-seq uses its own fixed processing profile; see Supported analyses.
- **No automatic aggregation or guessed joins.** Technical replicates are not averaged automatically, and ambiguous IDs are never matched by guesswork.

## Supported inputs

- **Formats.** CSV and JSON. Joining combines one measurement table with one sample-information table through a one-to-one ID key.
- **Numbers.** Values must be plain finite decimals: no exponent notation, at most 40 characters, and either zero or between 10⁻¹² and 10⁶ in magnitude.
- **Unit conversion.** One adopted same-dimension conversion with a fixed factor. Conversions that change dimension are refused.
- **Equipment-failure records.** One type of outcome-independent equipment failure in a two-independent-group study analyzed with Student’s t-test. That the failure was unrelated to the outcome is your declaration, not something SciGround proves; if it is unknown, only limited results are saved. General missing data, deaths, detection limits, and outcome-dependent quality control are out of scope.
- **Accepted data.** Data you have the right to use, that are not identifiable or re-identifiable, and that are not clinical. See Data and security.

## What the checks cannot establish

- **Who sent a file.** SciGround confirms the bytes it received, not who sent them. Keeping uploads to researchers depends on how you assign roles.
- **Measurement truth.** A clean quality report does not prove that the instrument measured correctly or that samples were labeled correctly at the bench.
- **Recovered values.** Analysis with missing values describes the observed data under stated conditions. It does not recover what the missing measurements would have been.
- **History.** Tidying data does not make an analysis pre-specified or a later correction eligible.

Back to [Prepare data](https://www.licklider.ai/sciground/data/).