# Explore, then confirm: limits and conditions

> What exploration results mean, what confirmation requires, and what remains outside the record.

Exploration and confirmation: limits and conditions.

## What exploration results mean

- **Not confirmatory evidence.** Ordinary p-values and intervals computed during exploration do not account for selection and are not issued as confirmatory results.
- **Logs are not a correction.** Recording the exploration does not adjust for selection.
- **A fixed set of operations.** Exploration offers a defined set of summaries, plots, and comparisons, not every exploratory method or data type.

## What confirmation requires

- **Independent, unused samples.** Confirmation needs samples independent of the exploration—separate animals or donors, not a re-split of data already seen.
- **“Unseen” is your assumption.** SciGround withholds confirmation data from the exploration side, but it cannot know whether you saw them elsewhere.
- **One candidate, one comparison.** The first confirmation tests one untransformed outcome with one two-sided Student’s t-test. Repeated confirmations without adjustment are not supported.
- **No changes after the result.** Changing the candidate after seeing confirmation results requires new confirmation data.

## Outside the record

- **Exploration elsewhere.** Analyses done outside SciGround are not captured. The product record is not assumed to be your complete selection history.
- **Methods not included.** Selective inference on the same data, simultaneous guarantees over many candidates, sequential confirmation, and arbitrary code execution are not provided.

Back to [Explore, then confirm](https://www.licklider.ai/sciground/exploration/).