# Check an existing analysis

> Bring existing data, analysis settings, and results. SciGround recomputes, compares, and returns what can be confirmed and what is missing—without restarting the study.

## Check past results. Keep the gaps visible.

Bring source data, settings and reported results. Recompute within a supported workflow, compare what was reported, and keep confirmed findings separate from missing history and unresolved conditions.

## Start with the evidence that exists.

| Available material | What it contributes | What remains separate |
| --- | --- | --- |
| Original data | Inputs for a supported recomputation. | Unknown preprocessing and study history. |
| Settings or an old script | Settings for you to confirm; scripts can inform a proposal. | Uploaded scripts are not executed or certified. |
| Reported results | A comparison under fixed rounding and tolerance rules. | Agreement does not prove how the original program ran. |
| Historical records | Evidence of design choices, exclusions and prior planning. | Absent history is not recreated by registering now. |

## Keep agreement and validity distinct.

1. **Match.** The reported value agrees under the comparison rules.

2. **Mismatch.** A specific reported value differs under those rules.

3. **Not comparable.** The materials or conditions do not support that comparison.

Value agreement is separate from an established, failed or unresolved verification check. Limited findings are not a full-chain guarantee.

## Bring what remains. Keep what can be checked.

1. **Bring — Upload what exists.** Data, settings, and results, with what is unknown marked as such.
2. **Recompute — Run the supported analysis.** The analysis is run on the original data with recorded versions.
3. **Compare — Match to the reported results.** Each value is classified as a match, a mismatch, or not comparable.
4. **Keep — Save what holds.** Confirmed findings and open gaps are saved separately.

## Useful findings, with their boundaries

- A finding-by-finding report: each reported value a match, mismatch, or not comparable; each check established, condition not met, or unresolved; unsupported items listed separately
- Limited results you can save, share, and recheck
- A list of the information that would close each gap
- A clear separation between what was reported then and what was checked now

## Historical checks and information gaps.

### Bring what you have.

Data alone, or data with the analysis settings and reported results. SciGround runs the supported analysis on the data and compares it with what was reported.

### Match or mismatch, precisely.

Reported values are compared using rounding and tolerance rules fixed in advance. Each result is a match, a mismatch, or not comparable—never a tolerance chosen afterward to fit.

### Partial answers still count.

Even when the full chain cannot be confirmed, the parts that do hold—input matching, numerical agreement, a limited claim—are returned and can be saved and shared.

### Missing context? Start anyway.

Answer what you know and mark the rest unknown. SciGround separates facts, mathematical assumptions, and analysis history, and shows which findings depend on what is missing.

### Reads your old scripts—without running them.

SciGround can propose the settings your old analysis likely used from an existing script. You confirm them, and the comparison runs through SciGround’s standard workflow.

### A clear hand-off.

Limited results come as a package that keeps confirmed findings, unconfirmed items, and the reasons apart—ready for your supervisor or co-authors.

## Limits and conditions

Some things about past work cannot be established after the fact. See what stays open. [Read the limits](https://www.licklider.ai/sciground/existing-analysis/limits/) ([Markdown](https://www.licklider.ai/sciground/existing-analysis/limits.md)).

## Related

- [Compare an original study with a replication](https://www.licklider.ai/sciground/analyses/#replication)
- [Write and report](https://www.licklider.ai/sciground/writing/)
- [Start with an analysis](https://www.licklider.ai/docs/sciground-start/#analysis)