# Test results: four checks on synthetic company data

Test results on synthetic data · DataBridge by Walnut

<https://walnut.sg/proof/>

Four results measured on synthetic company data with known answers, each with its month, and the same checks run on your files in a pilot.

We test DataBridge on synthetic company data with known answers, so every result can be checked. Each result carries its date, and a pilot runs the same checks on your files.

It starts with a demo on a sample of your own documents and data.

- 579,140 of 579,140 source links verified. Every link from a figure back to its source file resolved, and none was broken. Synthetic finance data, October 2026.
- 368 of 368 figures matched. Answers that combined several systems matched the source database exactly. Synthetic finance data, October 2026.
- 0.9 seconds to add a ninth source. No rebuild was needed; 12 of 12 true matches were linked, and 6 of 6 look-alike names were kept apart. Synthetic test data, August 2026.
- 26 of 26 planted attacks contained. Instructions hidden in emails and files were quoted as content; in the 26 planted tests, none was followed. Synthetic finance data, August 2026. The 26 plantings carried 14 distinct instructions across 7 sources.

## [How the results were measured](https://walnut.sg/proof/#method-head)

Synthetic data means company records we generated, where the right answer to every check is known in advance. No customer data was used.

## [The same checks, on your files](https://walnut.sg/proof/#pilot-head)

During a pilot we run these checks on your own files, so you see the results on your data before you decide what comes next.

## [What a source link shows](https://walnut.sg/proof/#limits-head)

A source link shows where a figure came from and whether the source still says it. It does not show that the source itself is right; that is what a pilot checks against your baseline.

It starts with a demo on a sample of your own documents and data.