How we test our tools

Published methods and known examples

Calculators identify the formulas or conventions they use. Where practical, automated checks compare their results with separately calculated reference values, rather than only repeating the same implementation. We test boundaries and invalid inputs as well as normal examples. Health and finance calculators remain estimates for general information; testing does not turn them into professional advice.

Data converters are tested with Unicode, empty values, quoted delimiters, multiline fields, duplicate keys, long numbers and resource limits. Round-trip tests help detect accidental value changes. A round trip is useful evidence, but it is not a guarantee that another program uses the same type rules.

Automated checks and real-browser checks

Automated checks cover PHP and JavaScript syntax, public routes, canonical links, page metadata, local assets and selected tool behavior. Some interaction checks run in a simulated document environment. Those checks can catch regressions, but they cannot prove rendering, camera access, graphics performance or file-download behavior on a real device.

Real-browser checks are a separate part of release verification. They include representative interactions, desktop and narrow viewport layouts, local data workers, portrait model processing, and PDF and image exports. A narrow desktop browser window helps find layout issues; it is not a replacement for testing every physical phone, browser or assistive technology.

Checking the exported file

For the PDF tools, sample outputs have been read with a separate PDF parser to verify page order, page dimensions, rotation and retained text or vector content. Multi-file split downloads are also checked as ZIP archives. This tests the generated bytes, not every possible PDF feature.

Image exports have been decoded independently to check format, dimensions, transparency where expected and print-density metadata. Real-browser portrait segmentation has been exercised with a reference portrait. Segmentation remains an estimate: fine hair, glasses, unusual poses, lighting and overlapping people can create imperfect edges.

The worked guides provide small sample inputs and expected results. Those examples are part of the content checks for this release. Please compare an important output in the application that will actually receive it.

Local processing and network exceptions

Most tools read and transform data in your browser. Their pages use a restrictive network policy. The portrait tools load a self-hosted model and runtime before processing; that is an asset download, not a photo upload. The internet speed test is an intentional exception because measuring a connection requires network traffic. Its page explains the external service it contacts.

Our checks do not certify the security of your device, browser extensions or third-party files. PDF processing does not sanitize untrusted documents, and browser-local tools cannot control what happens after you share an export. Read the Privacy Policy for the site’s current data practices.

What testing cannot promise

• Compatibility with every file, operating system, browser version or low-memory device.
• Government acceptance of a passport photo, preservation of a PDF signature or correctness of a production cron service.
• That a connection speed or animation-frame measurement describes your hardware under all conditions.
• That a hosting provider applies every routing, compression or security setting exactly as a local development server does.

This is an explanation of our internal checks, not an independent certification. New releases and hosting changes need their own verification. Report a problem through the Contact page, including the tool, browser, input size, expected result and steps to reproduce it. Remove passwords, personal records and confidential documents from the report.