How we test

What we have tested so far

As of the date above, we have published no original recovery benchmarks. The honest state is: protocol drafted, no data collected, no rankings. We will not publish rankings, scores or “best” labels until real, reviewed, reproducible measurement exists.

How future testing will work

• Protocol fixed and timestamped before any data collection; changes create a new version with reasons.
• Fixed synthetic corpora with published manifests: file counts, sizes, hashes, duplicate groups.
• Fixed loss scenarios per filesystem, fresh copies per attempt, multiple independent runs.
• Per-run results with denominators, including failed and crashed runs. Failure is data.
• Environment recorded: OS build, hardware, storage, image hashes, tool versions.
• Datasets downloadable as versioned CSV/JSON with a data dictionary.

What we will not claim

Virtual-image results cannot establish recovery from physical damage, SSD TRIM behaviour, or every real-world case, and we will not present them as if they do. Three runs measure repeatability, not the population. No universal “recovery chance” score exists anywhere honest, so we will never publish one.

Desk audits

Before expensive benchmarks we publish desk audits: documented terms checked against official vendor pages, with observation dates, regions and currency. An audit is not a performance test and is labelled as such.