Methods & limits
What our checks actually establish
Transparent formulas, small reproducible fixtures and a clear boundary between software tests and real-world recovery.
Image-file comparison experiment · October 15, 2026
The expanded copy-verification guide provides five original SVG files: a reference, an exact copy, a metadata-only change, a color change and an intentionally truncated file. The manifest lists byte lengths, SHA-256 references and five expected pair outcomes, including two identical invalid inputs. Run the cases in the checker or download them for the file pickers. SVG drawing instructions are inspectable; these are synthetic software fixtures, not camera photos, EXIF examples or cloud-transfer tests.
Reference digests are generated separately with Node’s crypto implementation. Browser comparisons use Web Crypto and direct byte comparison. Both implementation checks and their limits are visible; this is not independent third-party certification.
Three methods, three boundaries
| Tool | Method | Not established |
|---|---|---|
| Backup self-check | List every unchecked evidence item. All 64 combinations are covered by automated tests. | Actual account state, honesty of answers, complete recovery. |
| File copy checker | Compare selected bytes; calculate SHA-256 separately with browser Web Crypto. | Media health, authorship, absence of malware or completeness of an archive. |
| Space planner | (Current + growth × months + retained versions) / (1 − reserve fraction). | Future behavior, compression ratio, available space read from a device. |
Reproduce the file demonstration
Open the file copy checker and choose “Try known demo.” It uses UTF-8 text bytes abc, abc and abd. Expected outcomes: identical first pair; non-identical second pair despite the same size. The expected fingerprint for abc is published in the comparison guide.
The test suite also covers empty inputs, same-size different bytes, differing sizes, unsupported sizes, missing numeric inputs, non-finite values, reserve boundaries and unit conversion. These are synthetic software tests, not field trials on readers’ photo libraries.
Reproduce the storage example
Use 120 GB current, 4 GB/month, 12 months, 30 GB extra versions, 20% reserve and 200 GB free. Expected budget: 247.5 GB; shortfall: 47.5 GB; doubled-growth budget: 307.5 GB. Changing an input hides the old result until you calculate again.
Inspect or rerun the code
The site is static. You can inspect the tool implementation and the automated test suite in the public repository. With Node.js installed, run node scripts/test-focused-site.cjs from a copy of the repository. Nothing in that test sends personal files to KYD.
Evidence labels used on this site
- Provider fact: documented service behavior, linked to an official source with a check date.
- Hypothetical example: invented inputs to explain a method, never presented as a customer or performance result.
- Software fixture: a reproducible check of the implementation, with a known expected output.
- Reader observation: something you personally verified on your own copy. KYD does not collect it.
We do not turn a passing software test into a claim that iCloud, Google Photos, a drive or an entire recovery strategy is safe. Report a mismatch between these descriptions and the tool.