Testing methodology

Evidence levels

• Automated evidence: unit tests, integration tests, TypeScript checks, lint and the Astro production build.
• Reproducible content evidence: deterministic fixtures encoded by the codec versions used by the project, with exact output bytes recorded in JSON.
• Manual evidence: browser interaction, visual output inspection, cross-browser behavior and network-request review where automation cannot establish the complete result.
• External facts: official platform documentation, with the retrieval date and limitations stated next to the citation.

Compression benchmark procedure

Run `corepack pnpm --filter @batchmint/local-engine benchmark:content` from the repository root. The script regenerates two deterministic synthetic fixtures, encodes them with the project codecs, writes every output, and updates both public and source JSON reports.

• Opaque fixture: 1600×1200 JPEG quality 95 source, compared with JPEG quality 82 and WebP quality 82.
• Transparent fixture: 1200×1200 unoptimised PNG source, compared with OxiPNG level 2 and WebP quality 82 with alpha.
• The report measures encoded file bytes and percentage change from each fixture source. It does not measure speed or perceptual quality.

• View the measured benchmark

  Includes downloadable source and output files.

• View the machine-readable JSON result

  Generated by the same benchmark command.

Codec versions recorded on July 17, 2026

• JPEG: @jsquash/jpeg 1.5.0 using MozJPEG.
• PNG: @jsquash/png 3.1.1 with @jsquash/oxipng 2.3.0.
• WebP: @jsquash/webp 1.5.0 using libwebp.

Network privacy verification

Privacy checks inspect the workspace build and browser network activity while selecting, processing and exporting local test files. BatchMint application requests must not contain selected image bytes, filenames or watermark text, and the workspace must not include the AdSense site script. Requests for HTML, JavaScript, CSS, fonts, service-worker resources and codec assets are expected and are not image uploads.

• Read the privacy model

  Defines what remains local and which network requests can still exist.

Boundaries and interpretation

• Synthetic fixtures make byte results repeatable but do not represent every photograph, illustration or marketplace image.
• One quality value is not visually equivalent across different codecs, so equal numeric quality settings are not a universal quality comparison.
• A target file size is searched within encoder limits and cannot be guaranteed for every source.
• Platform rules and browser capabilities change; official platform requirements and current browsers must be checked again before operational use.