# Free Image Converter research data specification

This directory defines the publication format for reproducible research at FreeImageConverter.live. The live benchmark generates data locally; it does not upload or silently publish a visitor's files or results.

## Required publication bundle

Every published study should contain:

1. `README.md` — question, methodology, sample selection, test conditions, conclusions and limitations.
2. `manifest.json` — study ID, semantic version, tested date, authorship, source revision, environment and checksums.
3. `observations.csv` — one measurement per row, with stable column names and units.
4. `summary.csv` — clearly disclosed group-level aggregations derived from observations.
5. `charts/` — accessible SVG or PNG charts whose underlying values are present in the CSV files.
6. `CHANGELOG.md` — corrections, reruns and methodology changes.

## Minimum observation fields

`source_id`, `source_category`, `width_px`, `height_px`, `source_format`, `source_bytes`, `output_format`, `quality_setting`, `output_bytes`, `reduction_percent`, `encode_ms`, `quality_metric`, `quality_value`, `status`, `error_code`.

Do not publish personal filenames, EXIF location data or copyrighted source images without appropriate permission. Public corpus items should have a documented license and stable source identifier.

## Interpretation rules

- Failed or unsupported cases remain in the dataset.
- Missing values remain empty; they are never converted to zero.
- Browser-canvas results are identified as such and are not generalized to all encoders.
- Timing comparisons preserve the device, operating system, browser/version and background-load conditions.
- Quality metrics name the formula and implementation. No single metric is presented as a complete measure of perceived quality.
- Every headline conclusion links back to the rows or aggregation supporting it.

The interactive benchmark exports JSON with schema version `1.0.0` and a CSV containing all raw observations.