# G3D.AI generative 3D evidence method

Version: 1.0  
Status: blank public protocol; no benchmark result is implied.

## Test rule

Compare complete production attempts under the same frozen brief, time limit and acceptance gates. Retain every attempt, including failed generation and export attempts. Score the exported deliverable and its target-engine behavior, not only the hosted preview.

## Required record

1. Define asset role, scale, camera distance, target platform and acceptance gates.
2. Record provider, product, model or feature version, date, account tier, settings and seed when available.
3. Retain prompts or authorized inputs, raw previews, original exports and rejection reasons.
4. Measure retry, repair, retopology, material, rigging, import and review time.
5. Test accepted files in a representative engine scene on the named target.
6. Store input permissions, applicable terms reference, operator, reviewer and final file hash.

## Publication threshold

A public comparison must disclose the test date, systems and versions, sample briefs, attempt count, acceptance gates, rejected runs, cleanup method and limitations. It must not claim a universal winner from one selected output.