{ "run": "run3", "dataset": "yomi-chen/EditableDesignBench", "dataset_revision": "v4.0.0", "dataset_commit": "6b2b9258f50967d6f179325c5bca99ed88ce6e32", "code_repository": "https://github.com/Ch3nSir/EditableDesignBench", "code_commit": "8e087fccdc22798b9b131b62136314c943a2fecf", "models": [ "fibo2-edit-turbo", "joyai-image-edit-plus", "qwen-image21", "qwen-image21-turbo", "flux3-klein", "sensenova-u1.5-8b-mot" ], "text_cases": 100, "asset_cases": 100, "images": 1200, "mllm_scored": false, "pixel_metrics": [ "MSE", "PSNR", "SSIM", "LPIPS" ], "resolution_protocols": { "fibo2-edit-turbo": "native-pad32-v1", "joyai-image-edit-plus": "joyai-official-1024-bucket-resize-center-crop-v1", "qwen-image21": "native-pad32-v1", "qwen-image21-turbo": "native-pad32-v1", "flux3-klein": "native-pad32-v1", "sensenova-u1.5-8b-mot": "native-pad32-v1" }, "model_configuration_notices": { "joyai-image-edit-plus": "Official default bucket output,30steps,CFG4. Original source input without padding; omit width/height. Raw output is preserved. Pixel references/masks are mapped to the official output geometry. These pixel scores use a different resolution protocol from the other five models; report separately." }, "created_at": "2026-11-08T14:42:22.846136+00:00", "original_outputs": "Every raw_outputs PNG is byte-identical to the corresponding run3 model return.", "portable_exports": "Use predictions_index.jsonl, not recorded_reports local-path metadata, to resolve files after download.", "reports_note": "recorded_reports preserve historical reports with workspace root paths replaced by logical labels; they are reference records, not resumable caches.", "evaluation_code_commit": "2dc67b321c948d1976dc7bf5c00910e7cd41060e", "mllm_image_size_policy": "External source and prediction may differ in size; both full RGB images are sent unchanged, no crop/resize/padding/compositing. Pixel protocols are unchanged.", "evaluation_code_note": "Frozen generation/runtime snapshot with the independently sized MLLM input compatibility fix; original source_commit/code_commit identifies generation.", "mllm_smoke_test": { "path": "validation/joyai_gemini16", "model": "joyai-image-edit-plus", "judge_model": "gemini-4-argon", "cases": 16, "complete": true, "full_benchmark": false }, "other_models_mllm_smoke_test": { "path": "validation/other_models_gemini10", "models": [ "fibo2-edit-turbo", "qwen-image21", "qwen-image21-turbo", "flux3-klein", "sensenova-u1.5-8b-mot" ], "cases": 10, "cases_per_model": 2, "complete": true, "full_benchmark": false }, "evaluation_bundle_revision": 2, "release_audit": { "path": "validation/release_audit/summary.json", "offline_tests": 279, "input_pairs_verified": 1200, "real_api_cases": 26, "fresh_environment_install_tested": false }, "evaluation_bundle_note": "Packaging completeness fix adds public config examples and optional dependency manifests; inference/scoring code and model output archives are unchanged." }