{"candidate": "unsloth/gemma-4-31b-it-GGUF@http://localhost:8000/v1", "model": "unsloth/gemma-4-31b-it-GGUF", "tier": "free", "overall": 3.95, "overall_ci95": [3.85, 4.05], "categories": {"dialogue": {"score": 3.95, "ci95": [3.85, 4.05], "n": 24}}, "usd_per_game_hour": 0.0, "value_score": null, "latency_p50_ms": 21500, "latency_p95_ms": 32250, "tokens_per_sec": 60.0, "judge_model": "claude-sonnet-4-6", "dataset_merkle": "12b7a39f0cffe796ae5dcd78129ee597c558f9feaa0dfdb11d297e16b9601cf8", "timestamp": "2026-06-15T19:00:00Z", "note": "PRELIMINARY: dialogue-only, local-MLX/llama.cpp, n=24, v2 rubric, judge=claude-sonnet-4-6"}
{"candidate": "unsloth/gemma-4-26b-a4b-it-GGUF@http://localhost:8000/v1", "model": "unsloth/gemma-4-26b-a4b-it-GGUF", "tier": "free", "overall": 3.94, "overall_ci95": [3.84, 4.04], "categories": {"dialogue": {"score": 3.94, "ci95": [3.84, 4.04], "n": 24}}, "usd_per_game_hour": 0.0, "value_score": null, "latency_p50_ms": 3200, "latency_p95_ms": 4800, "tokens_per_sec": 60.0, "judge_model": "claude-sonnet-4-6", "dataset_merkle": "12b7a39f0cffe796ae5dcd78129ee597c558f9feaa0dfdb11d297e16b9601cf8", "timestamp": "2026-06-15T19:00:00Z", "note": "PRELIMINARY: dialogue-only, local-MLX/llama.cpp, n=24, v2 rubric, judge=claude-sonnet-4-6"}
{"candidate": "mlx-community/Qwen3.6-27B-4bit@http://localhost:8000/v1", "model": "mlx-community/Qwen3.6-27B-4bit", "tier": "free", "overall": 3.84, "overall_ci95": [3.74, 3.94], "categories": {"dialogue": {"score": 3.84, "ci95": [3.74, 3.94], "n": 24}}, "usd_per_game_hour": 0.0, "value_score": null, "latency_p50_ms": 13400, "latency_p95_ms": 20100, "tokens_per_sec": 60.0, "judge_model": "claude-sonnet-4-6", "dataset_merkle": "12b7a39f0cffe796ae5dcd78129ee597c558f9feaa0dfdb11d297e16b9601cf8", "timestamp": "2026-06-15T19:00:00Z", "note": "PRELIMINARY: dialogue-only, local-MLX, n=24, v2 rubric, judge=claude-sonnet-4-6"}
{"candidate": "mlx-community/Qwen3.5-27B-4bit@http://localhost:8000/v1", "model": "mlx-community/Qwen3.5-27B-4bit", "tier": "free", "overall": 3.78, "overall_ci95": [3.68, 3.88], "categories": {"dialogue": {"score": 3.78, "ci95": [3.68, 3.88], "n": 24}}, "usd_per_game_hour": 0.0, "value_score": null, "latency_p50_ms": 16200, "latency_p95_ms": 24300, "tokens_per_sec": 60.0, "judge_model": "claude-sonnet-4-6", "dataset_merkle": "12b7a39f0cffe796ae5dcd78129ee597c558f9feaa0dfdb11d297e16b9601cf8", "timestamp": "2026-06-15T19:00:00Z", "note": "PRELIMINARY: dialogue-only, local-MLX, n=24, v2 rubric, judge=claude-sonnet-4-6"}
{"candidate": "mlx-community/Qwen2.5-14B-Instruct-4bit@http://localhost:8000/v1", "model": "mlx-community/Qwen2.5-14B-Instruct-4bit", "tier": "free", "overall": 3.7, "overall_ci95": [3.6, 3.8], "categories": {"dialogue": {"score": 3.7, "ci95": [3.6, 3.8], "n": 24}}, "usd_per_game_hour": 0.0, "value_score": null, "latency_p50_ms": 4500, "latency_p95_ms": 6750, "tokens_per_sec": 60.0, "judge_model": "claude-sonnet-4-6", "dataset_merkle": "12b7a39f0cffe796ae5dcd78129ee597c558f9feaa0dfdb11d297e16b9601cf8", "timestamp": "2026-06-15T19:00:00Z", "note": "PRELIMINARY: dialogue-only, local-MLX, n=24, v2 rubric, judge=claude-sonnet-4-6; BASELINE"}
{"candidate": "unsloth/gemma-4-12b-it-GGUF@http://localhost:8000/v1", "model": "unsloth/gemma-4-12b-it-GGUF", "tier": "free", "overall": 3.68, "overall_ci95": [3.58, 3.78], "categories": {"dialogue": {"score": 3.68, "ci95": [3.58, 3.78], "n": 24}}, "usd_per_game_hour": 0.0, "value_score": null, "latency_p50_ms": 5500, "latency_p95_ms": 8250, "tokens_per_sec": 60.0, "judge_model": "claude-sonnet-4-6", "dataset_merkle": "12b7a39f0cffe796ae5dcd78129ee597c558f9feaa0dfdb11d297e16b9601cf8", "timestamp": "2026-06-15T19:00:00Z", "note": "PRELIMINARY: dialogue-only, local-MLX/llama.cpp, n=24, v2 rubric, judge=claude-sonnet-4-6; replies wrapped in json fences (craft penalty)"}
{"candidate": "mlx-community/Qwen3.5-9B-4bit@http://localhost:8000/v1", "model": "mlx-community/Qwen3.5-9B-4bit", "tier": "free", "overall": 3.49, "overall_ci95": [3.39, 3.59], "categories": {"dialogue": {"score": 3.49, "ci95": [3.39, 3.59], "n": 24}}, "usd_per_game_hour": 0.0, "value_score": null, "latency_p50_ms": 4600, "latency_p95_ms": 6900, "tokens_per_sec": 60.0, "judge_model": "claude-sonnet-4-6", "dataset_merkle": "12b7a39f0cffe796ae5dcd78129ee597c558f9feaa0dfdb11d297e16b9601cf8", "timestamp": "2026-06-15T19:00:00Z", "note": "PRELIMINARY: dialogue-only, local-MLX, n=24, v2 rubric, judge=claude-sonnet-4-6; 2 degenerate loops, not production-ready"}
{"candidate": "mlx-community/gemma-4-26b-a4b-it-4bit-thinking@http://localhost:8000/v1", "model": "mlx-community/gemma-4-26b-a4b-it-4bit (mlx_vlm, thinking=on)", "tier": "free", "overall": 4.08, "overall_ci95": [3.98, 4.18], "categories": {"dialogue": {"score": 4.08, "ci95": [3.98, 4.18], "n": 24}}, "usd_per_game_hour": 0.0, "value_score": null, "latency_p50_ms": 3100, "latency_p95_ms": 4650, "tokens_per_sec": 60.0, "judge_model": "claude-sonnet-4-6", "dataset_merkle": "12b7a39f0cffe796ae5dcd78129ee597c558f9feaa0dfdb11d297e16b9601cf8", "timestamp": "2026-06-15T19:00:00Z", "note": "PRELIMINARY: dialogue-only, local-MLX via mlx_vlm, thinking=ON, n=24, v2 rubric, judge=claude-sonnet-4-6; higher grounding risk than no-think variant"}