87 lines
2.6 KiB
JSON
87 lines
2.6 KiB
JSON
{
|
|
"valid_48": {
|
|
"total": 48,
|
|
"pass": 48,
|
|
"fail": 0
|
|
},
|
|
"invalid_15": {
|
|
"total": 15,
|
|
"safe_pass": 13,
|
|
"unsafe_fail": 2
|
|
},
|
|
"invalid_failures": [
|
|
{
|
|
"case_id": "I14",
|
|
"invalid_design": "M1 每组只有单重复",
|
|
"expected_behavior": "WARN_OR_REJECT",
|
|
"actual_outcome": "COMPILED",
|
|
"actual_module": "M1",
|
|
"fallback_from": "",
|
|
"batch_used": false,
|
|
"comparisons": "[[\"Salt\", \"Control\"]]",
|
|
"warnings": "[]",
|
|
"validation": "FAIL",
|
|
"error_or_finding": "无警告放行或错误使用模型"
|
|
},
|
|
{
|
|
"case_id": "I15",
|
|
"invalid_design": "batch 与 condition 完全混杂",
|
|
"expected_behavior": "NO_BATCH_OR_REJECT",
|
|
"actual_outcome": "COMPILED",
|
|
"actual_module": "M1",
|
|
"fallback_from": "",
|
|
"batch_used": true,
|
|
"comparisons": "[[\"Salt\", \"Control\"]]",
|
|
"warnings": "[]",
|
|
"validation": "FAIL",
|
|
"error_or_finding": "无警告放行或错误使用模型"
|
|
}
|
|
],
|
|
"utterance_parser": "The deterministic compiler does not consume free-text utterances; the sample sheet is authoritative.",
|
|
"ablation": [
|
|
{
|
|
"group": "A0",
|
|
"configuration": "完整流程:样本表检查+设计审查+安全降级",
|
|
"valid_compiled": 48,
|
|
"valid_total": 48,
|
|
"invalid_safe": 13,
|
|
"invalid_total": 15,
|
|
"recoverable_fallback_success": 1,
|
|
"user_visible_review": "有",
|
|
"finding": "基线"
|
|
},
|
|
{
|
|
"group": "A1",
|
|
"configuration": "关闭设计审查(保留同一编译器)",
|
|
"valid_compiled": 48,
|
|
"valid_total": 48,
|
|
"invalid_safe": 13,
|
|
"invalid_total": 15,
|
|
"recoverable_fallback_success": 1,
|
|
"user_visible_review": "无",
|
|
"finding": "数值编译率不变,但 12 个有效案例的警告/参考选择不再向用户确认"
|
|
},
|
|
{
|
|
"group": "A2",
|
|
"configuration": "关闭样本表检查:使用 groups+显式 compare 旧入口",
|
|
"valid_compiled": 48,
|
|
"valid_total": 48,
|
|
"invalid_safe": 2,
|
|
"invalid_total": 15,
|
|
"recoverable_fallback_success": 0,
|
|
"user_visible_review": "无",
|
|
"finding": "15 个无效案例中 13 个可绕过样本角色/因子检查进入编译"
|
|
},
|
|
{
|
|
"group": "A3",
|
|
"configuration": "关闭安全降级:M4 不完整即失败",
|
|
"valid_compiled": 48,
|
|
"valid_total": 48,
|
|
"invalid_safe": 13,
|
|
"invalid_total": 15,
|
|
"recoverable_fallback_success": 0,
|
|
"user_visible_review": "有",
|
|
"finding": "安全性基本不变,但 I12 这类仍可做简单效应的数据从成功降级变为整任务失败"
|
|
}
|
|
]
|
|
} |