.67 · 单轮详情
20261107-huanxi-0.8.14.67-t1
测量日 2026-11-07 · source SHA f29d351788d092073d2ae7d6c08755ef7cda9445
schema 缺测
provider: ppio；template: teams-work-v1；cpu: 4；memory_mib: 8192；os: Debian 13；deployment_kind: fresh_isolated
逐用户输入
| 输入 | 原生终态 | 回复延迟 / 秒 | CLI / 观察耗时 / 秒 | 模型响应 | 工具调用 |
|---|---|---|---|---|---|
| C01/U1 | failed | 173.6 | 174.4 | 18 | 14 |
| C02/U1 | failed | 306.2 | 306.4 | 35 | 34 |
| C03/U1 | completed | 134.1 | 135.8 | 17 | 18 |
| C04/U1 | completed | 384.4 | 384.7 | 38 | 34 |
| C05/U1 | completed | 145.9 | 146.2 | 12 | 11 |
| C06/U1 | completed | 478.4 | 479.6 | 41 | 37 |
| C07/U1 | completed | 43.6 | 45.4 | 4 | 3 |
| C07/U2 | completed | 70.6 | 72.2 | 6 | 5 |
| C08/U1 | failed | 150.5 | 152.4 | 16 | 12 |
| C08/U2 | completed | 46.9 | 缺测 | 5 | 4 |
| C09/U1 | completed | 169.4 | 171.0 | 19 | 19 |
| C09/U2 | completed | 341.6 | 342.3 | 31 | 29 |
| C10/U1 | completed | 268.3 | 270.2 | 31 | 34 |
| C10/U2 | completed | 305.6 | 307.3 | 28 | 27 |
逐题内容判分
| 题目 | 内容复核 | 原检查器 | 回复合计 / 秒 |
|---|---|---|---|
| C01 发票去重与对账 | PASS | PASS | 173.6 |
| C02 日志关联与事故时间线 | PASS | PASS | 306.2 |
| C03 多文件政策优先级 | PASS | PASS | 134.1 |
| C04 时区与稳定排序 | PASS | PASS | 384.4 |
| C05 小模块修复与回归 | PASS | PASS | 145.9 |
| C06 依赖图与关键路径 | PASS | PASS | 478.4 |
| C07 采购方案追加预算约束 | PASS | PASS | 114.2 |
| C08 销售汇总与增量更正 | PASS | FAIL | 197.4 |
| C09 命令行工具连续迭代 | PASS | PASS | 511.0 |
| C10 发布判断与新证据修正 | INCONCLUSIVE | FAIL | 573.9 |
指标与缺测
{
"artifact_grades": {
"INCONCLUSIVE": 1,
"PASS": 9
},
"invocation_or_observer": {
"expected": 14,
"observed": 13,
"observed_sum_s": 2987.926929453001,
"sum_s": null
},
"model_id_mismatch": 0,
"model_id_missing": 0,
"native_tool_calls": 281,
"observed_model_ids": {
"deepseek/deepseek-v4.1-flash-1017": 301
},
"observed_model_intervals": {
"expected": 301,
"observed": 301,
"observed_sum_s": 2831.535884521484,
"sum_s": 2831.535884521484
},
"observed_tmp_commands": 34,
"observed_tool_intervals": {
"expected": 281,
"observed": 281,
"observed_sum_s": 43.606087158203124,
"sum_s": 43.606087158203124
},
"raw_artifact_grades": {
"FAIL": 2,
"PASS": 8
},
"response_latency": {
"expected": 14,
"observed": 14,
"observed_sum_s": 3019.006586074829,
"sum_s": 3019.006586074829
},
"scope_observed_cases": 10,
"terminals": {
"completed": 11,
"failed": 3
},
"tmp_command_cases": [
"C01",
"C03",
"C04",
"C06",
"C07",
"C09",
"C10"
],
"tool_distribution": {
"call_tool": 23,
"capability_load": 6,
"capability_search": 6,
"shell": 246
},
"tool_input_availability": {
"available": 258,
"unavailable": 23
},
"tool_missing_arguments": 23,
"tool_missing_results": 23,
"tool_output_availability": {
"available": 258,
"unavailable": 23
},
"user_turns": 14,
"visible_model_responses": 301
}采集与覆盖说明（manifest 原文）
- All declared user turns have final reply timestamps
- Native model counts do not establish auxiliary/provider request completeness
- C08 U2 observer recovered read-only; 45.136s overlap with C09 U1;23 call_tool argument/result bodies missing
口径与限制
n=1；并发争用、模型随机性、跨日基线未受控；不能据此作统计排名或因果结论。输入终态、内容判分、范围遵守与采集完整度分开呈现。
固定基线为2026-11-07 Codex0.165.0 / OpenClaw2026.11.1；欢喜OS各版本单独新增试次。相同10题、14用户输入，liqian deepseek/deepseek-v4.1-flash-1017。 每次报告明确区分：输入→原生终态延迟、CLI/观察收尾、模型响应、工具步骤、内容判分、范围遵守与证据缺测。跨日固定基线不是同时随机对照。
- Relative to historical .68/.69 (teams-work-v1), this run uses a different teams-work-v2 environment. Main comparison uses same-day .69 v2 environment anchor.
- Six versions execute concurrently in separate 4C/8G sandboxes; cross-version provider contention and model nondeterminism are uncontrolled. One trial per case; no statistical ranking or causal attribution.
内容复核口径（comparisons/latest.md）：历史复核：C08 原检查器漏查子目录备份，原判保留、内容复核 PASS；C10 部署清单完整性与 JSON 层级题面有歧义，保留 INCONCLUSIVE。
主耗时不包含CLI终态后退出等待。模型/工具时间线允许重叠；不得相加当整体墙钟。单trial及跨日历史基线只支持探索性比较。