兩項目評估報告
- EvalRun: #77（suite #43
auto-e67d13e3-r1-053958483）
- Target: #1 production-baseline
- Cases: 30（N=1, attempts=30/30）
- Generated: 2026-05-16T12:25:25+08:00
本報告含兩個獨立評估項目。項目一 評估 bot 進入「知識與產品查詢」fallback
後的撈取 + 回答能力；項目二 評估 bot 跨全 enabled scenarios 的 routing →
tool → answer 三段 funnel 健康度。
項目一: 知識庫精準度
評估範圍 (prerequisite: 進對 scenario)
- 進對 scenario 的 attempts: 10 / 30 (33.3%)
- 只對這些 attempts 算 retrieval + answer
- 沒進對 scenario 的 attempts → 排除（routing 失敗不該污染 KB 訊號）
Per-scenario 評估
| Scenario |
qualifying / total attempts |
retrieval_relevance |
answer_correctness |
| 知識與產品查詢 |
10 / 10 |
⚪ 53.3% |
✅ 100.0% |
待修方向（worst-3）
- 知識與產品查詢 × retrieval_relevance = 53.3% — retrieval 略低 — 看 LLM relevance vs topic_match 哪邊弱
- 知識與產品查詢 × answer_correctness = 100.0% — 回答內容有問題 — 細看 hallucination_rate vs answer_quality 哪邊低
項目二: 情境調用與完成
整體 funnel（全 scenarios 加總）
| Stage |
Pass count |
% of total |
| Total attempts |
20 |
100.0% |
| Step 1: scenario_routing |
7 |
35.0% |
| Step 2: + tool_calling |
7 |
35.0% |
| Step 3: + answer |
5 |
25.0% |
Per-scenario funnel
| Scenario (n_attempts) |
Step 1 (routing) |
Step 2 (tool) |
Step 3 (answer) |
end-to-end |
| FAQ查詢 (1) |
❌ 0/1 (0.0%) |
❌ 0/1 (0.0%) |
❌ 0/1 (0.0%) |
❌ 0/1 (0.0%) |
| VESA Release System 功能確認 (1) |
✅ 1/1 (100.0%) |
✅ 1/1 (100.0%) |
✅ 1/1 (100.0%) |
✅ 1/1 (100.0%) |
| early_return (9) |
❌ 0/9 (0.0%) |
❌ 0/9 (0.0%) |
❌ 0/9 (0.0%) |
❌ 0/9 (0.0%) |
| 保固問題 (1) |
✅ 1/1 (100.0%) |
✅ 1/1 (100.0%) |
❌ 0/1 (0.0%) |
❌ 0/1 (0.0%) |
| 新增情境 (3) |
✅ 3/3 (100.0%) |
✅ 3/3 (100.0%) |
✅ 3/3 (100.0%) |
✅ 3/3 (100.0%) |
| 測試用 (1) |
❌ 0/1 (0.0%) |
❌ 0/1 (0.0%) |
❌ 0/1 (0.0%) |
❌ 0/1 (0.0%) |
| 產品查詢 (2) |
❌ 0/2 (0.0%) |
❌ 0/2 (0.0%) |
❌ 0/2 (0.0%) |
❌ 0/2 (0.0%) |
| 確認螢幕與支架的相容性 (1) |
✅ 1/1 (100.0%) |
✅ 1/1 (100.0%) |
❌ 0/1 (0.0%) |
❌ 0/1 (0.0%) |
| 訂單查詢 (1) |
✅ 1/1 (100.0%) |
✅ 1/1 (100.0%) |
✅ 1/1 (100.0%) |
✅ 1/1 (100.0%) |
Drop-off 最大的 5 個 scenario
- FAQ查詢 drop at step1 (-100.0pp)
- early_return drop at step1 (-100.0pp)
- 保固問題 drop at step3 (-100.0pp)
- 測試用 drop at step1 (-100.0pp)
- 產品查詢 drop at step1 (-100.0pp)
Audit
Reproduce combined report: bin/rails runner "puts Eval::EvaluationReport.call(run: EvalRun.find(77))"
Or fetch each item separately:
bin/rails runner "puts Eval::KbAccuracyReport.call(run: EvalRun.find(77))"bin/rails runner "puts Eval::ScenarioFunnelReport.call(run: EvalRun.find(77))"
Per-Scenario × Per-Dim — Run #77
Suite: BenQ _ Test (bulk R1) · scenarios: 10 · dims: 5 · populated cells: 39/50
| Scenario |
Scenario |
Tool |
Retrieval |
Faith |
AnsQ |
| FAQ查詢 |
0.0% (n=1) |
0.0% (n=1) |
— |
66.7% (n=1) |
40.0% (n=1) |
| VESA Release System 功能確認 |
100.0% (n=1) |
— |
0.0% (n=1) |
100.0% (n=1) |
100.0% (n=1) |
| early_return |
0.0% [0.0–0.0] (n=9) |
— |
66.7% [33.3–88.9] (n=9) |
100.0% [100.0–100.0] (n=8) |
98.1% [95.9–100.0] (n=9) |
| 保固問題 |
100.0% (n=1) |
— |
— |
— |
83.3% (n=1) |
| 新增情境 |
100.0% [100.0–100.0] (n=3) |
— |
0.0% [0.0–0.0] (n=3) |
66.7% [0.0–100.0] (n=3) |
96.7% [96.7–96.7] (n=3) |
| 測試用 |
100.0% (n=1) |
— |
— |
0.0% (n=1) |
16.7% (n=1) |
| 產品查詢 |
0.0% [0.0–0.0] (n=2) |
0.0% [0.0–0.0] (n=2) |
0.0% [0.0–0.0] (n=2) |
66.7% [66.7–66.7] (n=2) |
51.7% [50.0–53.3] (n=2) |
| 知識與產品查詢 |
100.0% [100.0–100.0] (n=10) |
0.0% [0.0–0.0] (n=10) |
70.0% [40.0–100.0] (n=10) |
100.0% [100.0–100.0] (n=10) |
97.7% [95.0–100.0] (n=10) |
| 確認螢幕與支架的相容性 |
100.0% (n=1) |
— |
0.0% (n=1) |
0.0% (n=1) |
43.3% (n=1) |
| 訂單查詢 |
100.0% (n=1) |
100.0% (n=1) |
— |
0.0% (n=1) |
83.3% (n=1) |
Worst-3 cells (lowest primary score)
- 訂單查詢 × Faith · 0.0% (n=1) · lowest sub_metric:
rule_compliance
- 確認螢幕與支架的相容性 × Faith · 0.0% (n=1) · lowest sub_metric:
rule_compliance
- 確認螢幕與支架的相容性 × Retrieval · 0.0% (n=1) · lowest sub_metric:
knowledges_source_recall