兩項目評估報告
- EvalRun: #128（suite #71
auto-9fd2a0a3-r1-053958777）
- Target: #2 production-canary
- Cases: 13（N=1, attempts=13/13）
- Generated: 2026-05-16T13:29:37+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: 12 / 13 (92.3%)
- 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 |
| early_return |
0 / 1 |
— |
— |
| 知識與產品查詢 |
11 / 11 |
❌ 0.0% |
⚪ 59.1% |
| 轉接真人客服 |
1 / 1 |
❌ 0.0% |
— |
待修方向（worst-3）
- 知識與產品查詢 × retrieval_relevance = 0.0% — bot 進對 scenario 但沒呼叫 search tool — 流程 / prompt 問題（不是 KB content gap）
- 轉接真人客服 × retrieval_relevance = 0.0% — bot 進對 scenario 但沒呼叫 search tool — 流程 / prompt 問題（不是 KB content gap）
- 知識與產品查詢 × answer_correctness = 59.1% — 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 |
13 |
100.0% |
| Step 1: scenario_routing |
12 |
92.3% |
| Step 2: + tool_calling |
12 |
92.3% |
| Step 3: + answer |
4 |
30.8% |
Per-scenario funnel
| Scenario (n_attempts) |
Step 1 (routing) |
Step 2 (tool) |
Step 3 (answer) |
end-to-end |
| early_return (1) |
❌ 0/1 (0.0%) |
❌ 0/1 (0.0%) |
❌ 0/1 (0.0%) |
❌ 0/1 (0.0%) |
| 知識與產品查詢 (11) |
✅ 11/11 (100.0%) |
✅ 11/11 (100.0%) |
❌ 4/11 (36.4%) |
❌ 4/11 (36.4%) |
| 轉接真人客服 (1) |
✅ 1/1 (100.0%) |
✅ 1/1 (100.0%) |
❌ 0/1 (0.0%) |
❌ 0/1 (0.0%) |
Drop-off 最大的 5 個 scenario
- early_return drop at step1 (-100.0pp)
- 轉接真人客服 drop at step3 (-100.0pp)
- 知識與產品查詢 drop at step3 (-63.6pp)
Audit
Reproduce combined report: bin/rails runner "puts Eval::EvaluationReport.call(run: EvalRun.find(128))"
Or fetch each item separately:
bin/rails runner "puts Eval::KbAccuracyReport.call(run: EvalRun.find(128))"bin/rails runner "puts Eval::ScenarioFunnelReport.call(run: EvalRun.find(128))"
Per-Scenario × Per-Dim — Run #128
Suite: HCG和成 (bulk R1) · scenarios: 3 · dims: 5 · populated cells: 11/15
| Scenario |
Scenario |
Tool |
Retrieval |
Faith |
AnsQ |
| early_return |
0.0% (n=1) |
— |
100.0% (n=1) |
— |
93.3% (n=1) |
| 知識與產品查詢 |
100.0% [100.0–100.0] (n=11) |
100.0% [100.0–100.0] (n=11) |
100.0% [100.0–100.0] (n=10) |
39.4% [27.3–51.5] (n=11) |
59.7% [49.4–69.4] (n=11) |
| 轉接真人客服 |
100.0% (n=1) |
— |
— |
0.0% (n=1) |
36.7% (n=1) |
Worst-3 cells (lowest primary score)
- early_return × Scenario · 0.0% (n=1) · lowest sub_metric:
scenario_precision
- 轉接真人客服 × Faith · 0.0% (n=1) · lowest sub_metric:
rule_compliance
- 轉接真人客服 × AnsQ · 36.7% (n=1) · lowest sub_metric:
answer_correctness