兩項目評估報告
- EvalRun: #276（suite #125
auto-a5f97610-r1-053959570）
- Target: #2 production-canary
- Cases: 16（N=1, attempts=16/16）
- Generated: 2026-05-16T14:50:19+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: 8 / 16 (50.0%)
- 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 / 3 |
— |
— |
| 知識與產品查詢 |
6 / 6 |
❌ 0.0% |
— |
| 訂單查詢 |
2 / 6 |
❌ 0.0% |
— |
| 轉接真人客服 |
0 / 1 |
— |
— |
待修方向（worst-3）
- 訂單查詢 × retrieval_relevance = 0.0% — bot 進對 scenario 但沒呼叫 search tool — 流程 / prompt 問題（不是 KB content gap）
- 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）
項目二: 情境調用與完成
整體 funnel（全 scenarios 加總）
| Stage |
Pass count |
% of total |
| Total attempts |
16 |
100.0% |
| Step 1: scenario_routing |
8 |
50.0% |
| Step 2: + tool_calling |
8 |
50.0% |
| Step 3: + answer |
3 |
18.8% |
Per-scenario funnel
| Scenario (n_attempts) |
Step 1 (routing) |
Step 2 (tool) |
Step 3 (answer) |
end-to-end |
| early_return (3) |
❌ 0/3 (0.0%) |
❌ 0/3 (0.0%) |
❌ 0/3 (0.0%) |
❌ 0/3 (0.0%) |
| 知識與產品查詢 (6) |
✅ 6/6 (100.0%) |
✅ 6/6 (100.0%) |
❌ 2/6 (33.3%) |
❌ 2/6 (33.3%) |
| 訂單查詢 (6) |
❌ 2/6 (33.3%) |
❌ 2/6 (33.3%) |
❌ 1/6 (16.7%) |
❌ 1/6 (16.7%) |
| 轉接真人客服 (1) |
❌ 0/1 (0.0%) |
❌ 0/1 (0.0%) |
❌ 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 step1 (-100.0pp)
- 知識與產品查詢 drop at step3 (-66.7pp)
- 訂單查詢 drop at step1 (-66.7pp)
Audit
Reproduce combined report: bin/rails runner "puts Eval::EvaluationReport.call(run: EvalRun.find(276))"
Or fetch each item separately:
bin/rails runner "puts Eval::KbAccuracyReport.call(run: EvalRun.find(276))"bin/rails runner "puts Eval::ScenarioFunnelReport.call(run: EvalRun.find(276))"
Per-Scenario × Per-Dim — Run #276
Suite: test (bulk R1) · scenarios: 4 · dims: 5 · populated cells: 14/20
| Scenario |
Scenario |
Tool |
Retrieval |
Faith |
AnsQ |
| early_return |
0.0% [0.0–0.0] (n=3) |
0.0% [0.0–0.0] (n=3) |
— |
— |
63.3% [43.3–83.3] (n=3) |
| 知識與產品查詢 |
100.0% [100.0–100.0] (n=6) |
100.0% [100.0–100.0] (n=6) |
— |
0.0% [0.0–0.0] (n=6) |
65.0% [50.0–81.1] (n=6) |
| 訂單查詢 |
66.7% [16.7–100.0] (n=6) |
33.3% [0.0–66.7] (n=6) |
— |
60.0% [20.0–100.0] (n=5) |
62.8% [47.8–80.6] (n=6) |
| 轉接真人客服 |
0.0% (n=1) |
— |
— |
0.0% (n=1) |
33.3% (n=1) |
Worst-3 cells (lowest primary score)
- early_return × Scenario · 0.0% (n=3) · lowest sub_metric:
scenario_precision
- early_return × Tool · 0.0% (n=3) · lowest sub_metric:
tools_recall
- 知識與產品查詢 × Faith · 0.0% (n=6) · lowest sub_metric:
rule_compliance