兩項目評估報告
- EvalRun: #230(suite #130
auto-03213b91-r1-053959613)
- Target: #2 production-canary
- Cases: 30(N=1, attempts=30/30)
- Generated: 2026-05-16T13:52:08+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 |
❌ 42.3% |
✅ 100.0% |
待修方向(worst-3)
- 知識與產品查詢 × retrieval_relevance = 42.3% — 撈到的 KB chunks 不夠相關 — KB content gap,跟 Neptune team 提 enrichment
- 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 |
9 |
45.0% |
| Step 2: + tool_calling |
9 |
45.0% |
| Step 3: + answer |
8 |
40.0% |
Per-scenario funnel
| Scenario (n_attempts) |
Step 1 (routing) |
Step 2 (tool) |
Step 3 (answer) |
end-to-end |
| 2026春季盃 (1) |
✅ 1/1 (100.0%) |
✅ 1/1 (100.0%) |
✅ 1/1 (100.0%) |
✅ 1/1 (100.0%) |
| JQT諮詢 (1) |
✅ 1/1 (100.0%) |
✅ 1/1 (100.0%) |
✅ 1/1 (100.0%) |
✅ 1/1 (100.0%) |
| early_return (10) |
❌ 0/10 (0.0%) |
❌ 0/10 (0.0%) |
❌ 0/10 (0.0%) |
❌ 0/10 (0.0%) |
| 官網訂單查詢 (1) |
✅ 1/1 (100.0%) |
✅ 1/1 (100.0%) |
✅ 1/1 (100.0%) |
✅ 1/1 (100.0%) |
| 官網購物-售後處理 (修改/退換/維修) (1) |
❌ 0/1 (0.0%) |
❌ 0/1 (0.0%) |
❌ 0/1 (0.0%) |
❌ 0/1 (0.0%) |
| 客製服飾進度詢問 (1) |
✅ 1/1 (100.0%) |
✅ 1/1 (100.0%) |
✅ 1/1 (100.0%) |
✅ 1/1 (100.0%) |
| 知識與產品查詢 (1) |
✅ 1/1 (100.0%) |
✅ 1/1 (100.0%) |
✅ 1/1 (100.0%) |
✅ 1/1 (100.0%) |
| 社會聯賽 (2) |
✅ 2/2 (100.0%) |
✅ 2/2 (100.0%) |
✅ 2/2 (100.0%) |
✅ 2/2 (100.0%) |
| 門市購物-售後處理 (1) |
✅ 1/1 (100.0%) |
✅ 1/1 (100.0%) |
✅ 1/1 (100.0%) |
✅ 1/1 (100.0%) |
| 預約取貨 (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)
- 官網購物-售後處理 (修改/退換/維修) drop at step1 (-100.0pp)
- 預約取貨 drop at step3 (-100.0pp)
- 2026春季盃 drop at step1 (-0.0pp)
- JQT諮詢 drop at step1 (-0.0pp)
Audit
Reproduce combined report: bin/rails runner "puts Eval::EvaluationReport.call(run: EvalRun.find(230))"
Or fetch each item separately:
bin/rails runner "puts Eval::KbAccuracyReport.call(run: EvalRun.find(230))"
bin/rails runner "puts Eval::ScenarioFunnelReport.call(run: EvalRun.find(230))"
Per-Scenario × Per-Dim — Run #230
Suite: TGF 客服小幫手 (bulk R1) · scenarios: 10 · dims: 5 · populated cells: 47/50
| Scenario |
Scenario |
Tool |
Retrieval |
Faith |
AnsQ |
| 2026春季盃 |
100.0% (n=1) |
100.0% (n=1) |
0.0% (n=1) |
0.0% (n=1) |
90.0% (n=1) |
| JQT諮詢 |
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=10) |
0.0% (n=1) |
66.7% [33.3–100.0] (n=9) |
100.0% [100.0–100.0] (n=10) |
79.7% [64.7–92.3] (n=10) |
| 官網訂單查詢 |
100.0% (n=1) |
0.0% (n=1) |
0.0% (n=1) |
0.0% (n=1) |
93.3% (n=1) |
| 官網購物-售後處理 (修改/退換/維修) |
100.0% (n=1) |
0.0% (n=1) |
0.0% (n=1) |
100.0% (n=1) |
83.3% (n=1) |
| 客製服飾進度詢問 |
100.0% (n=1) |
100.0% (n=1) |
100.0% (n=1) |
100.0% (n=1) |
83.3% (n=1) |
| 知識與產品查詢 |
100.0% [100.0–100.0] (n=11) |
0.0% [0.0–0.0] (n=11) |
63.6% [36.4–90.9] (n=11) |
97.0% [90.9–100.0] (n=11) |
93.9% [89.1–98.2] (n=11) |
| 社會聯賽 |
100.0% [100.0–100.0] (n=2) |
— |
0.0% [0.0–0.0] (n=2) |
100.0% [100.0–100.0] (n=2) |
98.3% [96.7–100.0] (n=2) |
| 門市購物-售後處理 |
100.0% (n=1) |
— |
0.0% (n=1) |
100.0% (n=1) |
100.0% (n=1) |
| 預約取貨 |
100.0% (n=1) |
100.0% (n=1) |
0.0% (n=1) |
0.0% (n=1) |
43.3% (n=1) |
Worst-3 cells (lowest primary score)
- 官網購物-售後處理 (修改/退換/維修) × Tool · 0.0% (n=1) · lowest sub_metric:
tools_recall
- 官網購物-售後處理 (修改/退換/維修) × Retrieval · 0.0% (n=1) · lowest sub_metric:
knowledges_source_precision
- 官網訂單查詢 × Faith · 0.0% (n=1) · lowest sub_metric:
rule_compliance