兩項目評估報告
- EvalRun: #208（suite #118
auto-056b78a4-r1-053959465）
- Target: #2 production-canary
- Cases: 30（N=1, attempts=30/30）
- Generated: 2026-05-16T13:51:31+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 |
❌ 44.6% |
✅ 85.0% |
待修方向（worst-3）
- 知識與產品查詢 × retrieval_relevance = 44.6% — 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 = 85.0% — 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 |
8 |
40.0% |
| Step 2: + tool_calling |
8 |
40.0% |
| Step 3: + answer |
2 |
10.0% |
Per-scenario funnel
| Scenario (n_attempts) |
Step 1 (routing) |
Step 2 (tool) |
Step 3 (answer) |
end-to-end |
| early_return (9) |
❌ 0/9 (0.0%) |
❌ 0/9 (0.0%) |
❌ 0/9 (0.0%) |
❌ 0/9 (0.0%) |
| 知識與產品查詢 (6) |
✅ 6/6 (100.0%) |
✅ 6/6 (100.0%) |
❌ 1/6 (16.7%) |
❌ 1/6 (16.7%) |
| 試騎推薦 (5) |
❌ 2/5 (40.0%) |
❌ 2/5 (40.0%) |
❌ 1/5 (20.0%) |
❌ 1/5 (20.0%) |
Drop-off 最大的 5 個 scenario
- early_return drop at step1 (-100.0pp)
- 知識與產品查詢 drop at step3 (-83.3pp)
- 試騎推薦 drop at step1 (-60.0pp)
Audit
Reproduce combined report: bin/rails runner "puts Eval::EvaluationReport.call(run: EvalRun.find(208))"
Or fetch each item separately:
bin/rails runner "puts Eval::KbAccuracyReport.call(run: EvalRun.find(208))"bin/rails runner "puts Eval::ScenarioFunnelReport.call(run: EvalRun.find(208))"
Per-Scenario × Per-Dim — Run #208
Suite: SOKER 官方騎行顧問 (bulk R1) · scenarios: 3 · dims: 5 · populated cells: 14/15
| Scenario |
Scenario |
Tool |
Retrieval |
Faith |
AnsQ |
| early_return |
0.0% [0.0–0.0] (n=9) |
— |
100.0% [100.0–100.0] (n=9) |
94.4% [83.3–100.0] (n=9) |
97.8% [95.9–99.3] (n=9) |
| 知識與產品查詢 |
100.0% [100.0–100.0] (n=16) |
100.0% [100.0–100.0] (n=16) |
86.7% [73.3–96.7] (n=15) |
71.9% [54.2–88.5] (n=16) |
78.1% [60.8–92.9] (n=16) |
| 試騎推薦 |
100.0% [100.0–100.0] (n=5) |
40.0% [0.0–80.0] (n=5) |
50.0% [10.0–90.0] (n=5) |
46.7% [13.3–80.0] (n=5) |
20.0% [0.0–60.0] (n=5) |
Worst-3 cells (lowest primary score)
- early_return × Scenario · 0.0% (n=9) · lowest sub_metric:
scenario_precision
- 試騎推薦 × AnsQ · 20.0% (n=5) · lowest sub_metric:
core_coverage
- 試騎推薦 × Tool · 40.0% (n=5) · lowest sub_metric:
tools_precision