兩項目評估報告
- EvalRun: #63（suite #39
auto-256cfb41-r1-053958414）
- Target: #1 production-baseline
- Cases: 27（N=1, attempts=27/27）
- Generated: 2026-05-16T12:25:26+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 / 27 (37.0%)
- 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 |
| 知識與產品查詢 |
10 / 10 |
❌ 40.1% |
✅ 100.0% |
待修方向（worst-3）
- 知識與產品查詢 × retrieval_relevance = 40.1% — 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 |
17 |
100.0% |
| Step 1: scenario_routing |
8 |
47.1% |
| Step 2: + tool_calling |
8 |
47.1% |
| Step 3: + answer |
4 |
23.5% |
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 (8) |
✅ 8/8 (100.0%) |
✅ 8/8 (100.0%) |
⚪ 4/8 (50.0%) |
⚪ 4/8 (50.0%) |
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 (-50.0pp)
Audit
Reproduce combined report: bin/rails runner "puts Eval::EvaluationReport.call(run: EvalRun.find(63))"
Or fetch each item separately:
bin/rails runner "puts Eval::KbAccuracyReport.call(run: EvalRun.find(63))"bin/rails runner "puts Eval::ScenarioFunnelReport.call(run: EvalRun.find(63))"
Per-Scenario × Per-Dim — Run #63
Suite: ASPADZ 客服 (bulk R1) · scenarios: 2 · dims: 5 · populated cells: 9/10
| Scenario |
Scenario |
Tool |
Retrieval |
Faith |
AnsQ |
| early_return |
0.0% [0.0–0.0] (n=9) |
— |
88.9% [66.7–100.0] (n=9) |
100.0% [100.0–100.0] (n=8) |
87.4% [65.2–99.6] (n=9) |
| 知識與產品查詢 |
100.0% [100.0–100.0] (n=18) |
100.0% [100.0–100.0] (n=18) |
88.2% [70.6–100.0] (n=17) |
87.0% [79.6–94.4] (n=18) |
83.0% [70.6–93.3] (n=18) |
Worst-3 cells (lowest primary score)
- early_return × Scenario · 0.0% (n=9) · lowest sub_metric:
scenario_precision
- 知識與產品查詢 × AnsQ · 83.0% (n=18) · lowest sub_metric:
critical_coverage
- 知識與產品查詢 × Faith · 87.0% (n=18) · lowest sub_metric:
rule_compliance