兩項目評估報告
- EvalRun: #212(suite #123
auto-34eb669a-r1-053959549)
- Target: #2 production-canary
- Cases: 30(N=1, attempts=30/30)
- Generated: 2026-05-16T13:52:06+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 |
❌ 38.5% |
✅ 100.0% |
待修方向(worst-3)
- 知識與產品查詢 × retrieval_relevance = 38.5% — 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 |
10 |
50.0% |
| Step 2: + tool_calling |
10 |
50.0% |
| Step 3: + answer |
9 |
45.0% |
Per-scenario funnel
| Scenario (n_attempts) |
Step 1 (routing) |
Step 2 (tool) |
Step 3 (answer) |
end-to-end |
| Bug回報情境 (6) |
✅ 5/6 (83.3%) |
✅ 5/6 (83.3%) |
✅ 5/6 (83.3%) |
✅ 5/6 (83.3%) |
| 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 (1) |
✅ 1/1 (100.0%) |
✅ 1/1 (100.0%) |
❌ 0/1 (0.0%) |
❌ 0/1 (0.0%) |
| 訂單查詢 (4) |
✅ 4/4 (100.0%) |
✅ 4/4 (100.0%) |
✅ 4/4 (100.0%) |
✅ 4/4 (100.0%) |
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 (-100.0pp)
- Bug回報情境 drop at step1 (-16.7pp)
- 訂單查詢 drop at step1 (-0.0pp)
Audit
Reproduce combined report: bin/rails runner "puts Eval::EvaluationReport.call(run: EvalRun.find(212))"
Or fetch each item separately:
bin/rails runner "puts Eval::KbAccuracyReport.call(run: EvalRun.find(212))"
bin/rails runner "puts Eval::ScenarioFunnelReport.call(run: EvalRun.find(212))"
Per-Scenario × Per-Dim — Run #212
Suite: SWAG Creator AI bot (bulk R1) · scenarios: 4 · dims: 5 · populated cells: 18/20
| Scenario |
Scenario |
Tool |
Retrieval |
Faith |
AnsQ |
| Bug回報情境 |
100.0% [100.0–100.0] (n=6) |
— |
0.0% [0.0–0.0] (n=5) |
77.8% [44.4–100.0] (n=6) |
68.3% [58.9–83.3] (n=6) |
| early_return |
0.0% [0.0–0.0] (n=9) |
100.0% (n=1) |
100.0% [100.0–100.0] (n=8) |
87.5% [62.5–100.0] (n=8) |
80.4% [59.6–95.6] (n=9) |
| 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) |
90.9% [72.7–100.0] (n=11) |
87.9% [72.7–97.0] (n=11) |
| 訂單查詢 |
100.0% [100.0–100.0] (n=4) |
100.0% [100.0–100.0] (n=4) |
— |
75.0% [25.0–100.0] (n=4) |
88.3% [86.7–90.0] (n=4) |
Worst-3 cells (lowest primary score)
- Bug回報情境 × Retrieval · 0.0% (n=5) · lowest sub_metric:
knowledges_source_recall
- early_return × Scenario · 0.0% (n=9) · lowest sub_metric:
scenario_precision
- Bug回報情境 × AnsQ · 68.3% (n=6) · lowest sub_metric:
critical_coverage