jina-v3 + deepseek-chat · k=20
run #19 · 2026-06-22 06:08 · git aa47cbb · deploy:pg-chunks_v2
Retrieval hit@k
80%
Answer accuracy
91%
Cases
45 (40 grounded)
pg FTS hybrid + planner + reranker on chunks_v2 (deploy validation)
Δ vs baseline
Compared to jina-v3 + deepseek-chat · k=20 (2026-06-22 05:20). A retrieval delta with the same generator isolates the embedder/retrieval change.
- ✓ Retrieval solved here (1): q040
- ⚠ Retrieval regressed (3): q008, q027, q029
- ✓ Answer solved here (4): q013, q021, q023, q040
- ⚠ Answer regressed (2): q027, q041
Failures in this run (11)
| Case | Question | Retrieval | Answer | Symptom |
|---|---|---|---|---|
| q008 | What was Berkshire Hathaway's shareholders' equity at the end of 2023? | FAIL | PASS | answered without the gold chunk (parametric leak?) |
| q014 | By how much did Berkshire Hathaway's insurance float grow from 2022 to 2023? | PASS | FAIL | retrieved but generated wrong (generation/arithmetic) |
| q017 | What kind of transportation business does Berkshire Hathaway operate? | FAIL | PASS | answered without the gold chunk (parametric leak?) |
| q022 | How much did Berkshire Hathaway's operating earnings grow from 2021 to 2023? | PASS | FAIL | retrieved but generated wrong (generation/arithmetic) |
| q025 | What was JPMorgan Chase's total net revenue in 2023? | FAIL | PASS | answered without the gold chunk (parametric leak?) |
| q027 | What were JPMorgan Chase's total assets at the end of 2023? | FAIL | FAIL | gold chunk not retrieved → answer ungrounded |
| q028 | What was JPMorgan Chase's CET1 capital ratio in 2023? | FAIL | PASS | answered without the gold chunk (parametric leak?) |
| q029 | What were JPMorgan Chase's cash dividends declared per share in 2023? | FAIL | PASS | answered without the gold chunk (parametric leak?) |
| q034 | What were Costco's net sales in fiscal 2023? | FAIL | PASS | answered without the gold chunk (parametric leak?) |
| q041 | How much greater was Costco's total revenue than Microsoft's in 2023? | PASS | FAIL | retrieved but generated wrong (generation/arithmetic) |
| q043 | Which had the highest net income in 2023: JPMorgan Chase, Microsoft, or Costco? | FAIL | PASS | answered without the gold chunk (parametric leak?) |