Skip to content

feat(eval): confusion clusters, and the corpus reaches 53 chunks #205

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
mrsibe merged 2 commits into from
Sep 30, 2026
Merged

feat(eval): confusion clusters, and the corpus reaches 53 chunks #205

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
20a8c75
feat(eval): a confusion cluster, and the corpus gets its first hard n…
mrsibe Sep 30, 2026
3ffee33
feat(eval): a second confusion cluster, and the corpus reaches 53 chunks
mrsibe Sep 30, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
998 changes: 858 additions & 140 deletions docs/eval/baseline-v1.6.json
View file
Open in desktop

Large diffs are not rendered by default.

43 changes: 22 additions & 21 deletions docs/eval/baseline-v1.6.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -11,23 +11,23 @@ Generated by `npm run eval`. The numbers below are harness output — do not edi
| Retrieval | `dense` |
| Ranks | `candidateK=20, threshold=0.5` |
| Context width | `contextK=3` |
| Corpus | `eval/corpus` (13 documents) |
| Split | `all` (44 questions, 38 answerable) |
| Index size | 19 chunks |
| Corpus | `eval/corpus` (21 documents) |
| Split | `all` (63 questions, 53 answerable) |
| Index size | 53 chunks |

## Metrics

| Metric | Value |
| --- | --- |
| Recall@1 | 0.6579 |
| Recall@5 | 0.9211 |
| Recall@10 | 1.0000 |
| MRR | 0.8009 |
| nDCG@10 | 0.8476 |
| Hit rate@5 | 0.9211 |
| MAP@10 | 0.7965 |
| Context precision@3 | 0.3246 |
| Context recall@3 | 0.9211 |
| Recall@1 | 0.5943 |
| Recall@5 | 0.8774 |
| Recall@10 | 0.9198 |
| MRR | 0.7548 |
| nDCG@10 | 0.7804 |
| Hit rate@5 | 0.9057 |
| MAP@10 | 0.7280 |
| Context precision@3 | 0.3082 |
| Context recall@3 | 0.8160 |

### By query type

Expand All @@ -37,10 +37,11 @@ The type comes from `type` in `questions.jsonl`; untagged questions report as

| Type | Questions | Recall@5 | nDCG@10 | Hit rate@5 | MAP@10 |
| --- | --- | --- | --- | --- | --- |
| cross-lingual | 9 | 0.6667 | 0.5032 | 0.6667 | 0.3446 |
| exact | 6 | 1.0000 | 1.0000 | 1.0000 | 1.0000 |
| multi-hop | 2 | 1.0000 | 0.9599 | 1.0000 | 0.9167 |
| semantic | 18 | 1.0000 | 0.9312 | 1.0000 | 0.9074 |
| cross-lingual | 9 | 0.6667 | 0.4173 | 0.6667 | 0.2994 |
| exact | 8 | 1.0000 | 0.8663 | 1.0000 | 0.8229 |
| hard-negative | 7 | 1.0000 | 0.8758 | 1.0000 | 0.8333 |
| multi-hop | 5 | 0.7000 | 0.7218 | 1.0000 | 0.6350 |
| semantic | 21 | 0.9048 | 0.8542 | 0.9048 | 0.8238 |
| zh | 3 | 1.0000 | 1.0000 | 1.0000 | 1.0000 |

### Unanswerable questions
Expand All @@ -62,13 +63,13 @@ with a small second one means the threshold filters nothing and the window is al

| Metric | Value |
| --- | --- |
| Unanswerable questions | 6 |
| Retrieval abstained | 0.0000 (0/6) |
| Mean candidates passing the threshold | 19.00 |
| Unanswerable questions | 10 |
| Retrieval abstained | 0.0000 (0/10) |
| Mean candidates passing the threshold | 20.00 |
| Mean passages in the context window | 3.00 |

Timing is informational only and is **not** frozen: indexing 1583 ms, query
p50 13.38 ms, p95 16.66 ms on the
Timing is informational only and is **not** frozen: indexing 2796 ms, query
p50 13.67 ms, p95 16.34 ms on the
machine that produced this file. Timing and index size depend on hardware and on the
corpus, so they must never be the reason two runs differ.

Expand Down
102 changes: 51 additions & 51 deletions docs/eval/retrieval-v1.6.json
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,79 +15,79 @@
"id": "dense",
"label": "dense (vector)",
"split": "validation",
"questions": 16,
"answerableCount": 13,
"questions": 24,
"answerableCount": 19,
"chunking": "1000/100",
"chunkCount": 19,
"recallAt1": 0.576923,
"recallAt5": 0.923077,
"recallAt10": 1,
"mrr": 0.778846,
"ndcgAt10": 0.827608,
"hitRateAt5": 0.923077,
"mapAt10": 0.766026,
"contextPrecision": 0.333333,
"contextRecall": 0.923077,
"latencyP95Ms": 16.02
"chunkCount": 53,
"recallAt1": 0.513158,
"recallAt5": 0.868421,
"recallAt10": 0.921053,
"mrr": 0.720175,
"ndcgAt10": 0.755611,
"hitRateAt5": 0.894737,
"mapAt10": 0.69386,
"contextPrecision": 0.315789,
"contextRecall": 0.802632,
"latencyP95Ms": 13.27
},
{
"id": "sparse",
"label": "sparse (BM25)",
"split": "validation",
"questions": 16,
"answerableCount": 13,
"questions": 24,
"answerableCount": 19,
"chunking": "1000/100",
"chunkCount": 19,
"recallAt1": 0.576923,
"recallAt5": 0.615385,
"recallAt10": 0.615385,
"mrr": 0.615385,
"ndcgAt10": 0.609209,
"hitRateAt5": 0.615385,
"mapAt10": 0.602564,
"contextPrecision": 0.230769,
"contextRecall": 0.615385,
"latencyP95Ms": 2.96
"chunkCount": 53,
"recallAt1": 0.460526,
"recallAt5": 0.671053,
"recallAt10": 0.684211,
"mrr": 0.600877,
"ndcgAt10": 0.610495,
"hitRateAt5": 0.684211,
"mapAt10": 0.576817,
"contextPrecision": 0.263158,
"contextRecall": 0.657895,
"latencyP95Ms": 2.26
},
{
"id": "hybrid",
"label": "hybrid (RRF of dense + BM25)",
"split": "validation",
"questions": 16,
"answerableCount": 13,
"questions": 24,
"answerableCount": 19,
"chunking": "1000/100",
"chunkCount": 19,
"recallAt1": 0.576923,
"recallAt5": 0.923077,
"recallAt10": 1,
"mrr": 0.778846,
"ndcgAt10": 0.827608,
"hitRateAt5": 0.923077,
"mapAt10": 0.766026,
"chunkCount": 53,
"recallAt1": 0.513158,
"recallAt5": 0.868421,
"recallAt10": 0.894737,
"mrr": 0.732456,
"ndcgAt10": 0.760265,
"hitRateAt5": 0.894737,
"mapAt10": 0.707018,
"contextPrecision": 0.333333,
"contextRecall": 0.923077,
"latencyP95Ms": 19.47
"contextRecall": 0.855263,
"latencyP95Ms": 13.61
}
],
"test": [
{
"id": "dense",
"label": "dense (vector)",
"split": "test",
"questions": 28,
"answerableCount": 25,
"questions": 39,
"answerableCount": 34,
"chunking": "1000/100",
"chunkCount": 19,
"recallAt1": 0.7,
"recallAt5": 0.92,
"recallAt10": 1,
"mrr": 0.812381,
"ndcgAt10": 0.858056,
"hitRateAt5": 0.92,
"mapAt10": 0.812381,
"contextPrecision": 0.32,
"contextRecall": 0.92,
"latencyP95Ms": 14.49
"chunkCount": 53,
"recallAt1": 0.639706,
"recallAt5": 0.882353,
"recallAt10": 0.919118,
"mrr": 0.774079,
"ndcgAt10": 0.794329,
"hitRateAt5": 0.911765,
"mapAt10": 0.747141,
"contextPrecision": 0.303922,
"contextRecall": 0.823529,
"latencyP95Ms": 12.89
}
]
}
8 changes: 4 additions & 4 deletions docs/eval/retrieval-v1.6.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,15 +16,15 @@ to.

| Strategy | Recall@1 | Recall@5 | MRR | nDCG@10 | MAP@10 | Context P | n | p95 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| dense (vector) | 0.5769 | 0.9231 | 0.7788 | 0.8276 | 0.7660 | 0.3333 | 16 | 16.02 ms |
| sparse (BM25) | 0.5769 | 0.6154 | 0.6154 | 0.6092 | 0.6026 | 0.2308 | 16 | 2.96 ms |
| hybrid (RRF of dense + BM25) | 0.5769 | 0.9231 | 0.7788 | 0.8276 | 0.7660 | 0.3333 | 16 | 19.47 ms |
| dense (vector) | 0.5132 | 0.8684 | 0.7202 | 0.7556 | 0.6939 | 0.3158 | 24 | 13.27 ms |
| sparse (BM25) | 0.4605 | 0.6711 | 0.6009 | 0.6105 | 0.5768 | 0.2632 | 24 | 2.26 ms |
| hybrid (RRF of dense + BM25) | 0.5132 | 0.8684 | 0.7325 | 0.7603 | 0.7070 | 0.3333 | 24 | 13.61 ms |

## Test — reported, not selected

| Strategy | Recall@1 | Recall@5 | MRR | nDCG@10 | MAP@10 | Context P | n | p95 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| dense (vector) | 0.7000 | 0.9200 | 0.8124 | 0.8581 | 0.8124 | 0.3200 | 28 | 14.49 ms |
| dense (vector) | 0.6397 | 0.8824 | 0.7741 | 0.7943 | 0.7471 | 0.3039 | 39 | 12.89 ms |

## Not evaluated

Expand Down
Loading
Loading