Update dataset README for reranking_hybrid candidates

README.md

@@ -1,173 +1,174 @@
 ---
 configs:
-- config_name: bm25
+- config_name: corpus
   data_files:
   - split: NanoArguAna
-    path: bm25/NanoArguAna-*
+    path: corpus/NanoArguAna.parquet
   - split: NanoClimateFEVER
-    path: bm25/NanoClimateFEVER-*
+    path: corpus/NanoClimateFEVER.parquet
   - split: NanoDBPedia
-    path: bm25/NanoDBPedia-*
+    path: corpus/NanoDBPedia.parquet
   - split: NanoFEVER
-    path: bm25/NanoFEVER-*
+    path: corpus/NanoFEVER.parquet
   - split: NanoFiQA2018
-    path: bm25/NanoFiQA2018-*
+    path: corpus/NanoFiQA2018.parquet
   - split: NanoHotpotQA
-    path: bm25/NanoHotpotQA-*
+    path: corpus/NanoHotpotQA.parquet
   - split: NanoMSMARCO
-    path: bm25/NanoMSMARCO-*
+    path: corpus/NanoMSMARCO.parquet
   - split: NanoNFCorpus
-    path: bm25/NanoNFCorpus-*
+    path: corpus/NanoNFCorpus.parquet
   - split: NanoNQ
-    path: bm25/NanoNQ-*
+    path: corpus/NanoNQ.parquet
   - split: NanoQuoraRetrieval
-    path: bm25/NanoQuoraRetrieval-*
+    path: corpus/NanoQuoraRetrieval.parquet
   - split: NanoSCIDOCS
-    path: bm25/NanoSCIDOCS-*
+    path: corpus/NanoSCIDOCS.parquet
   - split: NanoSciFact
-    path: bm25/NanoSciFact-*
+    path: corpus/NanoSciFact.parquet
   - split: NanoTouche2020
-    path: bm25/NanoTouche2020-*
-- config_name: corpus
+    path: corpus/NanoTouche2020.parquet
+- config_name: queries
   data_files:
   - split: NanoArguAna
-    path: corpus/NanoArguAna-*
+    path: queries/NanoArguAna.parquet
   - split: NanoClimateFEVER
-    path: corpus/NanoClimateFEVER-*
+    path: queries/NanoClimateFEVER.parquet
   - split: NanoDBPedia
-    path: corpus/NanoDBPedia-*
+    path: queries/NanoDBPedia.parquet
   - split: NanoFEVER
-    path: corpus/NanoFEVER-*
+    path: queries/NanoFEVER.parquet
   - split: NanoFiQA2018
-    path: corpus/NanoFiQA2018-*
+    path: queries/NanoFiQA2018.parquet
   - split: NanoHotpotQA
-    path: corpus/NanoHotpotQA-*
+    path: queries/NanoHotpotQA.parquet
   - split: NanoMSMARCO
-    path: corpus/NanoMSMARCO-*
+    path: queries/NanoMSMARCO.parquet
   - split: NanoNFCorpus
-    path: corpus/NanoNFCorpus-*
+    path: queries/NanoNFCorpus.parquet
   - split: NanoNQ
-    path: corpus/NanoNQ-*
+    path: queries/NanoNQ.parquet
   - split: NanoQuoraRetrieval
-    path: corpus/NanoQuoraRetrieval-*
+    path: queries/NanoQuoraRetrieval.parquet
   - split: NanoSCIDOCS
-    path: corpus/NanoSCIDOCS-*
+    path: queries/NanoSCIDOCS.parquet
   - split: NanoSciFact
-    path: corpus/NanoSciFact-*
+    path: queries/NanoSciFact.parquet
   - split: NanoTouche2020
-    path: corpus/NanoTouche2020-*
-- config_name: harrier_oss_v1_270m
+    path: queries/NanoTouche2020.parquet
+  default: true
+- config_name: qrels
   data_files:
   - split: NanoArguAna
-    path: harrier_oss_v1_270m/NanoArguAna-*
+    path: qrels/NanoArguAna.parquet
   - split: NanoClimateFEVER
-    path: harrier_oss_v1_270m/NanoClimateFEVER-*
+    path: qrels/NanoClimateFEVER.parquet
   - split: NanoDBPedia
-    path: harrier_oss_v1_270m/NanoDBPedia-*
+    path: qrels/NanoDBPedia.parquet
   - split: NanoFEVER
-    path: harrier_oss_v1_270m/NanoFEVER-*
+    path: qrels/NanoFEVER.parquet
   - split: NanoFiQA2018
-    path: harrier_oss_v1_270m/NanoFiQA2018-*
+    path: qrels/NanoFiQA2018.parquet
   - split: NanoHotpotQA
-    path: harrier_oss_v1_270m/NanoHotpotQA-*
+    path: qrels/NanoHotpotQA.parquet
   - split: NanoMSMARCO
-    path: harrier_oss_v1_270m/NanoMSMARCO-*
+    path: qrels/NanoMSMARCO.parquet
   - split: NanoNFCorpus
-    path: harrier_oss_v1_270m/NanoNFCorpus-*
+    path: qrels/NanoNFCorpus.parquet
   - split: NanoNQ
-    path: harrier_oss_v1_270m/NanoNQ-*
+    path: qrels/NanoNQ.parquet
   - split: NanoQuoraRetrieval
-    path: harrier_oss_v1_270m/NanoQuoraRetrieval-*
+    path: qrels/NanoQuoraRetrieval.parquet
   - split: NanoSCIDOCS
-    path: harrier_oss_v1_270m/NanoSCIDOCS-*
+    path: qrels/NanoSCIDOCS.parquet
   - split: NanoSciFact
-    path: harrier_oss_v1_270m/NanoSciFact-*
+    path: qrels/NanoSciFact.parquet
   - split: NanoTouche2020
-    path: harrier_oss_v1_270m/NanoTouche2020-*
-- config_name: qrels
+    path: qrels/NanoTouche2020.parquet
+- config_name: bm25
   data_files:
   - split: NanoArguAna
-    path: qrels/NanoArguAna-*
+    path: bm25/NanoArguAna.parquet
   - split: NanoClimateFEVER
-    path: qrels/NanoClimateFEVER-*
+    path: bm25/NanoClimateFEVER.parquet
   - split: NanoDBPedia
-    path: qrels/NanoDBPedia-*
+    path: bm25/NanoDBPedia.parquet
   - split: NanoFEVER
-    path: qrels/NanoFEVER-*
+    path: bm25/NanoFEVER.parquet
   - split: NanoFiQA2018
-    path: qrels/NanoFiQA2018-*
+    path: bm25/NanoFiQA2018.parquet
   - split: NanoHotpotQA
-    path: qrels/NanoHotpotQA-*
+    path: bm25/NanoHotpotQA.parquet
   - split: NanoMSMARCO
-    path: qrels/NanoMSMARCO-*
+    path: bm25/NanoMSMARCO.parquet
   - split: NanoNFCorpus
-    path: qrels/NanoNFCorpus-*
+    path: bm25/NanoNFCorpus.parquet
   - split: NanoNQ
-    path: qrels/NanoNQ-*
+    path: bm25/NanoNQ.parquet
   - split: NanoQuoraRetrieval
-    path: qrels/NanoQuoraRetrieval-*
+    path: bm25/NanoQuoraRetrieval.parquet
   - split: NanoSCIDOCS
-    path: qrels/NanoSCIDOCS-*
+    path: bm25/NanoSCIDOCS.parquet
   - split: NanoSciFact
-    path: qrels/NanoSciFact-*
+    path: bm25/NanoSciFact.parquet
   - split: NanoTouche2020
-    path: qrels/NanoTouche2020-*
-- config_name: queries
+    path: bm25/NanoTouche2020.parquet
+- config_name: harrier_oss_v1_270m
   data_files:
   - split: NanoArguAna
-    path: queries/NanoArguAna.parquet
+    path: harrier_oss_v1_270m/NanoArguAna.parquet
   - split: NanoClimateFEVER
-    path: queries/NanoClimateFEVER.parquet
+    path: harrier_oss_v1_270m/NanoClimateFEVER.parquet
   - split: NanoDBPedia
-    path: queries/NanoDBPedia.parquet
+    path: harrier_oss_v1_270m/NanoDBPedia.parquet
   - split: NanoFEVER
-    path: queries/NanoFEVER.parquet
+    path: harrier_oss_v1_270m/NanoFEVER.parquet
   - split: NanoFiQA2018
-    path: queries/NanoFiQA2018.parquet
+    path: harrier_oss_v1_270m/NanoFiQA2018.parquet
   - split: NanoHotpotQA
-    path: queries/NanoHotpotQA.parquet
+    path: harrier_oss_v1_270m/NanoHotpotQA.parquet
   - split: NanoMSMARCO
-    path: queries/NanoMSMARCO.parquet
+    path: harrier_oss_v1_270m/NanoMSMARCO.parquet
   - split: NanoNFCorpus
-    path: queries/NanoNFCorpus.parquet
+    path: harrier_oss_v1_270m/NanoNFCorpus.parquet
   - split: NanoNQ
-    path: queries/NanoNQ.parquet
+    path: harrier_oss_v1_270m/NanoNQ.parquet
   - split: NanoQuoraRetrieval
-    path: queries/NanoQuoraRetrieval.parquet
+    path: harrier_oss_v1_270m/NanoQuoraRetrieval.parquet
   - split: NanoSCIDOCS
-    path: queries/NanoSCIDOCS.parquet
+    path: harrier_oss_v1_270m/NanoSCIDOCS.parquet
   - split: NanoSciFact
-    path: queries/NanoSciFact.parquet
+    path: harrier_oss_v1_270m/NanoSciFact.parquet
   - split: NanoTouche2020
-    path: queries/NanoTouche2020.parquet
+    path: harrier_oss_v1_270m/NanoTouche2020.parquet
 - config_name: reranking_hybrid
   data_files:
   - split: NanoArguAna
-    path: reranking_hybrid/NanoArguAna-*
+    path: reranking_hybrid/NanoArguAna.parquet
   - split: NanoClimateFEVER
-    path: reranking_hybrid/NanoClimateFEVER-*
+    path: reranking_hybrid/NanoClimateFEVER.parquet
   - split: NanoDBPedia
-    path: reranking_hybrid/NanoDBPedia-*
+    path: reranking_hybrid/NanoDBPedia.parquet
   - split: NanoFEVER
-    path: reranking_hybrid/NanoFEVER-*
+    path: reranking_hybrid/NanoFEVER.parquet
   - split: NanoFiQA2018
-    path: reranking_hybrid/NanoFiQA2018-*
+    path: reranking_hybrid/NanoFiQA2018.parquet
   - split: NanoHotpotQA
-    path: reranking_hybrid/NanoHotpotQA-*
+    path: reranking_hybrid/NanoHotpotQA.parquet
   - split: NanoMSMARCO
-    path: reranking_hybrid/NanoMSMARCO-*
+    path: reranking_hybrid/NanoMSMARCO.parquet
   - split: NanoNFCorpus
-    path: reranking_hybrid/NanoNFCorpus-*
+    path: reranking_hybrid/NanoNFCorpus.parquet
   - split: NanoNQ
-    path: reranking_hybrid/NanoNQ-*
+    path: reranking_hybrid/NanoNQ.parquet
   - split: NanoQuoraRetrieval
-    path: reranking_hybrid/NanoQuoraRetrieval-*
+    path: reranking_hybrid/NanoQuoraRetrieval.parquet
   - split: NanoSCIDOCS
-    path: reranking_hybrid/NanoSCIDOCS-*
+    path: reranking_hybrid/NanoSCIDOCS.parquet
   - split: NanoSciFact
-    path: reranking_hybrid/NanoSciFact-*
+    path: reranking_hybrid/NanoSciFact.parquet
   - split: NanoTouche2020
-    path: reranking_hybrid/NanoTouche2020-*
+    path: reranking_hybrid/NanoTouche2020.parquet
 language:
 - ko
 tags:
@@ -176,6 +177,8 @@ tags:
 - nano
 - bm25
 - hakari-bench
+- dense-retrieval
+- reranking
 dataset_info:
 - config_name: bm25
   features:
@@ -466,173 +469,99 @@ dataset_info:
   download_size: 1476357
   dataset_size: 1458705
 ---
-
 # NanoBEIR-ko
 
-This dataset is a Nano-style retrieval dataset. Nano-series evaluation can
-be run easily with [HAKARI-Bench](https://github.com/hotchpotch/hakari-bench).
-
-NanoBEIR-ko is derived from MNanoBEIR / NanoBEIR. It follows the
-Hugging Face Datasets layout convention used by
-[sentence-transformers/NanoBEIR-en](https://huggingface.co/datasets/sentence-transformers/NanoBEIR-en):
-each Nano split has separate `corpus`, `queries`, and `qrels` tables, and BM25
-candidates are provided separately in a `bm25` table. This layout follows
-the NanoBEIR-style evaluation approach summarized in
-[NanoBEIR](https://huggingface.co/blog/sionic-ai/eval-sionic-nano-beir).
+This dataset is a Nano-style retrieval dataset for [HAKARI-bench](https://github.com/hakari-bench/hakari-bench).
 
 NanoBEIR-ko is the Korean language-specific component of MNanoBEIR. It groups compact BEIR-derived retrieval tasks for efficient evaluation of document ranking in that language.
 
+## Usage
 
+```python
+from datasets import load_dataset
 
-## Source Links
+dataset_id = "hakari-bench/NanoBEIR-ko"
+split = "NanoArguAna"
 
-- Final dataset: [hakari-bench/NanoBEIR-ko](https://huggingface.co/datasets/hakari-bench/NanoBEIR-ko)
+queries = load_dataset(dataset_id, "queries", split=split)
+corpus = load_dataset(dataset_id, "corpus", split=split)
+qrels = load_dataset(dataset_id, "qrels", split=split)
+reranking_candidates = load_dataset(dataset_id, "reranking_hybrid", split=split)
+```
 
 ## Data Layout
 
-This dataset uses four Hugging Face Datasets configs:
+This dataset uses six Hugging Face Datasets configs:
 
 - `corpus`: documents with `_id` and `text`
 - `queries`: queries with `_id` and `text`
 - `qrels`: positive relevance labels with `query-id` and `corpus-id`
 - `bm25`: BM25 candidate lists with `query-id` and `corpus-ids`
+- `harrier_oss_v1_270m`: dense candidate lists from `microsoft/harrier-oss-v1-270m`
+- `reranking_hybrid`: RRF candidate lists built from `bm25` and `harrier_oss_v1_270m`
 
 Each config has the same Nano split names.
 
-The `qrels` config is positive-only. Source rows with `score <= 0` are treated
-as non-relevant or hard-negative annotations and are not included in `qrels`.
-When the source provides such rows, their documents are preferentially used as
-hard negatives in the `corpus` config before generic corpus-fill documents.
-Source hard negatives are sampled deterministically with query round-robin so
-one query's negative pool does not dominate the corpus.
-
-## Split Statistics
-
-Length statistics are computed with `len(str(text))` over the `queries` and
-`corpus` tables. `std` is the population standard deviation over the rows in
-each split.
-
-| Nano split | Queries | Corpus | Qrels | Query avg | Query std | Query median | Query p25 | Query p75 | Doc avg | Doc std | Doc median | Doc p25 | Doc p75 |
-|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|
-| NanoArguAna | 50 | 3635 | 50 | 619.4 | 212.6 | 604.0 | 471.5 | 768.2 | 519.6 | 290.0 | 464.0 | 317.0 | 659.5 |
-| NanoClimateFEVER | 50 | 3408 | 148 | 66.0 | 26.4 | 64.5 | 44.0 | 82.8 | 779.7 | 427.6 | 704.0 | 473.0 | 1005.0 |
-| NanoDBPedia | 50 | 6045 | 1158 | 16.8 | 7.7 | 15.5 | 11.0 | 21.8 | 187.6 | 89.4 | 198.0 | 130.0 | 244.0 |
-| NanoFEVER | 50 | 4996 | 57 | 26.4 | 9.2 | 25.0 | 21.0 | 29.0 | 648.1 | 441.0 | 551.5 | 327.8 | 870.0 |
-| NanoFiQA2018 | 50 | 4598 | 123 | 29.6 | 11.9 | 29.5 | 20.5 | 36.8 | 490.3 | 440.7 | 361.0 | 195.2 | 622.8 |
-| NanoHotpotQA | 50 | 5090 | 100 | 49.5 | 18.5 | 46.0 | 34.5 | 59.8 | 197.1 | 142.0 | 170.0 | 88.0 | 269.0 |
-| NanoMSMARCO | 50 | 5043 | 50 | 19.1 | 11.7 | 17.0 | 11.0 | 23.8 | 169.2 | 71.0 | 152.0 | 122.0 | 202.5 |
-| NanoNFCorpus | 50 | 2953 | 1651 | 10.8 | 7.5 | 10.0 | 5.0 | 14.8 | 752.7 | 268.7 | 758.0 | 591.0 | 895.0 |
-| NanoNQ | 50 | 5035 | 57 | 29.3 | 13.7 | 25.0 | 21.0 | 31.8 | 274.2 | 227.0 | 231.0 | 93.0 | 394.5 |
-| NanoQuoraRetrieval | 50 | 5046 | 70 | 28.7 | 11.5 | 28.0 | 21.2 | 32.0 | 32.8 | 29.6 | 28.0 | 22.0 | 37.0 |
-| NanoSCIDOCS | 50 | 2210 | 244 | 32.1 | 9.0 | 30.0 | 25.0 | 39.0 | 452.8 | 339.0 | 444.0 | 281.0 | 609.8 |
-| NanoSciFact | 50 | 2919 | 56 | 46.3 | 18.7 | 42.0 | 32.0 | 59.2 | 723.6 | 305.0 | 679.0 | 532.0 | 869.0 |
-| NanoTouche2020 | 49 | 5745 | 932 | 21.7 | 6.8 | 20.0 | 17.0 | 26.0 | 1032.8 | 1085.4 | 543.0 | 176.0 | 1629.0 |
-
-## Construction Steps
-
-This dataset is constructed as follows.
-
-1. Use MNanoBEIR / NanoBEIR as the upstream benchmark or dataset family.
-2. Load source datasets from the `hakari-bench/NanoBEIR-ko` corpus, queries, and qrels tables.
-3. Source evaluation split policy: the NanoBEIR split set.
-4. Create one Nano split for each selected source retrieval task.
-5. Keep up to 200 eligible queries per Nano split.
-6. Treat source relevance rows with `score > 0` as qrels-positive documents.
-   If the source has no score column, treat its qrels as positive-only only when
-   that is the source task convention.
-7. Exclude source rows with `score <= 0` from `qrels`. When such rows are
-   available for selected queries, use their documents as hard-negative corpus
-   candidates before generic fill documents.
-8. Include all qrels-positive documents for the selected queries.
-9. Use the included corpus tables for each Nano split; no additional document resampling is performed.
-10. Remove exact duplicate query text and document text within each split. If a
-   removed document duplicate was referenced by qrels,
-   the qrels row was removed.
-11. Store corpus text as `title` plus body text when available.
-12. Generate BM25 top-100 candidates with
-    `wordseg:ko` tokenization.
-13. If a qrels-positive document is missing from the raw BM25 result, insert it
-    into the final `bm25` candidate list by replacing a tail non-positive
-    candidate.
+## Candidate Construction
 
-The `qrels` config is positive-only. When needed for top-k reranking coverage, positive qrels are capped per query to the BM25 top-k.
+- `bm25`: local BM25 top-500 with automatic language-aware tokenization. The resolved tokenizer is shown in the Candidate Quality table, for example `wordseg@ja`.
+- `harrier_oss_v1_270m`: dense top-500 from `microsoft/harrier-oss-v1-270m`. In tables this is shown as `Dense`; Dense means `microsoft/harrier-oss-v1-270m` with the `web_search_query` prompt for queries and cosine similarity over normalized embeddings.
+- `reranking_hybrid`: RRF over `bm25` and `harrier_oss_v1_270m` using `rrf_k=100`, keeping the RRF top-100.
 
-The `bm25` candidate subset is generated from the included corpus for each split.
+Safeguard means rank 101 is appended only when RRF top-100 contains no qrels-positive document.
 
-For top-100 reranking diagnostics, positive qrels are capped to at most 100 documents per query before BM25 positive forcing. This makes full relevant coverage possible for splits whose upstream qrels contain more than 100 positives for a single query.
-
-## BM25 Subset Policy
-
-The `bm25` config is a candidate subset for first-stage retrieval and reranking.
-It is not a separate source dataset. Each row contains one query id and a ranked
-list of up to 100 corpus ids.
-
-BM25 candidates are generated from the selected corpus for each split. When a
-qrels-positive document is not present in the raw BM25 top-100
-results, the missing positive is forced into the final candidate list by
-replacing a tail candidate that is not positive for that query. Candidate ids
-are kept unique after replacement.
-
-Concretely, each `bm25` row is produced by tokenizing the selected split corpus
-and query texts with `wordseg:ko`, ranking the corpus with BM25,
-then writing the ranked corpus ids as `corpus-ids` for that query. The list is a
-candidate subset for downstream evaluation, not a full-corpus ranking.
-
-Source hard negatives, including documents referenced by source rows with
-`score <= 0`, may appear in the selected corpus and can naturally appear in BM25
-candidates. They are still non-relevant and are not listed in `qrels`.
-
-When source hard negatives are available, the default corpus sampling policy is
-query round-robin: group hard negatives by selected query, preserve source
-rank/order within each query, add at most one new hard negative from each query
-per pass, remove duplicate IDs and exact duplicate text, then fill any remaining
-slots from source corpus order.
-
-## Split Mapping
-
-Each Nano split maps to one source retrieval task unless noted otherwise.
-
-| Nano split | Source task | Source dataset | Queries | Corpus | Qrels |
-|---|---|---|---:|---:|---:|
-| NanoArguAna | NanoArguAna | [hakari-bench/NanoBEIR-ko](https://huggingface.co/datasets/hakari-bench/NanoBEIR-ko) | 50 | 3635 | 50 |
-| NanoClimateFEVER | NanoClimateFEVER | [hakari-bench/NanoBEIR-ko](https://huggingface.co/datasets/hakari-bench/NanoBEIR-ko) | 50 | 3408 | 148 |
-| NanoDBPedia | NanoDBPedia | [hakari-bench/NanoBEIR-ko](https://huggingface.co/datasets/hakari-bench/NanoBEIR-ko) | 50 | 6045 | 1158 |
-| NanoFEVER | NanoFEVER | [hakari-bench/NanoBEIR-ko](https://huggingface.co/datasets/hakari-bench/NanoBEIR-ko) | 50 | 4996 | 57 |
-| NanoFiQA2018 | NanoFiQA2018 | [hakari-bench/NanoBEIR-ko](https://huggingface.co/datasets/hakari-bench/NanoBEIR-ko) | 50 | 4598 | 123 |
-| NanoHotpotQA | NanoHotpotQA | [hakari-bench/NanoBEIR-ko](https://huggingface.co/datasets/hakari-bench/NanoBEIR-ko) | 50 | 5090 | 100 |
-| NanoMSMARCO | NanoMSMARCO | [hakari-bench/NanoBEIR-ko](https://huggingface.co/datasets/hakari-bench/NanoBEIR-ko) | 50 | 5043 | 50 |
-| NanoNFCorpus | NanoNFCorpus | [hakari-bench/NanoBEIR-ko](https://huggingface.co/datasets/hakari-bench/NanoBEIR-ko) | 50 | 2953 | 1651 |
-| NanoNQ | NanoNQ | [hakari-bench/NanoBEIR-ko](https://huggingface.co/datasets/hakari-bench/NanoBEIR-ko) | 50 | 5035 | 57 |
-| NanoQuoraRetrieval | NanoQuoraRetrieval | [hakari-bench/NanoBEIR-ko](https://huggingface.co/datasets/hakari-bench/NanoBEIR-ko) | 50 | 5046 | 70 |
-| NanoSCIDOCS | NanoSCIDOCS | [hakari-bench/NanoBEIR-ko](https://huggingface.co/datasets/hakari-bench/NanoBEIR-ko) | 50 | 2210 | 244 |
-| NanoSciFact | NanoSciFact | [hakari-bench/NanoBEIR-ko](https://huggingface.co/datasets/hakari-bench/NanoBEIR-ko) | 50 | 2919 | 56 |
-| NanoTouche2020 | NanoTouche2020 | [hakari-bench/NanoBEIR-ko](https://huggingface.co/datasets/hakari-bench/NanoBEIR-ko) | 49 | 5745 | 932 |
-
-## BM25 nDCG@10
-
-`nDCG@10` is computed from the included BM25 ranking against the included qrels.
-
-Coverage is measured against included qrels at the same BM25 top-k used for reranking diagnostics. The included BM25 candidate subset has 100.00% query coverage and 100.00% relevant coverage for every split.
+## Split Statistics
 
-| Nano split | Tokenizer | Forced BM25 positives | Query cov | Relevant cov | BM25 nDCG@10 |
-|---|---|---:|---:|---:|---:|
-| NanoArguAna | wordseg:ko | 5 | 100.00% | 100.00% | 0.3666 |
-| NanoClimateFEVER | wordseg:ko | 60 | 100.00% | 100.00% | 0.2457 |
-| NanoDBPedia | wordseg:ko | 403 | 100.00% | 100.00% | 0.5322 |
-| NanoFEVER | wordseg:ko | 5 | 100.00% | 100.00% | 0.5723 |
-| NanoFiQA2018 | wordseg:ko | 54 | 100.00% | 100.00% | 0.3433 |
-| NanoHotpotQA | wordseg:ko | 13 | 100.00% | 100.00% | 0.5966 |
-| NanoMSMARCO | wordseg:ko | 6 | 100.00% | 100.00% | 0.3320 |
-| NanoNFCorpus | wordseg:ko | 1420 | 100.00% | 100.00% | 0.3112 |
-| NanoNQ | wordseg:ko | 12 | 100.00% | 100.00% | 0.4301 |
-| NanoQuoraRetrieval | wordseg:ko | 2 | 100.00% | 100.00% | 0.7099 |
-| NanoSCIDOCS | wordseg:ko | 96 | 100.00% | 100.00% | 0.2688 |
-| NanoSciFact | wordseg:ko | 4 | 100.00% | 100.00% | 0.6835 |
-| NanoTouche2020 | wordseg:ko | 245 | 100.00% | 100.00% | 0.5034 |
+Length statistics are character counts computed with `len(str(text))`.
+
+| Nano split | Queries | Corpus | Qrels | Query chars avg | Query chars p50 | Query chars p75 | Doc chars avg | Doc chars p50 | Doc chars p75 |
+|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|
+| NanoArguAna | 50 | 3635 | 50 | 619.4 | 604.0 | 768.2 | 519.6 | 464.0 | 659.5 |
+| NanoClimateFEVER | 50 | 3408 | 148 | 66.0 | 64.5 | 82.8 | 779.7 | 704.0 | 1005.0 |
+| NanoDBPedia | 50 | 6045 | 1158 | 16.8 | 15.5 | 21.8 | 187.6 | 198.0 | 244.0 |
+| NanoFEVER | 50 | 4996 | 57 | 26.4 | 25.0 | 29.0 | 648.1 | 551.5 | 870.0 |
+| NanoFiQA2018 | 50 | 4598 | 123 | 29.6 | 29.5 | 36.8 | 490.3 | 361.0 | 622.8 |
+| NanoHotpotQA | 50 | 5090 | 100 | 49.5 | 46.0 | 59.8 | 197.1 | 170.0 | 269.0 |
+| NanoMSMARCO | 50 | 5043 | 50 | 19.1 | 17.0 | 23.8 | 169.2 | 152.0 | 202.5 |
+| NanoNFCorpus | 50 | 2953 | 1651 | 10.8 | 10.0 | 14.8 | 752.7 | 758.0 | 895.0 |
+| NanoNQ | 50 | 5035 | 57 | 29.3 | 25.0 | 31.8 | 274.2 | 231.0 | 394.5 |
+| NanoQuoraRetrieval | 50 | 5046 | 70 | 28.7 | 28.0 | 32.0 | 32.8 | 28.0 | 37.0 |
+| NanoSCIDOCS | 50 | 2210 | 244 | 32.1 | 30.0 | 39.0 | 452.8 | 444.0 | 609.8 |
+| NanoSciFact | 50 | 2919 | 56 | 46.3 | 42.0 | 59.2 | 723.6 | 679.0 | 869.0 |
+| NanoTouche2020 | 49 | 5745 | 932 | 21.7 | 20.0 | 26.0 | 1032.8 | 543.0 | 1629.0 |
+
+## Candidate Quality
+
+`nDCG@10` and `Recall@100` are computed from the included candidate rankings against the included qrels, then reported as 0-100 scores such as `52.45`. `Recall@100` uses only the top 100 candidates; an optional rank-101 safeguard positive is not counted in `Recall@100`.
+
+Dense means `microsoft/harrier-oss-v1-270m` with the `web_search_query` prompt and cosine similarity.
+
+| Nano split | BM25 tokenizer | BM25 nDCG@10 | Dense nDCG@10 | Hybrid nDCG@10 | BM25 Recall@100 | Dense Recall@100 | Hybrid Recall@100 | Hybrid candidates | Safeguard positives |
+|---|---|---:|---:|---:|---:|---:|---:|---:|---:|
+| Mean | - | 44.80 | 49.88 | 50.23 | 74.94 | 78.51 | 81.26 | - | 31 |
+| NanoArguAna | wordseg@ko | 36.61 | 40.82 | 42.17 | 90.00 | 94.00 | 96.00 | 100-101 | 2 |
+| NanoClimateFEVER | wordseg@ko | 24.57 | 30.03 | 29.83 | 63.87 | 68.30 | 66.10 | 100-101 | 3 |
+| NanoDBPedia | wordseg@ko | 53.22 | 59.28 | 57.87 | 72.53 | 76.53 | 79.55 | 100 | 0 |
+| NanoFEVER | wordseg@ko | 57.23 | 73.35 | 70.01 | 92.00 | 98.33 | 99.00 | 100 | 0 |
+| NanoFiQA2018 | wordseg@ko | 34.15 | 37.13 | 42.91 | 60.57 | 74.03 | 73.29 | 100-101 | 7 |
+| NanoHotpotQA | wordseg@ko | 59.66 | 62.69 | 63.16 | 87.00 | 84.00 | 93.00 | 100-101 | 2 |
+| NanoMSMARCO | wordseg@ko | 33.20 | 41.64 | 43.71 | 88.00 | 96.00 | 96.00 | 100-101 | 2 |
+| NanoNFCorpus | wordseg@ko | 24.66 | 23.32 | 24.40 | 17.26 | 20.15 | 22.32 | 100-101 | 9 |
+| NanoNQ | wordseg@ko | 43.01 | 58.05 | 50.33 | 78.00 | 93.00 | 99.00 | 100 | 0 |
+| NanoQuoraRetrieval | wordseg@ko | 70.62 | 81.33 | 76.32 | 97.33 | 96.00 | 100.00 | 100 | 0 |
+| NanoSCIDOCS | wordseg@ko | 26.73 | 33.10 | 33.80 | 60.93 | 64.17 | 64.27 | 100-101 | 1 |
+| NanoSciFact | wordseg@ko | 68.35 | 62.07 | 68.38 | 92.00 | 84.00 | 90.00 | 100-101 | 5 |
+| NanoTouche2020 | wordseg@ko | 50.33 | 45.64 | 50.13 | 74.71 | 72.08 | 77.89 | 100 | 0 |
+
+## Hybrid Safeguard Summary
+
+- Safeguard positives: 31
+- Rows limited by corpus size: 0
+- Metadata file: `reranking_hybrid_metadata.json`
 
-## Skipped Tasks
+## Source Links
 
-No source tasks were skipped.
+- Final dataset: [hakari-bench/NanoBEIR-ko](https://huggingface.co/datasets/hakari-bench/NanoBEIR-ko)
 
 ## License