v0.2.0 — Tier 3 top-20 union (230,964 judged pairs)

Extends v0.1.0 (top-3, 36,418 pairs) to the full top-20 union of all 6 retrievers. Same judge (Qwen3.5-397B-A17B-FP8, v2 graded rubric). Top-3 pairs from v0.1.0 are a strict subset, recoverable via rankings.rank<=3.

README.md

@@ -18,7 +18,7 @@ tags:
   - llm-as-judge
   - rag
 size_categories:
-  - 1K<n<10K
+  - 100K<n<1M
 configs:
   - config_name: queries
     data_files:
@@ -42,15 +42,17 @@ configs:
         path: "audit/judgments_with_reasoning.parquet"
 ---
 
-# MamaRetrieval — v0.1.0
+# MamaRetrieval — v0.2.0
 
 A retrieval evaluation benchmark for medical RAG systems serving midwives and
 doctors. 3,185 clinical queries on midwifery / OBGYN topics, evaluated against
-the top-3 results of 6 retrievers, with per `(query, chunk)` pair labels graded
+the top-20 results of 6 retrievers, with per `(query, chunk)` pair labels graded
 by an LLM judge under a four-dimension rubric.
 
-This release is the **Tier 2** split (top-3 union of 6 retrievers, 36,418
-labelled `(q, c)` pairs). The Tier 3 split (top-20 union) will land as v0.2.0.
+This release is the **Tier 3** split (top-20 union of 6 retrievers, **230,964**
+labelled `(q, c)` pairs). It extends the v0.1.0 Tier 2 release (top-3 union,
+36,418 pairs) — Tier 2's pairs are a strict subset of v0.2.0's, recoverable from
+`rankings` with `rank <= 3`.
 
 ## Quick start
 
@@ -62,7 +64,7 @@ rankings  = load_dataset("nmrenyi/mamaretrieval", "rankings",  split="test")
 judgments = load_dataset("nmrenyi/mamaretrieval", "judgments", split="test")
 chunks    = load_dataset("nmrenyi/mamaretrieval", "chunks",    split="test")
 
-# Optional — the same judgments + the judge's per-row reasoning trace (~117 MB)
+# Optional — the same judgments + the judge's per-row reasoning trace (~770 MB)
 judgments_full = load_dataset("nmrenyi/mamaretrieval",
                               "judgments_with_reasoning", split="test")
 ```
@@ -72,22 +74,22 @@ judgments_full = load_dataset("nmrenyi/mamaretrieval",
 | Config | Rows | Columns | What it is |
 |---|---:|---|---|
 | `queries` | 3,185 | `query_id`, `query_text`, `seed_chunk_id` | The benchmark queries, each generated by an LLM from a single chunk of the corpus. |
-| `rankings` | 57,330 | `query_id`, `retriever`, `rank`, `chunk_id`, `score` | For every query × retriever combination, the top-3 `chunk_id`s with the retriever's similarity score. 6 retrievers × 3,185 queries × 3 = 57,330. |
-| `judgments` | 36,418 | `query_id`, `chunk_id`, `d1_topic`, `d2_meaningful`, `d3_actionable`, `d4_density`, `score` | One label per unique `(query, chunk)` pair in the pooled top-3 union. `score = d1 × (d2 + d3 + d4) ∈ [0..6]`. |
-| `judgments_with_reasoning` | 36,418 | (same as `judgments`) + `thinking` | The same labels with the judge model's reasoning trace per row. Ships in `audit/` because it's ~117 MB and not needed to use the benchmark. |
-| `chunks` | 17,827 | `chunk_id`, `text` | The chunk text for every `chunk_id` referenced by `queries.seed_chunk_id` or any retriever's top-3 result. Drawn from the producer corpus (see Provenance). |
+| `rankings` | 382,200 | `query_id`, `retriever`, `rank`, `chunk_id`, `score` | For every query × retriever combination, the top-20 `chunk_id`s with the retriever's similarity score. 6 retrievers × 3,185 queries × 20 = 382,200. |
+| `judgments` | 230,964 | `query_id`, `chunk_id`, `d1_topic`, `d2_meaningful`, `d3_actionable`, `d4_density`, `score` | One label per unique `(query, chunk)` pair in the pooled top-20 union. `score = d1 × (d2 + d3 + d4) ∈ [0..6]`. |
+| `judgments_with_reasoning` | 230,964 | (same as `judgments`) + `thinking` | The same labels with the judge model's reasoning trace per row. Ships in `audit/` because it's ~770 MB and not needed to use the benchmark. |
+| `chunks` | 41,298 | `chunk_id`, `text` | The chunk text for every `chunk_id` referenced by `queries.seed_chunk_id` or any retriever's top-20 result. Drawn from the producer corpus (see Provenance). |
 
 ### Schema notes
 
 - **`chunk_id`** is the 16-character hexadecimal identifier from the
-  producer corpus. Every `chunk_id` that appears in `rankings`, `judgments`,
-  `judgments_with_reasoning`, or `queries.seed_chunk_id` is guaranteed to be
-  resolvable in `chunks`.
+  producer corpus. Every `chunk_id` that appears in `rankings`,
+  `judgments`, `judgments_with_reasoning`, or `queries.seed_chunk_id`
+  is guaranteed to be resolvable in `chunks`.
 - **`score`** in `judgments` is computed downstream from the four dimensions
   via `score = d1 × (d2 + d3 + d4)`. The judge emits only `d1..d4`.
 - **`seed_chunk_id`** records which chunk an LLM was given when it
   synthesised the query. It's provenance, **not** a gold label — seed
-  chunks may not appear in any retriever's top-3, and when they do they
+  chunks may not appear in any retriever's top-20, and when they do they
   are not always the highest-rated chunk for that query.
 
 ## Rubric
@@ -124,8 +126,8 @@ calibration — is shipped verbatim at
 | `gecko` | `gecko-1024-quant-v0.2.0` (on-device TFLite, deployed retriever) |
 
 All retrievers were run on the producer corpus (see Provenance) and their
-top-20 results stored. This release exposes the **top-3** of each — the
-deployment-honest depth for the RAG system this benchmark was built for.
+top-20 results are exposed here. v0.1.0 exposed only the top-3; v0.2.0 ships
+the full top-20 of each, with judgments covering every chunk in the union pool.
 
 ## How the dataset was made
 
@@ -137,8 +139,8 @@ deployment-honest depth for the RAG system this benchmark was built for.
    [`audit/query_generation_prompt.txt`](audit/query_generation_prompt.txt).
 2. **Retrieval.** Each query was run against the producer corpus by every
    retriever. Top-20 candidates per retriever were stored.
-3. **Pooling.** For each query, the union of every retriever's top-3 was
-   deduped (~11.4 unique chunks per query at this scale).
+3. **Pooling.** For each query, the union of every retriever's top-20 was
+   deduped (~72 unique chunks per query on average at this scale).
 4. **Judging.** Every `(query, chunk)` pair in the pool was scored by
    `Qwen/Qwen3.5-397B-A17B-FP8` against the four-dimension rubric. The
    judge's reasoning was captured separately and is shipped in
@@ -156,8 +158,10 @@ reference labels on a 62-pair pilot, with 95% threshold agreement at score ≥
   The 63,650-chunk corpus the retrievers were run against. Built from a mix
   of WHO guidelines, Tanzania / Zanzibar MOH documents, and a small set of
   midwifery references.
-- **Versioning**: `v0.1.0` = Tier 2 (top-3 union). `v0.2.0` will add Tier 3
-  (top-20 union) on the same query set when judging finishes.
+- **Versioning**: `v0.1.0` = Tier 2 (top-3 union, 36,418 pairs). `v0.2.0`
+  (this release) = Tier 3 (top-20 union, 230,964 pairs). Tier-2 pairs are
+  a strict subset of Tier-3 pairs; reproducing v0.1.0 from v0.2.0 amounts
+  to filtering `rankings` by `rank <= 3` and inner-joining `judgments`.
 - **Audit trail**: [`manifest.json`](manifest.json) pins exact judge and
   generator model IDs, prompt hashes, and schema versions.
 
@@ -195,16 +199,18 @@ warranty disclaimers — are in [LICENSE](LICENSE).
 
 ## Citation
 
-Ren, Yi. *MamaRetrieval* v0.1.0. 2026. <https://huggingface.co/datasets/nmrenyi/mamaretrieval>
+Ren, Yi. *MamaRetrieval* v0.2.0. 2026. <https://huggingface.co/datasets/nmrenyi/mamaretrieval>
 
 ## Limitations
 
 - **Scope**: midwifery / OBGYN / neonatal care, framed for guidelines deployed
   in Zanzibar. Performance numbers do not transfer cleanly to general
   medical retrieval.
-- **Depth-3 ceiling**: ~25% of queries have no `score ≥ 5` chunk in any
-  retriever's top-3, even from the strongest retriever. This is an inherent
-  depth-3 pool limit, not a retriever failure.
+- **Depth-20 ceiling**: ~6% of queries have no `score ≥ 5` chunk in the
+  6-retriever top-20 union, even from the strongest retriever. This is an
+  inherent retrieval ceiling for the producer corpus + retriever set, not
+  a per-retriever failure. The Tier 3 labels make this measurable directly
+  (`196 / 3185` queries fall outside the strict-relevance pool).
 - **Single relevance judge**: every `(query, chunk)` relevance label in
   this dataset is produced by one LLM (`Qwen/Qwen3.5-397B-A17B-FP8`) under
   the four-dimension rubric. That judge was calibrated against Claude Opus

audit/judgments_with_reasoning.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ac827a0c75bff5f35a1b3488a3ffe4aca46fc7e2c4ed53a7f13604ab04e0db71
-size 122884735
+oid sha256:4ba2969a0e2fe9ac8c2f1dcb1388e76d6d3be494e06811dd9f2d3cad69ae5305
+size 768951185

data/chunks.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e1c90755e14891479aea49f898d0166dd7d1953e3502e9f4e16120667574cf2b
-size 5642002
+oid sha256:86f62784d90c269863898240288ef7dace519715bc99cf89ae6305732b4ec0e5
+size 12142532

data/judgments.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:86495f5767e68aed71471866cc069d8212b269294aab93d5504d07e784133049
-size 479037
+oid sha256:55faeb775bcf31450b21d1c3cd6d7e84367ceac1c02168945546e85913ff5c84
+size 1644468

data/rankings.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:54f643b4cbca594bb5de38c71a06cc3dea77a5dc4d4d544e83e3235c055deaa4
-size 791597
+oid sha256:79bc6f6d420687e5462e2e90c2a0de8f747f2d73a22680d84533fc39d610612e
+size 3271956

manifest.json

@@ -1,9 +1,14 @@
 {
   "name": "mamaretrieval",
-  "version": "v0.1.0",
-  "release_date": "2026-05-19",
+  "version": "v0.2.0",
+  "release_date": "2026-05-24",
   "description": "Per-retriever evaluation of 6 retrievers on 3,185 midwifery / OBGYN queries against the rag-bundle-v0.2.0 corpus, graded by an LLM judge under a 4-dimension rubric (D1 topic, D2 meaningful, D3 actionable, D4 density; score = d1 \u00d7 (d2 + d3 + d4) \u2208 [0..6]).",
-  "scope": "Tier 2: union of top-3 from 6 retrievers. v0.2.0 will extend to top-20 union when Tier 3 lands.",
+  "scope": "Tier 3: union of top-20 from 6 retrievers \u2014 230,964 judged (q, c) pairs. Top-3 union (Tier 2's scope) is a strict subset, recoverable from rankings.parquet with rank <= 3.",
+  "previous_version": {
+    "version": "v0.1.0",
+    "scope": "Tier 2: union of top-3, 36,418 judged pairs.",
+    "build_commit": "02127e3"
+  },
   "judge": {
     "model": "Qwen/Qwen3.5-397B-A17B-FP8",
     "prompt_hash": "9d2abdfb76b030ea",
@@ -42,7 +47,7 @@
       "model": "gecko-1024-quant-v0.2.0 (on-device TFLite)"
     }
   ],
-  "top_k": 3,
+  "top_k": 20,
   "corpus": {
     "bundle": "rag-bundle-v0.2.0",
     "producer_commit": "a1abe003cce742b46954375d17abb28a3e27110f"
@@ -52,20 +57,20 @@
       "rows": 3185
     },
     "rankings": {
-      "rows": 57330,
-      "depth": 3
+      "rows": 382200,
+      "depth": 20
     },
     "judgments": {
-      "rows": 36418
+      "rows": 230964
     },
     "judgments_with_reasoning": {
-      "rows": 36418,
+      "rows": 230964,
       "extra_columns": [
         "thinking"
       ]
     },
     "chunks": {
-      "rows": 17827
+      "rows": 41298
     }
   }
 }