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-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 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

from datasets import load_dataset

queries   = load_dataset("nmrenyi/mamaretrieval", "queries",   split="test")
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 (~770 MB)
judgments_full = load_dataset("nmrenyi/mamaretrieval",
                              "judgments_with_reasoning", split="test")

Configs

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	382,200	query_id, retriever, rank, chunk_id, score	For every query × retriever combination, the top-20 chunk_ids 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.
• 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-20, and when they do they are not always the highest-rated chunk for that query.

Rubric

The judge scores each (query, chunk) pair on four dimensions:

• D1 — Topic (bool): does the chunk address the same clinical problem as the query (same condition, intervention, and clinical-timing context)? If D1 = false, D2 = D3 = D4 = 0 automatically.
• D2 — Meaningful clinical content (0–2): how rich is the chunk's clinical content, independent of whether it specifically answers the query?
• D3 — Actionable guidance (0–2): how specific is the actionable guidance — vague advice (0), general direction (1), exact doses/thresholds/steps (2)?
• D4 — Density (0–2): what fraction of the chunk is directly useful for answering this specific query?

score = d1 × (d2 + d3 + d4) ∈ [0..6].

The full prompt — including four worked examples that anchor the calibration — is shipped verbatim at audit/judge_relevance_prompt.txt. Its prompt_hash is recorded in manifest.json.

Retrievers

name	model
bm25	BM25 (lexical baseline)
medcpt	ncbi/MedCPT (Query + Article encoders)
octen	Octen/Octen-Embedding-8B
voyage	voyage-4-large
lateon	lightonai/GTE-ModernColBERT-v1 (late-interaction ColBERT)
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 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

1. Query generation. For each clinically-relevant chunk in the producer corpus, an LLM (Qwen/Qwen3.6-27B-FP8) was prompted to produce one ≤20-word clinical question the chunk could answer. Chunks judged non-clinical (e.g. course outlines, references, learning objectives) were skipped. The full prompt is shipped at 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-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 judgments_with_reasoning.

Validation: the judge model was calibrated against Claude Opus 4.7 reference labels on a 62-pair pilot, with 95% threshold agreement at score ≥ 3 and 85% at score ≥ 5.

Provenance

• Producer corpus: rag-bundle-v0.2.0, produced at commit a1abe003 of nmrenyi/mamai-medical-guidelines. 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, 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 pins exact judge and generator model IDs, prompt hashes, and schema versions.

License — Research use only

This dataset is released for non-commercial academic research and retrieval-evaluation benchmarking only. By downloading or using it, you agree to all of the following:

Permitted

• Academic research, including publication of aggregate metrics, qualitative analysis, ablations, and methodology comparisons.
• Use as an evaluation benchmark for retrieval systems.
• Re-running the rubric or running new judges against the included (query, chunk) pairs for methodology research.

Not permitted without explicit written permission

• Any commercial use, including evaluation as part of internal product decisions at for-profit organisations.
• Use of the chunk text as training data for any model — generative, embedding, retrieval, or otherwise.
• Redistribution of the chunk text, in whole or in part, outside the form shipped here (i.e. do not extract chunks.parquet, repackage, mirror, or re-host the chunk content).
• Production deployment of any system whose retrieval or judging behaviour has been tuned on this data.
• Clinical use of the chunk text. None of the chunk content has been reviewed for clinical accuracy in the form presented here; do not surface it to patients or clinicians.

Full terms — including upstream-licensing constraints, attribution, and warranty disclaimers — are in LICENSE.

Citation

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-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 4.7 on a 62-pair pilot — 95% threshold agreement at score ≥ 3, 85% at ≥ 5 — but that's a small LLM-vs-LLM sanity check, not a human-annotated gold standard. Practical consequences: retriever-vs-retriever rankings tend to be stable across reasonable relevance judges, but absolute score distributions and per-row labels will shift if you re-grade the same (query, chunk) pairs with a different judge. Treat each label as one judge's calibrated opinion, not ground truth.