Simplify dataset card

README.md

@@ -42,63 +42,75 @@ configs:
 
 # Local GraphRAG Cascade Artifacts
 
-This dataset contains the fixed corpus and final scientific artifacts for a
-controlled GraphRAG experiment with 12 local builder models and three cumulative
-graph conditions:
+This repository contains the data and output artifacts produced for my master's
+thesis research on knowledge graph construction in local GraphRAG systems. It
+is an evidence package for the study, rather than the code used to run the
+experiment.
+
+The collection includes the study corpus, model extraction outputs, final
+knowledge graphs, retrieved evidence, generated answers, and evaluation results.
+Together, these files make it possible to inspect the experiment from source
+documents through to question-level outcomes.
+
+The complete experiment harness, including the code, configuration, and commands
+used to produce these artifacts, will be published separately on
+[my GitHub profile](https://github.com/georgijkutivadze).
+
+## What was compared
+
+The experiment used 12 local language models to build graphs under three
+cumulative conditions:
 
 ```text
 Native -> ER -> ER+RR
 ```
 
-Entity Resolution (ER) is applied to the Native graph. Relation Recovery (RR)
-is then applied to the ER graph. The dataset does not contain a Native+RR
-condition and does not estimate an independent RR effect.
+`Native` is the original LightRAG graph. `ER` applies Entity Resolution to that
+graph. `ER+RR` then applies Relation Recovery to the ER result. Because the
+conditions form a sequence, this collection does not include a separate
+Native+RR condition or measure RR as an independent intervention.
 
-## Dataset scope
+## What is included
 
 - 120 MultiHop-RAG questions: 30 inference, 30 comparison, 30 temporal, and 30
   unanswerable.
 - 155 documents: 125 gold documents and 30 hard negatives.
 - 12 builder models and 36 graph conditions.
-- 36 retrieval outputs, 36 answer outputs, and 4,320 question-level evaluation
-  records.
-- Raw and normalized extraction records, final graph node/edge exports, paired
-  cascade comparisons, and exploratory cross-builder analyses.
+- Raw model calls and normalized extraction records.
+- Final graph nodes and edges for every model and condition.
+- 36 retrieval outputs, 36 generated-answer outputs, and 4,320 question-level
+  evaluation records.
+- Paired comparisons between the cumulative conditions and exploratory
+  cross-model analyses.
 
-Builders: `gemma3_12b`, `gemma3_1b`, `gemma3_270m`, `gemma3_4b`, `qwen35_0_8b`, `qwen35_2b`, `qwen35_4b`, `qwen35_9b`, `qwen3_0_6b`, `qwen3_1_7b`, `qwen3_4b`, `qwen3_8b`.
+<details>
+<summary>Builder model identifiers</summary>
 
-## Repository structure
+`gemma3_270m`, `gemma3_1b`, `gemma3_4b`, `gemma3_12b`, `qwen3_0_6b`,
+`qwen3_1_7b`, `qwen3_4b`, `qwen3_8b`, `qwen35_0_8b`, `qwen35_2b`,
+`qwen35_4b`, and `qwen35_9b`.
 
-```text
-README.md
-manifest.json
-checksums.sha256
-corpus/
-extraction/
-graphs/
-retrieval/
-answers/
-evaluation/
-```
+</details>
+
+## Finding the files
 
 | Directory | Contents |
 | --- | --- |
-| `corpus/` | Questions, documents, subset manifest, audit, and selection provenance |
-| `extraction/<builder>/` | Raw model calls and normalized chunks, entities, and relations |
-| `graphs/<builder>/<regime>/` | Final graph nodes and edges for each cumulative condition |
-| `retrieval/<builder>/` | One 120-record JSONL file per graph condition |
-| `answers/<builder>/` | One 120-record JSONL file per graph condition |
-| `evaluation/question_metrics/<builder>/` | Question-level retrieval and QA metrics |
-| `evaluation/builders/<builder>/` | Builder summaries and three paired cascade contrasts |
-| `evaluation/aggregate/` | Prespecified and exploratory cross-builder results |
+| `corpus/` | Questions, source documents, and subset provenance |
+| `extraction/` | Raw model calls and normalized chunks, entities, and relations |
+| `graphs/` | Final graph nodes and edges |
+| `retrieval/` | Evidence retrieved for each question |
+| `answers/` | Answers generated from the retrieved evidence |
+| `evaluation/` | Question-level metrics, paired comparisons, and aggregate results |
 
-The top-level `manifest.json` records every published payload file, its role,
-record count, source artifact, byte size, and SHA-256 digest. Paths are relative
-and contain no machine-specific workspace locations.
+Files are grouped first by builder model and then, where applicable, by graph
+condition. The top-level `manifest.json` provides the role, record count, source,
+size, and checksum of every published payload file.
 
-## Loading examples
+## Loading the data
 
-Load a homogeneous JSONL file with 🤗 Datasets:
+Each JSONL file has one consistent schema and can be loaded separately with
+🤗 Datasets. For example:
 
 ```python
 from datasets import load_dataset
@@ -116,9 +128,9 @@ metrics = load_dataset(
 )
 ```
 
-The artifact families have different schemas and should be loaded separately.
-Join question-level files on `question_id`; use `builder_model`, `base_run_id`,
-`variant_run_id`, and `graph_regime` for lineage across stages.
+The different artifact families use different schemas. Question-level files can
+be joined on `question_id`. Model, run, and condition fields preserve lineage
+between extraction, graphs, retrieval, answers, and evaluation.
 
 ## Integrity
 
@@ -134,17 +146,18 @@ On macOS:
 shasum -a 256 -c checksums.sha256
 ```
 
-## Provenance and reproducibility
+## Study design and provenance
 
-The corpus is a fixed, stratified subset derived from MultiHop-RAG. Extraction
-was performed once per builder. ER and RR reuse the saved extraction records,
-and retrieval and answering use the same questions and fixed downstream models
-across all conditions. Exact model identities, hashes, schema versions, and run
-lineage are preserved in the records and `manifest.json`.
+The corpus is a fixed, question-type-balanced subset derived from MultiHop-RAG.
+Extraction was performed once per builder, after which ER and RR reused the
+saved extraction records. Retrieval and answering used the same questions and
+fixed downstream models in all three conditions. The records and manifest
+preserve model identities, hashes, schema versions, and run lineage.
 
-The package excludes mutable LightRAG workspaces, vector stores, SQLite state,
-logs, caches, and model weights. Those files are execution infrastructure rather
-than scientific exchange artifacts.
+Mutable LightRAG workspaces, vector stores, logs, caches, and model weights are
+not included because they are execution infrastructure rather than the research
+artifacts intended for analysis and reuse. See the experiment harness linked
+above for the implementation and reproduction workflow.
 
 ## Limitations