dataset_card.md

Dataset Card: curatorkit_run

Summary

• Run timestamp: 2026-09-01T05:33:22.078875Z
• Wall clock time: 50.8s
• Pipeline config hash: 3b235ce9a22decbf

Source Documents

• Unknown

Pipeline Stages

Step	Input	Passed	Recovered	Rejected	Forward
JSONLReader	40	40	—	0	40
SchemaGate	40	39	0	1	39
ExactDeduplicator	39	39	—	—	39
TextCleaner	39	39	—	—	39
SecretsGate	39	38	0	1	38
PIIPseudonymizer	38	38	—	—	38
ToxicityGate	38	30	0	8	30
AlpacaExporter	—	30	—	—	30

Gate Pass Rates and Rejection Breakdown

• JSONLReader: 40/0 passed (N/A), 0 rejected
• SchemaGate: 39/40 passed (97.5%), 1 rejected
• SecretsGate: 38/39 passed (97.4%), 1 rejected
• ToxicityGate: 30/38 passed (78.9%), 8 rejected

Rejection reasons:

• toxic_content:classifier:0.996: 4
• below_min_tokens:9: 1
• secret_detected:AWS Access Key: 1
• toxic_content:classifier:0.993: 1
• toxic_content:classifier:0.995: 1
• toxic_content:classifier:0.997: 1
• toxic_content:classifier:0.992: 1

Deduplication

• exact_duplicates_removed: 0

Known Limitations

• No MinHash deduplication applied.
• Table extraction requires extract_tables: true on the PDF reader and MinerU with GPU support. Disabled by default.
• Token counting uses a whitespace tokenizer by default. For exact LLM context window enforcement, enable use_tiktoken: true in the SchemaGate config.
• Both within-run (MinHash/Diversity) and cross-run (EmbeddingDeduplicator) deduplication are available. Cross-run dedup persists an embedding index across runs; use embedding_reset_index: true to start fresh.

Reproduction

pip install curatorkit
curatorkit run pipeline.yaml --output-dir ./out/

Tool Versions

• curatorkit: 0.1.6
• python: 3.13.15