Hugging Face's logo

Datasets:
gss1147
/
P_P_GHOST_1Million

Tags:
prompt-orchestration
rag
react
chain-of-thought
self-consistency
self-refine
Dataset card Files
xet
Community
Dataset Viewer

The dataset viewer is not available because its heuristics could not detect any supported data files. You can try uploading some data files, or configuring the data files location manually.

P_P_GHOST — 1,000,000 record pack (sharded)

This repository contains 1,000,000 JSONL records (gzipped shards) to train/evaluate prompt orchestration that creates the illusion of instant fine-tuning via:

  • prompt packets / compilation
  • routing + retrieval grounding (RAG)
  • tool-use loops (ReAct)
  • schema enforcement + reasking/repair loops
  • end-of-pass meta-optimization (self-refine style)

Truth policy

Each record references at least one evidence capsule with a public source URL in capsules/evidence_capsules.jsonl. Failure records include a validator trace and a corrected repair output.

Files

  • data/train-00000.jsonl.gz ... train-00019.jsonl.gz (20 shards × 50,000 = 1,000,000)
  • capsules/evidence_capsules.jsonl
  • schema/ppg_record.schema.json
  • eval/validators.py

Generated: 2026-01-04T19:58:05.653029Z

Downloads last month
40

Collections including gss1147/P_P_GHOST_1Million