Skip to content

README.md

Latest commit

 

History

History
27 lines (17 loc) · 1.71 KB

README.md

File metadata and controls

27 lines (17 loc) · 1.71 KB
Raw

Integration Examples

Paired JSONL files showing each handoff between the three packages. Use them as fixtures or as references for the exact field set required at each boundary. See docs/integration.md and docs/data-contracts.md for the contract.

Ingestion → Observability and Foundry

  • ingestion_documents_for_dataset_foundry.jsonl — normalized documents emitted by ingestion.
  • ingestion_chunks_for_observability.jsonl — chunks that runtime serving indexes and that observability references by chunk_id / document_id.
  • ingestion_lineage_for_dataset_foundry.jsonl — chunk-to-document lineage consumed by the foundry.

Observability → Foundry

  • output_for_dataset_foundry_interactions.jsonl — interaction events with query_id, trace_id, prompt_text, response_text, model_context, retrieval_references.
  • output_for_dataset_foundry_retrieval_traces.jsonl — retrieval traces with query_id, trace_id, references[], model_version, dataset_version.

Foundry outputs

  • output_prompt_response_sample.jsonl — curated prompt-response pairs.
  • output_retrieval_eval_sample.jsonl — curated retrieval-evaluation records.
  • output_manifest_sample.json — dataset manifest.

"Input from" mirrors

  • input_from_ingestion_documents.jsonl, input_from_ingestion_chunks.jsonl — the same ingestion outputs framed as foundry/observability inputs.
  • input_from_observability_interactions.jsonl, input_from_observability_retrieval_traces.jsonl — the observability outputs framed as foundry inputs.

These mirror pairs exist so each consumer can be tested in isolation against the exact shape it will see at runtime.