Preserve the source. Reproduce the transformation.

Raw protects evidence. Clean creates canonical research state. Neither becomes production data without a separate approval record.

01

Evidence layer

Raw Data

Source-faithful, append-only snapshots carrying request parameters, retrieval time, checksum, provider identity and licensing class.

  • No silent value rewrites
  • Immutable after manifest creation
  • Partitioned by source and ingestion date
  • Never read directly by production
02

Canonical layer

Clean Data

Deterministic transformations with canonical identifiers, units, timestamps, schemas, availability semantics and complete Raw lineage.

  • Versioned transformation code
  • Point-in-time available_at
  • Quality report attached
  • Immutable snapshot ID

Lineage contract

Every output points backward.

A Clean snapshot must be reproducible from declared Raw snapshots and an immutable transformation version.

INPUTRaw Snapshot IDs
+
CODETransform Version
OUTPUTClean Snapshot
+
EVIDENCEQuality Report

Landing registry

Current implementation state.

This table is a transparent build dashboard, not a fabricated feed monitor.

DatasetZoneContractData statusProduction
A500 constituentsRaw → CleanDefinedSource pendingDenied
Daily OHLCVRaw → CleanDefinedSource pendingDenied
FundamentalsRaw → Clean PITDraftSource pendingDenied
News & sentimentRaw → Clean NLPRequirementsNot connectedDenied