Provenance-Grade Ingestion

Every data vendor will sell you ticks. None of them will sell you the ability to prove, two years later, that the file you backtested is byte-identical to the file you ingested — or that you registered your hypothesis before you saw the data that confirms it. That discipline is the product.

What it does

• Hash at the door. Every dataset is SHA-256 content-hashed at the moment of ingestion, before anyone has looked at it. From that moment, any modification — cleaning, resampling, “fixing” — produces a new hash and a recorded lineage step. Silent mutation becomes structurally impossible.
• Registration timestamps. When data arrives relative to when hypotheses about it were registered is evidential bedrock. Data fetched after a hypothesis was formed carries different weight than data that predates it, and the ledger records which is which — permanently.
• Gap ledgers. Missing sessions, backfilled ranges, and vendor corrections are first-class recorded events, not silent patches. A dataset’s honesty is measured by how visible its wounds are.
• The burned-door rail. Our own Burned Door incident — data fetched minutes before registration, destroying its evidential value — is the founding case. The instrument exists so that class of destruction is caught at ingestion, not discovered in an audit.

What it does not do

It does not clean your data or improve its quality. It records what the data is, when it arrived, and everything that happened to it since — provenance, not polish.

Who it is for

Quant teams whose backtests must survive hostile review: their own, a regulator’s, or an allocator’s.