§5 Chunk 04 — LINEAGE

LINEAGE0x04B1

The LINEAGE chunk is the proof. Every result carries its derivation — the source fingerprints and step hashes that produced it — so the path from raw input to final number is always recoverable. The lineage travels with the work: a result is never separated from how it was derived.

Why it matters

A workflow is usually scattered: a file here, a notebook there, a query in someone's history, a result no one can trace back. The .dat0 file collapses all of that into one attachable object — and this chunk is what makes the object answerable. Hand someone the file and they get the result, the steps, and the reason each step exists — not just a number to take on faith.

A chain of hashes

Lineage is a sha-256 chain. Each transform step is hashed with its SQL and its inputs; each source carries a fingerprint; the artifact's seal covers every preceding byte. Entries address other chunks by content — a source fingerprint plus a byte range, the way content-addressed systems do it — so provenance survives rewrites of the artifact itself:

events.parquet ──┐
                 ├──▶ join ──▶ filter ──▶ rollup ──▶ result
customers (pg) ──┘

The chunk

The payload is the chain itself — sha-256 chain: step hashes plus source fingerprints addressing other chunks by content and byte range. Diff two artifacts and the differing links name what changed; replay one and the chain verifies every step on the way down.

That is the end of the file. §6 closes with what "draft" means for all of this — continue to §6 Versioning & status.