§2 Chunk 01 — TRANSFORMS

TRANSFORMS0x02A0

The TRANSFORMS chunk is the workflow, remembered. Filter, join, pivot — each transform is sealed into the file in order, with the SQL that produced it and a hash of the step itself, so the workflow replays exactly: same steps, same order, same result. The concept story's payload holds 14 steps; the count is the story, the shape is the spec.

Why it matters

A result you cannot replay is a rumor. Work usually lives in someone's session history — half-remembered filters, a join that took three tries, the one pivot that made the chart make sense. Sealing the steps with their SQL turns the workflow into an object: open the artifact months later, on different data, and the same steps run in the same order.

The chunk

Steps are stored in application order — the replay order — and each carries its own hash, so a tampered or truncated artifact fails loudly instead of silently drifting.

Compute portability

Compute is portable. The same workbench runs against a local file, a local database, or an attached MotherDuck workspace — and which one it uses is your call, made per query, not a setting you commit to up front. Local is the default; the cloud is an option you reach for, not a requirement you start with.

Most work belongs on your machine, where it is fast and private. Some work — a very large scan, a long-running aggregation — outgrows a laptop. Compute portability lets a single heavy query burst to the cloud while everything else stays local. You get the headroom when you need it without moving your whole workflow somewhere else, and without a heavy query blocking the rest of your session.

The query you write does not change when the engine does. Only where it runs moves — from your machine to an attached cloud workspace and back:

engine: duckdb · local        # default — runs on your machine
        ─────────────────
        ↓  attach when a query outgrows the laptop
        ─────────────────
engine: motherduck · cloud    # opt-in — only when you choose

The choice is per query and reversible. Nothing is pinned to the cloud, and nothing is uploaded by default — the TRANSFORMS chunk records where each step ran, so the artifact remembers the engine split as faithfully as the steps.

Continue to §3 Chunk 02 — QUERIES.