Explanation#

Understanding-oriented pages. They answer why rather than how: what each design decision buys, what it costs, and which failure it was chosen to prevent. Nothing here is needed to use the tool — How-to guides and Reference are for that — but it is what makes the tool’s behaviour predictable instead of surprising.

Page

The question it answers

What vs. where

Why moving a stage to a cluster is config and not code

The stage contract

Why a stage can refuse, and why a non-zero exit is not an error

The funnel, and what stopping means

Why the pipeline fans out then narrows, and what “stopping” means

Scheduling and restart

Why an interrupted run resumes instead of starting over

Remote execution

Why a remote machine holds none of MatMelt’s code

What the stages actually compute

What each stage actually computes, and how far to trust it

One idea runs through all six: the tool decides what to compute and never decides where; it judges results and never invents them; and everything it learns is written down before it is acted on. Each page below is that idea applied to one part of the system.