The Layers

In superbrief:

1. Content — filesystems, and their hashes.
2. Execution — hashes go in, something is exec’d, hashes come out.
3. Plans — multiple steps of execution are described in advance.
4. Intentions — a templating system can crank out plans.

L0: Content

• nuff said. WareIDs. That's it.
• A WareID is a content-addressable identifier of a Ware. A Ware is just a fancy name for (packed) filesystem snapshot.
• A Ware has no implied or presumed structure. It’s just files.

L1: Hashed / Crystallized / Executable

• Ready to invoke — all inputs are content-addressed wareIDs.
• No catalog references, nothing.
• Internal graph structure no longer allowed either. No pipes. Just one action.
• Is the basis of memoization. You can hash this whole document.
  • Should have no extraneous non-load-bearing contents, in order for this to work ideally. No semantic tagging should float down this far.
  • One exception to this: if mounts have been used here, memoization should probably not be applied.
• Mostly not seen by a user, in practice, it turns out.
  • If you engage some real verbose debugging flags, sure.

L2: Flattened / Resolved

• The replayable.
• Still has catalog references.