Answer audits from the model: query, diff and harvest #55

Parent

#35

What to build

Three operations the chapters do by hand become use-cases over the same core:

• Query: the readers of a Secret Store path, the consumers of a surface, the readers of an exposure. Today chapter 10 answers these with grep -r over env files.
• Diff: two Resolved Deployments compared by element identity and trace, not by line. This is the review surface 0029 describes.
• Harvest: live manifests read into candidate Project Intent, with whatever cannot be expressed written to a ledger. It serves adoption (chapter 60) and 0005's render-and-diff. Harvest reads live state at adoption time only; chapter 60 says so, which keeps 0006's rule about renders intact.

Parity

Not a parity case. Query, diff and harvest are TypeScript use-cases only; the Java implementation under emf/ implements the graded pipeline (parse, validate, resolve, render) and nothing else.

Acceptance criteria

• Each query answers, over fixtures, what the chapter's grep answers.
• A diff of two projections reports the changed decisions by element.
• Harvesting one rendered example tree yields intent that renders back to that tree, apart from ledgered residue.

Blocked by

• Evaluate identity and reference invariants over a composed union #46
• Specify the trace and check chapter 16's three properties on it #48