Keeping objects on purpose

Some objects are unused by the reports ripbi can see on purpose: a measure only an Excel pivot or a thin report in another workspace reads, or a column kept for an upcoming report. Mark them in the model itself with a ripbi_keep annotation (issue #151). Its value is the reason:

measure 'Excel Margin %' = DIVIDE([Margin], [Revenue])
    annotation ripbi_keep = Used by the Finance Excel pivot (FIN-231)

• A kept object is a reachability root, like a report binding: it is live, and so is everything it references — [Margin], [Revenue], and the columns behind them, which the Excel pivot needs too. None of them is a finding, and none gates.
• Any object can carry it: a table, column, measure, hierarchy, calculation item, relationship, shared expression, or function. A kept table keeps its members. A kept relationship counts as activated, so its key columns stay live even when it is inactive; like every relationship, it does not keep its tables alive.
• Every model format reads it: TMDL, model.bim, and the model inside a PBIX, PBIT, or .abf. The name is case-insensitive. Add it in Power BI Desktop's TMDL view, Tabular Editor, or the TMDL file; these tools and deployment pipelines carry annotations along, and a rename carries the annotation with the object. ripbi never writes it.
• Write the reason. An empty value (annotation ripbi_keep =) still keeps the object, but the next reader will not know why.
• Output. Kept objects are not listed by scan. --json counts them in summary.kept. ripbi deps --impact shows the annotation's reason in a Kept section, so "why is this alive?" answers with the reason (see deps.md).

Prefer the annotation over a [scan].ignore pattern for an intentionally unused object: the reason sits beside the object, it survives renames, and it keeps the object's inputs alive where an ignore pattern would leave them reported as dead. [scan].ignore remains the tool for findings you want silenced without a model edit, such as whole families of generated objects.