ADR 007: Don't scaffold a builder chain ahead of an implemented operation

Status

Accepted

Context

Cdm().Policies().Mappings().Inputs().Resolved() could be constructed — both Inputs() and Resolved() existed as request-builder accessors — but neither had a verb method (Get/Post/etc.) or a To*RequestInformation method. The resolved-inputs endpoint was advertised in the fluent chain and in docs/snippets, but was actually unreachable: calling it did nothing, because there was nothing to call. Inputs() itself existed only as a stepping stone to reach Resolved() (#489).

Alternatives considered:

1. Implement the missing verb methods to make the chain functional — rejected for this PR specifically because the endpoint's actual behavior wasn't yet scoped; forcing an implementation just to match the existing (accidental) shape risked guessing at the wrong contract.
2. Leave the dead-end chain in place as a documented "coming soon" — rejected; an SDK builder chain that compiles and can be navigated but silently does nothing on every verb is worse than a compile error, and contradicts the nil-receiver-guard principle in ADR-006 of failing loud and early rather than silently.
3. Remove the dead-end chain entirely (Inputs()/Resolved() and their builder types, 177 lines) until the endpoint is implemented additively.

Decision

Remove builder-chain segments that have no reachable verb operation, rather than leaving them in place as a stub. A request-builder accessor should only exist once the operations it leads to are actually implemented — not speculatively, to reserve a URL segment or match an API's documented shape in advance.

Consequences

• Pros: every reachable point in a fluent chain does something; a builder existing is a reliable signal that its operations work, matching the "fail loud" spirit of ADR-006 rather than silently advertising unreachable surface area.
• Cons: breaking removal for anyone who had already started depending on the chain shape (even though no verb ever worked). Re-adding the Inputs/Resolved endpoint later is an additive change, not a revert.
• Rule for new modules and reviews: don't scaffold a full URL-segment chain "for completeness" ahead of the operation being implemented. Add the builder when the verb methods it exposes are ready to ship together.