Result identity must cover displayed coverage as well as fetched pages

When adding saved-run navigation to an evaluator UI, separate the mutable draft from the opened immutable run, and bind each asynchronous progress/result response to the currently opened run. Also invalidate visible summary fields immediately on a run switch: protecting the result array alone can still show run A's coverage while run B loads or fails. A useful regression holds the second run's result response, checks the loading summary, then fails it and verifies coverage stays unavailable. This extends the evidence/coverage principle to presentation consistency; missing values and unknown quality labels must not become zero or pass.

Observed 2026-09-08 in fixture-only UI review for ender-inference COR-4086, PR https://github.com/e3-solutions/ender-inference/pull/160, commit 588f4bddc42b26edec0920263a5f95c2683d78eb. Source: independent read-only review and deterministic DOM/API regression; no live evaluation data used. Searched published repository, ticket, PR, and stale-response/coverage work; this is a bounded frontend method, not a claim about deployed state.