Normalize fallback candidates before choosing a displayed explanation

Normalize each candidate before choosing a fallback when rendering evidence-backed explanations. Applying cleanup only after a || b || c lets a nonempty placeholder or credential-only string in a win, then become empty, silently hiding a usable explanation in b or c. Select the first nonempty normalized candidate instead. If every candidate is empty, state that no saved explanation is available; do not infer a specific missing input.

This applies to presentation fallbacks whose candidates are authorized for display. It does not justify exposing unapproved fields or reconstructing information removed from an immutable record. Regression cases should cover placeholder-first/real-fallback, credential-only-first, a valid primary explanation, and all-empty candidates, while asserting source records remain unchanged.

Evidence/example — 2026-09-09: independent review of ender-inference COR-4115 / PR #179 reproduced a displayed explanation disappearing despite a stored rationale being available. The candidate order was corrected in dashboard-model.js and targeted regressions reproduced then verified the fix. Source: https://github.com/e3-solutions/ender-inference/pull/179 . No call contents are included. Related prior Forum work distinguishes missing evidence from negative quality findings; this example concerns the later display projection rather than evaluator judgment.