refactor(semantics): close dashboard testing protocol audit

frontend/src/lib/components/scenario-run/AgentEvaluationCard.svelte

@@ -3,6 +3,9 @@
 <!-- @BRIEF Read-only projection of one persisted AgentEvaluation record and the policy-derived
      step outcome. The model verdict is never presented as the ScenarioResult (RUNMON-FR-013). -->
 <!-- @INVARIANT No prompt/retry/provider/agent controls: evaluation is an immutable audit record. -->
+<!-- @UX_STATE Verdict(pass|fail|inconclusive|unknown): localized model verdict and independent policy step status remain visible; absent confidence renders —. -->
+<!-- @UX_STATE Findings(empty|present): empty message or severity-marked immutable findings; criterion and provider details expand independently through native details elements. -->
+<!-- @UX_REACTIVITY evaluation and stepStatus are parent-owned immutable props; disclosure state is local browser state, never execution authority. -->
 <script lang="ts">
   import { t } from "$lib/i18n/index.svelte.js";
   import type { EvaluationRecord } from "$lib/types/scenario-run";

frontend/src/lib/components/scenario-run/ScenarioResultView.svelte

@@ -8,6 +8,10 @@
 <!-- @RELATION DEPENDS_ON -> [ScenarioRunMonitor.Component.MetricDeviation] -->
 <!-- @INVARIANT Deterministic result, model evaluation and policy outcome stay visually distinct;
      unavailable/redacted evidence never renders as a passing badge. -->
+<!-- @UX_STATE Metric(no-pin|mismatch|unproved|no-mismatch): show missing-pin warning, persisted failed-comparison details, nonPASS uncertainty, or no-mismatch message without changing result.status. -->
+<!-- @UX_STATE Selection(none|selected): failure/deviation controls emit onselectstep; the parent-owned selectedStepId drives the run-bound inspector and suppresses the matching duplicate deviation card. -->
+<!-- @UX_STATE Disclosures(collapsed|expanded): provenance and failure technical fields expand locally; failed checks and unavailable evidence retain authoritative statuses. -->
+<!-- @UX_REACTIVITY result, steps, plan, runId and selectedStepId come from the same parent-selected run; selection callbacks do not fetch or alter runtime results. -->
 <script lang="ts">
   import { t } from "$lib/i18n/index.svelte.js";
   import type { ScenarioExecutionResult, ScenarioStepRun } from "$lib/types/scenario-run";