diff --git a/test/unit/advisory-live-parity.test.ts b/test/unit/advisory-live-parity.test.ts
new file mode 100644
index 0000000000..00ac216114
--- /dev/null
+++ b/test/unit/advisory-live-parity.test.ts
@@ -0,0 +1,148 @@
+// Locks advisory==live computed disposition parity (#2190). Advisory mode posts the non-enforcing
+// conclusion while dry-run exposes the byte-identical would-be verdict that live enforcement applies.
+import { describe, expect, it, vi } from "vitest";
+import { evaluateGateCheck, type GateCheckConclusion, type GateCheckPolicy } from "../../src/rules/advisory";
+import { planAgentMaintenanceActions, type AgentActionPlanInput } from "../../src/settings/agent-actions";
+import type { Advisory } from "../../src/types";
+
+function missingIssueAdvisory(): Advisory {
+  return {
+    id: "advisory-live-parity",
+    targetType: "pull_request",
+    targetKey: "owner/repo#7",
+    repoFullName: "owner/repo",
+    pullNumber: 7,
+    headSha: "sha7",
+    conclusion: "neutral",
+    severity: "warning",
+    title: "Gittensory advisory available",
+    summary: "1 advisory finding generated.",
+    findings: [
+      {
+        code: "missing_linked_issue",
+        title: "No linked issue detected",
+        severity: "warning",
+        detail: "No closing reference.",
+        action: "Link the issue.",
+      },
+    ],
+    generatedAt: "2026-06-13T00:00:00.000Z",
+  };
+}
+
+function cleanPass(): Advisory {
+  return { ...missingIssueAdvisory(), findings: [] };
+}
+
+function aiFail(): Advisory {
+  return {
+    ...missingIssueAdvisory(),
+    findings: [
+      {
+        code: "ai_consensus_defect",
+        title: "AI consensus defect",
+        severity: "warning",
+        detail: "both models flagged a real defect",
+        action: "fix it",
+      },
+    ],
+  };
+}
+
+function advisoryWouldBeConclusion(advisory: Advisory, policy: GateCheckPolicy): GateCheckConclusion {
+  const out = evaluateGateCheck(advisory, { ...policy, dryRun: true });
+  expect(out.displayConclusion).toBeDefined();
+  return out.displayConclusion!;
+}
+
+function liveConclusion(advisory: Advisory, policy: GateCheckPolicy): GateCheckConclusion {
+  return evaluateGateCheck(advisory, policy).conclusion;
+}
+
+function agentInput(overrides: Partial<AgentActionPlanInput> & { conclusion: GateCheckConclusion }): AgentActionPlanInput {
+  return {
+    blockerTitles: [],
+    autonomy: {},
+    autoMaintain: { requireApprovals: 0, mergeMethod: "squash" },
+    slopGateMinScore: 60,
+    changedPaths: [],
+    hardGuardrailGlobs: [],
+    authorIsOwner: false,
+    authorIsAdmin: false,
+    authorIsAutomationBot: false,
+    ciState: "passed",
+    pr: { labels: [], mergeableState: "clean", headSha: "head" },
+    ...overrides,
+  };
+}
+
+const terminalClass = (actions: ReturnType<typeof planAgentMaintenanceActions>) =>
+  actions.find((action) => action.actionClass === "merge" || action.actionClass === "close")?.actionClass ?? "hold";
+
+describe("advisory==live gate disposition parity (#2190)", () => {
+  const advisoryModes: GateCheckPolicy = { aiReviewGateMode: "advisory", linkedIssueGateMode: "advisory" };
+  const liveModes: GateCheckPolicy = { aiReviewGateMode: "block", linkedIssueGateMode: "block" };
+
+  it("PASS: clean PR would-be verdict matches live enforcement", () => {
+    const advisory = cleanPass();
+    expect(advisoryWouldBeConclusion(advisory, advisoryModes)).toBe(liveConclusion(advisory, liveModes));
+    expect(liveConclusion(advisory, liveModes)).toBe("success");
+  });
+
+  it("FAIL: AI consensus defect would-be close matches live enforcement", () => {
+    const advisory = aiFail();
+    expect(advisoryWouldBeConclusion(advisory, advisoryModes)).toBe(liveConclusion(advisory, liveModes));
+    expect(liveConclusion(advisory, liveModes)).toBe("failure");
+  });
+
+  it("advisory mode posts a non-blocking conclusion while dry-run exposes the would-be verdict", () => {
+    const out = evaluateGateCheck(aiFail(), { ...advisoryModes, dryRun: true });
+    expect(out.conclusion).toBe("success");
+    expect(out.displayConclusion).toBe("failure");
+  });
+});
+
+describe("advisory==live execution boundary (#2190)", () => {
+  it("only live autonomy schedules terminal merge/close actions for the same verdict", () => {
+    const executeTerminal = vi.fn();
+    const cases = [
+      {
+        title: "passing PR",
+        expected: "merge" as const,
+        facts: agentInput({
+          conclusion: "success",
+          pr: { labels: [], mergeableState: "clean", headSha: "pass-head" },
+        }),
+        liveAutonomy: { merge: "auto" as const },
+      },
+      {
+        title: "failing PR",
+        expected: "close" as const,
+        facts: agentInput({
+          conclusion: "failure",
+          blockerTitles: ["review gate failed"],
+          pr: { labels: [], headSha: "fail-head" },
+        }),
+        liveAutonomy: { close: "auto" as const },
+      },
+    ];
+
+    for (const { title, expected, facts, liveAutonomy } of cases) {
+      const livePlan = planAgentMaintenanceActions({ ...facts, autonomy: liveAutonomy });
+      const advisoryPlan = planAgentMaintenanceActions({ ...facts, autonomy: {} });
+
+      expect(terminalClass(livePlan), title).toBe(expected);
+      expect(terminalClass(advisoryPlan), title).toBe("hold");
+      expect(advisoryPlan.some((action) => action.actionClass === "merge" || action.actionClass === "close"), title).toBe(
+        false,
+      );
+
+      for (const action of livePlan.filter((candidate) => candidate.actionClass === expected)) {
+        executeTerminal(action);
+      }
+    }
+
+    expect(executeTerminal).toHaveBeenCalledTimes(cases.length);
+    expect(executeTerminal.mock.calls.map(([action]) => action.actionClass)).toEqual(["merge", "close"]);
+  });
+});