diff --git a/.agents/plans/22-flow10-guidance-release/README.md b/.agents/plans/22-flow10-guidance-release/README.md
index ef831866..be528639 100644
--- a/.agents/plans/22-flow10-guidance-release/README.md
+++ b/.agents/plans/22-flow10-guidance-release/README.md
@@ -66,3 +66,20 @@ No source guard removal, reauthentication or credential fallback is authorized.
   historical policy preservation and 100/126 dispatch arithmetic were checked.
 - This document and the count draft authorize no paid starts. The final proposal
   still requires exact merged-source/artifact/evaluator bindings and approval.
+
+## First campaign: retained failure
+
+The frozen 4d08d36b / f670ad4f candidate stopped after three required attempts:
+two passed and one failed the current-handoff parser; 78 primary cells remained
+missing. Four top-level starts were consumed, leaving 122 unused and not
+transferred. The original report remains NOT VERIFIED and all failed receipts
+are preserved. No canary or publication followed.
+
+Offline reproduction identifies an evaluator false positive: the known command
+claim `passed unchanged; independent review passed with no findings` was parsed
+as one command suffix. The same retained evidence passes when only that semicolon
+becomes a period. The correction extends the existing review-sibling boundary
+rule, retaining native command/review proof, script-integrity and contradiction
+checks. Diagnostic acceptance is not qualification evidence and does not
+complete or pool the stopped campaign. Further real-account qualification needs
+a freshly bound approval after corrected evaluator preparation.
diff --git a/evals/delivery-presentation.ts b/evals/delivery-presentation.ts
index 559d801c..0ab72415 100644
--- a/evals/delivery-presentation.ts
+++ b/evals/delivery-presentation.ts
@@ -416,10 +416,11 @@ function sentenceBoundary(
 			/^,\s*host\b/i.test(text.slice(index))
 		)
 			return { end: index, next: index, unterminatedQuote: false };
-		if (mode === "command" && character === ",") {
-			const sibling = /^,\s+and\s+(?=Independent review(?::|\s))/i.exec(
-				text.slice(index),
-			);
+		if (mode === "command" && (character === "," || character === ";")) {
+			const sibling =
+				/^(?:,\s+and\s+|;\s*)(?=Independent review(?::|\s))/i.exec(
+					text.slice(index),
+				);
 			if (sibling)
 				return {
 					end: index,
diff --git a/tests/delivery-passed-unchanged.test.ts b/tests/delivery-passed-unchanged.test.ts
index 75ad5b77..c5464054 100644
--- a/tests/delivery-passed-unchanged.test.ts
+++ b/tests/delivery-passed-unchanged.test.ts
@@ -51,6 +51,84 @@ test("passed unchanged claims both script and invocation integrity", () => {
 		},
 	]);
 });
+
+function compoundFixture(separator = "; independent review") {
+	const input = fixture();
+	input.finalText = input.finalText.replace(
+		"passed unchanged.",
+		`passed unchanged${separator} passed with no findings.`,
+	);
+	return input;
+}
+
+test("supported separators preserve command integrity and independent review evidence", () => {
+	for (const separator of [
+		"; independent review",
+		";Independent review:",
+		"; Independent review",
+		", and Independent review",
+		". Independent review",
+	]) {
+		const facts = currentHandoffFacts(
+			`${gate} passed unchanged${separator} passed with no findings.`,
+			[gate],
+		);
+		expect(facts.unsupported).toEqual([]);
+		expect(facts.observations).toEqual([
+			{
+				command: gate,
+				exitCode: 0,
+				qualification: "claimed-pass",
+				integrity: "script-and-invocation-unchanged",
+			},
+		]);
+		expect(facts.independentReview).toEqual([
+			{ kind: "passed", findings: "none" },
+		]);
+		expect(scenario.check(compoundFixture(separator))).toEqual([]);
+	}
+});
+
+for (const suffix of [
+	"independent review passed if it ran",
+	"independent review passed and deployed",
+	"independent review passed with no findings; exited 1",
+	"independent review passed with no findings; its script changed",
+	"independent review passed with no findings; this observation does not claim a pass",
+]) {
+	test(`semicolon review composition still rejects ${suffix}`, () => {
+		const input = compoundFixture();
+		input.finalText = input.finalText.replace(
+			"independent review passed with no findings.",
+			`${suffix}.`,
+		);
+		expect(scenario.check(input)).toContain(
+			"Unsupported or conflicting current handoff assertions.",
+		);
+	});
+}
+
+for (const change of ["nonzero", "incomplete", "source", "unsubmitted"]) {
+	test(`a semicolon review claim cannot rescue ${change} native evidence`, () => {
+		const input = compoundFixture();
+		const runs = object(input.archives[0]).runs;
+		if (!Array.isArray(runs)) throw new Error("Missing native runs.");
+		const run = object(runs[0]);
+		if (!Array.isArray(run.validations) || !Array.isArray(run.reviews))
+			throw new Error("Missing native proof.");
+		const validation = run.validations
+			.map(object)
+			.find((value) => value.command === gate);
+		if (!validation) throw new Error("Missing gate.");
+		if (change === "nonzero") validation.exitCode = 1;
+		if (change === "incomplete") validation.outputComplete = false;
+		if (change === "source")
+			validation.sourceDigest = `sha256:${"b".repeat(64)}`;
+		if (change === "unsubmitted") object(run.reviews[0]).result = null;
+		expect(scenario.check(input)).not.toEqual([]);
+	});
+}
+
 for (const paths of [null, ["src/parser.mjs", "scripts/verify.mjs"]]) {
 	test(`passed unchanged requires immutable script evidence ${paths}`, () => {
 		const input = fixture();