pi-pr-review 1.17.0 → 1.17.2

package/CHANGELOG.md

@@ -1,5 +1,19 @@
 # Changelog
 
+## [1.17.2](https://github.com/10ego/pi-pr-review/compare/v1.17.1...v1.17.2) (2026-08-31)
+
+
+### Bug Fixes
+
+* **review:** keep degraded comments concise ([#126](https://github.com/10ego/pi-pr-review/issues/126)) ([912e474](https://github.com/10ego/pi-pr-review/commit/912e474d04be9db504d2ea83bcd9a7c86db2ca5b))
+
+## [1.17.1](https://github.com/10ego/pi-pr-review/compare/v1.17.0...v1.17.1) (2026-08-31)
+
+
+### Bug Fixes
+
+* **review:** start batch deadline at reviewer dispatch ([#124](https://github.com/10ego/pi-pr-review/issues/124)) ([26a38af](https://github.com/10ego/pi-pr-review/commit/26a38af6559223c6403913de32026c7d2eee5d2b))
+
 ## [1.17.0](https://github.com/10ego/pi-pr-review/compare/v1.16.0...v1.17.0) (2026-08-29)
 
 

package/README.md

@@ -191,13 +191,13 @@ Example:
 }
 ```
 
-Every invocation has a host-owned monotonic 15-minute hard cap, including the two GitHub identity/lifecycle preflights, synthesis, termination grace, and reserved cleanup. The dependent preflight calls share the one invocation budget, so they cannot each add an independent command timeout before review timing begins. Defaults bound light/medium/heavy attempts to 3/6/8 minutes, a fallback attempt to 3 minutes, and the concurrent batch to 12 minutes. The complete `deadlines` object may be replaced at user scope or by a trusted project; partial, malformed, non-integer, out-of-range, or internally inconsistent objects are rejected as a unit and the last valid/default finite budget remains active. Supported inclusive ranges are: attempts 30–720 seconds, fallback 30–360 seconds, batch 60–840 seconds, synthesis 10–120 seconds, total 120–1200 seconds, TERM grace 0.1–15 seconds, cleanup reserve 1–30 seconds, and minimum useful fallback 10–120 seconds. Minimum fallback must not exceed its attempt cap, and batch + synthesis + termination grace + cleanup must fit inside total.
+Every invocation has a host-owned monotonic 15-minute hard cap, including the two GitHub identity/lifecycle preflights, parent orchestration, synthesis, termination grace, and reserved cleanup. The dependent preflight calls share the one invocation budget, so they cannot each add an independent command timeout before review timing begins. The reviewer batch window activates once, at the first reviewer dispatch: preflight and diff capture remain charged to the total cap but cannot exhaust `batchMs` before any reviewer starts. The activated batch is still truncated by the original total deadline and its synthesis/termination/cleanup reserves. Defaults bound light/medium/heavy attempts to 3/6/8 minutes, a fallback attempt to 3 minutes, and the concurrent batch to 12 minutes. The complete `deadlines` object may be replaced at user scope or by a trusted project; partial, malformed, non-integer, out-of-range, or internally inconsistent objects are rejected as a unit and the last valid/default finite budget remains active. Supported inclusive ranges are: attempts 30–720 seconds, fallback 30–360 seconds, batch 60–840 seconds, synthesis 10–120 seconds, total 120–1200 seconds, TERM grace 0.1–15 seconds, cleanup reserve 1–30 seconds, and minimum useful fallback 10–120 seconds. Minimum fallback must not exceed its attempt cap, and batch + synthesis + termination grace + cleanup must fit inside total.
 
 A timed-out or retryable quota/rate-limit/capacity lane may start at most one configured fallback attempt. It starts only when at least `minimumFallbackMs` plus cleanup reserve remains; the host never changes the configured model, thinking level, or tool policy to save time. If a tier is unset, its existing nearest-configured-tier/Pi-default behavior is unchanged.
 
-On an attempt deadline the host records timeout separately from user cancellation, sends TERM to the original child, waits only `terminationGraceMs`, then sends KILL if no exit was observed and stops draining after the cleanup reserve. Partial assistant text and telemetry survive this lifecycle. The synthesis cap arms only once review work goes quiet: any turn that starts or any review tool that runs again while the cap is armed defers it, and it re-arms from the next turn end, so early review-tool turns (for example verification discovery) cannot starve later heavy lanes. Batch/total expiry stops queued work and waiting lanes; completed and partial artifacts proceed to Markdown synthesis or deterministic lane assembly, identify every incomplete reviewer, and remain eligible for the safe body-only `COMMENT` publication path. A timed-out lane is never reported as `NO FINDINGS` or full coverage, and a lane ended by the host total or synthesis deadline is disclosed with that kind (`deadline_expired`) instead of being mistaken for its own attempt deadline expiring. Host lane artifacts are authoritative for completeness in both directions: a false assistant completion claim cannot upgrade incomplete lanes, and a paraphrased or omitted `Lane completeness` line cannot downgrade a host-complete batch away from the concise renderer.
+On an attempt deadline the host records timeout separately from user cancellation, sends TERM to the original child, waits only `terminationGraceMs`, then sends KILL if no exit was observed and stops draining after the cleanup reserve. Partial assistant text and telemetry survive this lifecycle. The synthesis cap arms only once review work goes quiet: any turn that starts or any review tool that runs again while the cap is armed defers it, and it re-arms from the next turn end, so early review-tool turns (for example verification discovery) cannot starve later heavy lanes. Batch/total expiry stops queued work and waiting lanes; completed and partial artifacts proceed to Markdown synthesis or deterministic lane assembly, identify every incomplete reviewer, and remain eligible for the concise body-only `COMMENT` publication path. A timed-out lane is never reported as `NO FINDINGS` or full coverage, and a lane ended by the host total or synthesis deadline is disclosed with that kind (`deadline_expired`) instead of being mistaken for its own attempt deadline expiring. Host lane artifacts are authoritative for completeness in both directions: a false assistant completion claim cannot upgrade incomplete lanes, and a paraphrased or omitted `Lane completeness` line cannot downgrade a host-complete batch away from the concise renderer.
 
-Initial operating targets are ordinary-review p50 ≤ 6 minutes and p95 ≤ 12 minutes, and large-review p50 ≤ 10 minutes and p95 ≤ 14 minutes, with the 15-minute hard cap authoritative. Invocation telemetry starts before GitHub preflight and records configured deadline source/caps, termination grace, cleanup reserve, and active wall time; batch details record first event/output timing, lifecycle counts, configured and effective batch-truncated attempt deadlines, fallback starts/budget rejections, external total/synthesis deadline expiries per lane, and termination grace/escalation data. These are initial production targets, not a promise that every provider completes before its host deadline.
+Initial operating targets are ordinary-review p50 ≤ 6 minutes and p95 ≤ 12 minutes, and large-review p50 ≤ 10 minutes and p95 ≤ 14 minutes, with the 15-minute hard cap authoritative. Invocation telemetry starts before GitHub preflight and records configured deadline source/caps, termination grace, cleanup reserve, and active wall time; batch details record invocation time consumed before each attempt, batch/total time remaining at dispatch, first event/output timing, lifecycle counts, configured and effective batch-truncated attempt deadlines, fallback starts/budget rejections, external total/synthesis deadline expiries per lane, and termination grace/escalation data. These are initial production targets, not a promise that every provider completes before its host deadline.
 
 Ordinary reviewer output is reconstructed from every text part of the authoritative final assistant message in order. A zero process exit is not enough to mark a lane complete: completion also requires a terminal `stop` and the expected lane sections (or an explicitly emitted `NO FINDINGS.` where that response is allowed). Empty success is never synthesized into `NO FINDINGS.`; length-limited, malformed, timed-out, and failed attempts retain their raw text with an explicit lifecycle. Batch details include raw text, attempt artifacts, lifecycle counts and elapsed totals, while the displayed batch remains in deterministic input order.
 
@@ -227,11 +227,11 @@ Publishing is off by default.
 
 The extension owns normal publishing. Before review execution it captures repository, hostname, PR number/title, reviewed head, lifecycle state, posting/stale authority, and invocation identity independently of assistant text. After synthesis, it caches one validated completed review (the host-owned canonical artifact) per repository and PR in the current Pi session. `autoPostReviews` and `--comment` publish that cached review after completion; `--no-comment` suppresses publication for the run.
 
-Review semantics are Markdown-first. A deterministic tolerant parser normalizes line endings, extracts complete findings when possible, and ignores heading-like text inside CommonMark fenced-code and HTML-block contexts. Every degraded synthesis — ambiguous, partially parsed, incomplete lane coverage, or absent terminal synthesis — publishes one deterministic host-rendered `COMMENT` body with generic code-review labels: a `Coverage` section with the exact lane lifecycle disclosure, a `Findings` section with host-formatted parsed findings (never a clean-review claim when nothing parsed), and the complete original synthesis plus every retained lane artifact preserved verbatim under `Retained synthesis` / `Retained lane output` with heading levels shifted so they nest. A contradictory completion claim inside retained text is reconciled to the host verdict. Reserved markers are sanitized, size limits are enforced, and the canonical marker is appended only by host code. Optional formatting repair is never required and can never suppress the degraded fallback.
+Review semantics are Markdown-first. A deterministic tolerant parser normalizes line endings, extracts complete findings when possible, and ignores heading-like text inside CommonMark fenced-code and HTML-block contexts. Every degraded synthesis — ambiguous, partially parsed, incomplete lane coverage, or absent terminal synthesis — is retained internally as one deterministic host-rendered artifact with generic code-review labels: a `Coverage` section with the exact lane lifecycle disclosure, a `Findings` section with host-formatted parsed findings (never a clean-review claim when nothing parsed), and the complete original synthesis plus every retained lane artifact preserved under `Retained synthesis` / `Retained lane output`. GitHub publication never dumps that artifact: it emits a concise `COMMENT` with a short incomplete-coverage or unstructured-output warning, validated inline findings, and summary-only `Other Notes`. Reserved markers are sanitized, size limits are enforced, and the canonical marker is appended only by host code. Optional formatting repair is never required and can never suppress the degraded fallback.
 
 You can publish the cache later with `/pr-review-publish 123`, or directly ask the agent to “post the inline review,” “post it as an inline review,” or “publish the review for PR #123.” The extension handles that request directly before an agent turn. `/pr-review-publish` and a matching direct request publish only the cache; they never start or rerun review agents. Unnumbered direct requests select the latest cached review for the current repository. Only fresh interactive/RPC input can use the direct path.
 
-Every authorized publish path builds one GitHub review payload and sends at most one review `POST`; it never submits `REQUEST_CHANGES` or retries a rejected write with a fallback POST. Fully parsed Markdown with one complete retained artifact for every host-registered dispatch may emit a gated `APPROVE` through the same host binding, priority, lifecycle, self-author, and stale checks as strict JSON. Host lane evidence is authoritative for completeness: a paraphrased or omitted `All requested lanes completed.` disclosure cannot downgrade a host-complete batch, and a false disclosure cannot upgrade incomplete or missing lane coverage; the assistant's canonical line is consulted only when no batch evidence exists. Verdict fields outside the document preamble, hidden in CommonMark code/HTML blocks, or inside lazy container continuations, severity-tagged headings outside `Findings`, incomplete or missing host lane coverage, malformed or unsafe output, and lane-fallback artifacts remain `COMMENT` reviews; their safely parsed findings keep inline placement, while unparsable output publishes body-only. Cache restore reclassifies persisted lane output and requires the exact host-recorded artifact key set, with every complete artifact matching the frozen invocation generation, tier, and minor-hygiene contract, and re-synthesizes the retained raw text under that frozen binding before retaining approval eligibility. For a current, open PR, the first 50 eligible P0–P3 findings with valid, unique diff anchors are inline. The concise top-level body starts with the verdict, points readers to inline findings, and places nits, off-diff findings, unavailable diff metadata, duplicate anchors, and overflow under `Other Notes`; overview, verification, strengths, and transport diagnostics stay out of the public summary. The complete original Markdown remains retained internally and is never substituted for a fully parsed concise review. If the concise body plus its canonical marker exceeds GitHub's limit, publication fails closed rather than dumping the full report. Stale or authorized non-open reviews are body-only. A stale approval additionally requires `allowStaleApprovals: true`.
+Every authorized publish path builds one GitHub review payload and sends at most one review `POST`; it never submits `REQUEST_CHANGES` or retries a rejected write with a fallback POST. Fully parsed Markdown with one complete retained artifact for every host-registered dispatch may emit a gated `APPROVE` through the same host binding, priority, lifecycle, self-author, and stale checks as strict JSON. Host lane evidence is authoritative for completeness: a paraphrased or omitted `All requested lanes completed.` disclosure cannot downgrade a host-complete batch, and a false disclosure cannot upgrade incomplete or missing lane coverage; the assistant's canonical line is consulted only when no batch evidence exists. Verdict fields outside the document preamble, hidden in CommonMark code/HTML blocks, or inside lazy container continuations, severity-tagged headings outside `Findings`, incomplete or missing host lane coverage, malformed or unsafe output, and lane-fallback artifacts remain `COMMENT` reviews; their safely parsed findings keep inline placement, while unparsable output publishes body-only. Cache restore reclassifies persisted lane output and requires the exact host-recorded artifact key set, with every complete artifact matching the frozen invocation generation, tier, and minor-hygiene contract, and re-synthesizes the retained raw text under that frozen binding before retaining approval eligibility. For a current, open PR, the first 50 eligible P0–P3 findings with valid, unique diff anchors are inline. The concise top-level body starts with the verdict, points readers to inline findings, and places nits, off-diff findings, unavailable diff metadata, duplicate anchors, and overflow under `Other Notes`; overview, verification, strengths, and transport diagnostics stay out of the public summary. The complete original Markdown and retained lane evidence remain internal and are never substituted for a concise GitHub review, including degraded, incomplete, cached, direct, and stale `COMMENT` publication. If the concise body plus its canonical marker exceeds GitHub's limit, publication fails closed rather than dumping the full report. Stale or authorized non-open reviews are body-only. A stale approval additionally requires `allowStaleApprovals: true`.
 
 All publication paths apply host-enforced safety gates: captured posting authority, repository and requested-PR binding, reviewed/current-head and stale policy, bounded bodies and payloads, draft and lifecycle checks, non-open authorization, authenticated-identity same-head duplicate detection, and a final head check. Fully or partially parsed review paths additionally enforce safe inline locations, including degraded incomplete-lane reviews whose findings still parse. Raw and lane-assembled fallbacks without parsed findings are body-only; every degraded path is `COMMENT`-only, and assistant text cannot select event, commit, repository, hostname, API path, or inline anchors. Unknown or invalid host states fail closed before a write.
 

package/extensions/pr-review-subagent.ts

@@ -56,7 +56,7 @@ import {
 	type ReviewLaneLifecycle,
 } from "../lib/pr-review-artifacts.ts";
 import { runWithConcurrency } from "../lib/pr-review-concurrency.ts";
-import { attemptDeadline, fallbackBudget, type ReviewBudget } from "../lib/pr-review-deadlines.ts";
+import { activateReviewBatch, attemptDeadline, fallbackBudget, type ReviewBudget } from "../lib/pr-review-deadlines.ts";
 import { buildExtractionSystemPrompt, buildExtractionTask, MAX_EXTRACTION_OUTPUT_BYTES } from "../lib/pr-review-extract.ts";
 import { loadReviewContext } from "../lib/pr-review-context.ts";
 import {
@@ -910,6 +910,12 @@ interface ModelAttemptReport {
 	forcedTermination?: boolean;
 	deadlineMs?: number;
 	configuredDeadlineMs?: number;
+	/** Invocation time already consumed before this model attempt was scheduled. */
+	budgetElapsedBeforeAttemptMs?: number;
+	/** Remaining absolute reviewer-batch window at scheduling time; may be negative. */
+	batchRemainingBeforeAttemptMs?: number;
+	/** Remaining total invocation window at scheduling time. */
+	totalRemainingBeforeAttemptMs?: number;
 }
 
 interface SubagentPassResult {
@@ -986,16 +992,32 @@ async function runSubagentAttempt(
 	onText?: (text: string) => void,
 	beforeSpawn?: () => boolean,
 	budget?: ReviewBudget,
-): Promise<{ result: RunResult; notice: string; elapsedMs: number; deadlineMs?: number }> {
+): Promise<{
+	result: RunResult;
+	notice: string;
+	elapsedMs: number;
+	deadlineMs?: number;
+	budgetElapsedBeforeAttemptMs?: number;
+	batchRemainingBeforeAttemptMs?: number;
+	totalRemainingBeforeAttemptMs?: number;
+}> {
 	let tmp: { dir: string; filePath: string } | undefined;
 	const startedAt = monotonicNow();
+	const budgetElapsedBeforeAttemptMs = budget ? startedAt - budget.startedAtMs : undefined;
+	const batchRemainingBeforeAttemptMs = budget ? budget.batchDeadlineMs - startedAt : undefined;
+	const totalRemainingBeforeAttemptMs = budget ? budget.totalDeadlineMs - startedAt : undefined;
 	const deadlineAtMs = budget
 		? attemptDeadline(budget, pass.tier, attempt.kind === "fallback", () => startedAt)
 		: undefined;
 	const deadlineMs = deadlineAtMs === undefined ? undefined : Math.max(0, deadlineAtMs - startedAt);
 	try {
 		if (deadlineAtMs !== undefined && deadlineAtMs <= startedAt) {
-			throw new Error("Review attempt timed out before it could start within the batch deadline.");
+			throw new Error(
+				`Review attempt timed out before it could start within the batch deadline ` +
+				`(invocation elapsed ${Math.round(budgetElapsedBeforeAttemptMs!)}ms; ` +
+				`batch remaining ${Math.round(batchRemainingBeforeAttemptMs!)}ms; ` +
+				`total remaining ${Math.round(totalRemainingBeforeAttemptMs!)}ms).`,
+			);
 		}
 		const args = buildReviewBaseArgs();
 		if (attempt.spec) args.push("--model", attempt.spec);
@@ -1064,13 +1086,24 @@ async function runSubagentAttempt(
 		fileBackedAccessObserved = !!pass.fileBackedRequiredReads?.length &&
 			pass.fileBackedRequiredReads.every((range) => completedFileBackedReads.has(`${range.offset}:${range.limit}`));
 		result.fileBackedAccessObserved = fileBackedAccessObserved;
-		return { result, notice: noticeForAttempt(pass.tier, attempt), elapsedMs: monotonicNow() - startedAt, deadlineMs };
+		return {
+			result,
+			notice: noticeForAttempt(pass.tier, attempt),
+			elapsedMs: monotonicNow() - startedAt,
+			deadlineMs,
+			budgetElapsedBeforeAttemptMs,
+			batchRemainingBeforeAttemptMs,
+			totalRemainingBeforeAttemptMs,
+		};
 	} catch (e) {
 		return {
 			result: { text: "", exitCode: 1, stderr: "", errorMessage: errMessage(e), toolElapsedMs: 0, toolCallCount: 0 },
 			notice: noticeForAttempt(pass.tier, attempt),
 			elapsedMs: monotonicNow() - startedAt,
 			deadlineMs,
+			budgetElapsedBeforeAttemptMs,
+			batchRemainingBeforeAttemptMs,
+			totalRemainingBeforeAttemptMs,
 		};
 	} finally {
 		if (tmp) {
@@ -1219,6 +1252,9 @@ function retainPassArtifact(pass: SubagentPassRequest, result: SubagentPassResul
 			forcedTermination: attempt.forcedTermination,
 			deadlineMs: attempt.deadlineMs,
 			configuredDeadlineMs: attempt.configuredDeadlineMs,
+			budgetElapsedBeforeAttemptMs: attempt.budgetElapsedBeforeAttemptMs,
+			batchRemainingBeforeAttemptMs: attempt.batchRemainingBeforeAttemptMs,
+			totalRemainingBeforeAttemptMs: attempt.totalRemainingBeforeAttemptMs,
 		})),
 		fallbackUsed: result.fallbackUsed,
 		elapsedMs: result.elapsedMs,
@@ -1263,7 +1299,15 @@ async function runSubagentPass(
 			fallbackBudgetRejected = true;
 			break;
 		}
-		const { result, notice, elapsedMs, deadlineMs } = await runSubagentAttempt(
+		const {
+			result,
+			notice,
+			elapsedMs,
+			deadlineMs,
+			budgetElapsedBeforeAttemptMs,
+			batchRemainingBeforeAttemptMs,
+			totalRemainingBeforeAttemptMs,
+		} = await runSubagentAttempt(
 			config,
 			ctx,
 			pass,
@@ -1320,6 +1364,9 @@ async function runSubagentPass(
 			configuredDeadlineMs: budget
 				? (attempt.kind === "fallback" ? budget.config.fallbackAttemptMs : budget.config.attemptMs[tier])
 				: undefined,
+			budgetElapsedBeforeAttemptMs,
+			batchRemainingBeforeAttemptMs,
+			totalRemainingBeforeAttemptMs,
 		});
 
 		if (lifecycle === "complete") {
@@ -1767,6 +1814,7 @@ export default function registerPrReviewSubagents(
 			});
 			const artifactPublisher = loopCoordinator.createArtifactPublisher(lease, ctx);
 			const config = loadConfig(ctx);
+			const reviewBudget = lease.budget ? activateReviewBatch(lease.budget) : undefined;
 			const result = await runSubagentPass(
 				config,
 				ctx,
@@ -1785,7 +1833,7 @@ export default function registerPrReviewSubagents(
 				executionSignal,
 				(text) => onUpdate?.({ content: [{ type: "text", text }] }),
 				() => loopCoordinator.isLeaseActive(lease, ctx),
-				lease.budget,
+				reviewBudget,
 			);
 
 			const warnings = [...thinkingWarnings(config, [tier]), ...(lease.budget?.warnings ?? [])];
@@ -1978,6 +2026,7 @@ export default function registerPrReviewSubagents(
 			// reviewer may run concurrently, and no caller/model tuning knob exists.
 			const reviewerConcurrency = passes.length;
 			const config = loadConfig(ctx);
+			const reviewBudget = lease.budget ? activateReviewBatch(lease.budget) : undefined;
 			const batchStartedAt = monotonicNow();
 
 			const tierPriority: Record<Tier, number> = { heavy: 0, medium: 1, light: 2 };
@@ -1993,7 +2042,7 @@ export default function registerPrReviewSubagents(
 					executionSignal,
 					undefined,
 					() => loopCoordinator.isLeaseActive(lease, ctx),
-					lease.budget,
+					reviewBudget,
 				);
 				const endOffsetMs = monotonicNow() - batchStartedAt;
 				result.startOffsetMs = startOffsetMs;

package/extensions/review-table.ts

@@ -432,6 +432,16 @@ type ReviewPublicationOrigin =
 	| { readonly kind: "publish-command"; readonly stalePolicy: "frozen" | "allow-stale" }
 	| { readonly kind: "direct-request" };
 
+function conciseDegradedPublicationNotice(record: CompletedReviewRecord): string | undefined {
+	if (record.completeness === "incomplete") {
+		return "> [!WARNING]\n> Review coverage was incomplete. This COMMENT is not evidence of a clean review.";
+	}
+	if (record.synthesisQuality !== undefined && record.synthesisQuality !== "fully_parsed") {
+		return "> [!WARNING]\n> The review output was not fully structured. This COMMENT is not evidence of a clean review.";
+	}
+	return undefined;
+}
+
 async function publishCompletedReview(
 	record: CompletedReviewRecord,
 	origin: ReviewPublicationOrigin,
@@ -459,8 +469,9 @@ async function publishCompletedReview(
 		ctx.ui.notify("PR review was not posted: cached review artifact is missing pr.head_sha", "error");
 		return undefined;
 	}
-	const useDegradedPublicationBody = typeof record.publicationBody === "string" &&
-		(record.synthesisQuality !== "fully_parsed" || record.completeness === "incomplete");
+	const degradedPublication = record.completeness === "incomplete" ||
+		(record.synthesisQuality !== undefined && record.synthesisQuality !== "fully_parsed");
+	const publicationNotice = conciseDegradedPublicationNotice(record);
 	const result = await publishPullReview({
 		cwd: ctx.cwd,
 		prNumber: record.invocation.prNumber,
@@ -471,21 +482,18 @@ async function publishCompletedReview(
 		approveMaxPriorityLevel: record.invocation.approveMaxPriorityLevel,
 		expectedRepository: record.repository,
 		review: record.review,
-		...(useDegradedPublicationBody ? { publicationBody: record.publicationBody } : {}),
-		// Fully parsed reviews publish only the concise host-rendered body plus
-		// inline findings. If that payload cannot fit, publication fails closed;
-		// retained raw synthesis is never substituted as a public report dump.
-		// A degraded synthesis stays COMMENT-only, but its safely parsed findings
-		// still earn inline placement; only unparsable output is forced body-only.
+		...(publicationNotice ? { publicationNotice } : {}),
+		// Every review publishes only the concise host-rendered body plus validated
+		// inline findings. Retained raw synthesis and lane evidence remain private
+		// completion diagnostics, including for degraded/incomplete COMMENT runs.
+		// If the concise payload cannot fit, publication fails closed.
 		forceBodyOnly: record.synthesisQuality !== undefined &&
 			record.synthesisQuality !== "fully_parsed" &&
 			!(Array.isArray(record.review.findings) && record.review.findings.length > 0),
 		// A publication body identifies Markdown-derived or degraded synthesis.
 		// Enforce COMMENT again at the final publication boundary so restored
 		// canonical artifacts created by an older parser cannot inherit APPROVE.
-		forceComment: record.mergeApprovalEligible === false || useDegradedPublicationBody ||
-			(record.synthesisQuality !== undefined &&
-				(record.synthesisQuality !== "fully_parsed" || record.completeness === "incomplete")),
+		forceComment: record.mergeApprovalEligible === false || degradedPublication,
 	});
 	notifyPublishResult(result, source, ctx);
 	return result;

package/lib/pr-review-artifacts.ts

@@ -29,6 +29,12 @@ export interface ReviewLaneAttemptArtifact {
 	readonly deadlineMs?: number;
 	/** Configured tier/fallback cap before batch/total truncation. */
 	readonly configuredDeadlineMs?: number;
+	/** Invocation time already consumed before this attempt was scheduled. */
+	readonly budgetElapsedBeforeAttemptMs?: number;
+	/** Remaining reviewer-batch window at scheduling time; negative means dispatch was already late. */
+	readonly batchRemainingBeforeAttemptMs?: number;
+	/** Remaining total invocation window at scheduling time. */
+	readonly totalRemainingBeforeAttemptMs?: number;
 }
 
 export interface ExpectedReviewLane {
@@ -676,6 +682,9 @@ function attemptArtifactSnapshot(value: unknown): ReviewLaneAttemptArtifact | un
 		const forcedTermination = source.forcedTermination;
 		const deadlineMs = source.deadlineMs;
 		const configuredDeadlineMs = source.configuredDeadlineMs;
+		const budgetElapsedBeforeAttemptMs = source.budgetElapsedBeforeAttemptMs;
+		const batchRemainingBeforeAttemptMs = source.batchRemainingBeforeAttemptMs;
+		const totalRemainingBeforeAttemptMs = source.totalRemainingBeforeAttemptMs;
 		// --- Validation phase (locals only). ---
 		if (!isInteger(ordinal)) return undefined;
 		if (!isString(rawText)) return undefined;
@@ -700,6 +709,9 @@ function attemptArtifactSnapshot(value: unknown): ReviewLaneAttemptArtifact | un
 		if (!optional(forcedTermination, isBoolean)) return undefined;
 		if (!optional(deadlineMs, isFiniteNumber)) return undefined;
 		if (!optional(configuredDeadlineMs, isFiniteNumber)) return undefined;
+		if (!optional(budgetElapsedBeforeAttemptMs, isFiniteNumber)) return undefined;
+		if (!optional(batchRemainingBeforeAttemptMs, isFiniteNumber)) return undefined;
+		if (!optional(totalRemainingBeforeAttemptMs, isFiniteNumber)) return undefined;
 		// --- Snapshot phase. ---
 		const snapshot: Record<string, unknown> = {
 			ordinal,
@@ -726,6 +738,9 @@ function attemptArtifactSnapshot(value: unknown): ReviewLaneAttemptArtifact | un
 		withOptional(snapshot, "forcedTermination", forcedTermination);
 		withOptional(snapshot, "deadlineMs", deadlineMs);
 		withOptional(snapshot, "configuredDeadlineMs", configuredDeadlineMs);
+		withOptional(snapshot, "budgetElapsedBeforeAttemptMs", budgetElapsedBeforeAttemptMs);
+		withOptional(snapshot, "batchRemainingBeforeAttemptMs", batchRemainingBeforeAttemptMs);
+		withOptional(snapshot, "totalRemainingBeforeAttemptMs", totalRemainingBeforeAttemptMs);
 		return Object.freeze(snapshot) as ReviewLaneAttemptArtifact;
 	} catch {
 		// A throwing getter is a malformed attempt, never a crash at this boundary.

package/lib/pr-review-deadlines.ts

@@ -105,6 +105,8 @@ export function resolveReviewDeadlines(user: unknown, project?: unknown): Deadli
 
 export interface ReviewBudget {
 	readonly startedAtMs: number;
+	/** First reviewer dispatch time. Undefined until the batch window is activated. */
+	readonly batchStartedAtMs?: number;
 	readonly totalDeadlineMs: number;
 	readonly batchDeadlineMs: number;
 	readonly config: ReviewDeadlineConfig;
@@ -130,6 +132,37 @@ export function createReviewBudget(
 	});
 }
 
+const activatedBatchBudgets = new WeakMap<ReviewBudget, ReviewBudget>();
+
+/**
+ * Start the reviewer-batch window once, at the first reviewer dispatch.
+ *
+ * PR binding, diff capture, and parent orchestration remain charged to the total
+ * invocation deadline, but no longer consume `batchMs` before a reviewer exists.
+ * The original synthesis and termination reserves remain hard upper bounds.
+ */
+export function activateReviewBatch(
+	budget: ReviewBudget,
+	now: MonotonicNow = monotonicNow,
+): ReviewBudget {
+	if (budget.batchStartedAtMs !== undefined) return budget;
+	const existing = activatedBatchBudgets.get(budget);
+	if (existing) return existing;
+	const batchStartedAtMs = now();
+	const terminationReserveMs = budget.config.terminationGraceMs + budget.config.cleanupReserveMs;
+	const batchDeadlineMs = Math.min(
+		batchStartedAtMs + budget.config.batchMs,
+		budget.totalDeadlineMs - budget.config.synthesisMs - terminationReserveMs,
+	);
+	const activated = Object.freeze({
+		...budget,
+		batchStartedAtMs,
+		batchDeadlineMs,
+	});
+	activatedBatchBudgets.set(budget, activated);
+	return activated;
+}
+
 export function attemptDeadline(
 	budget: ReviewBudget,
 	tier: ReviewTier,

package/lib/pr-review-markdown.ts

@@ -840,9 +840,9 @@ export function synthesizeReviewArtifact(input: {
 		}
 		return reasons.length > 0 ? reasons : ["terminal synthesis was not structurally parseable; preserved as body-only Markdown"];
 	})();
-	// Markdown is the durable semantic product. A fully parsed complete synthesis
-	// publishes verbatim; every degraded synthesis publishes the deterministic
-	// host-rendered body so labels stay readable while all content is retained.
+	// Markdown is the durable semantic product. Keep the complete deterministic
+	// body for local rendering, cache diagnostics, and extraction. GitHub
+	// publication independently renders a concise host summary for every quality.
 	const body = quality === "fully_parsed"
 		? safeReviewBody(raw)
 		: buildDegradedReviewBody({

package/lib/pr-review-publish.ts

@@ -1257,7 +1257,11 @@ function findingAnchor(finding: ReviewFindingLike): string | undefined {
 	return `${path}:${side}:${start}:${end}`;
 }
 
-export function buildReviewSummary(review: ReviewLike, inlineComments: PublishComment[] = []): string {
+export function buildReviewSummary(
+	review: ReviewLike,
+	inlineComments: PublishComment[] = [],
+	publicationNotice?: string,
+): string {
 	const findings = Array.isArray(review.findings) ? review.findings : [];
 	const inlineAnchors = new Map<string, number>();
 	for (const comment of inlineComments) {
@@ -1281,6 +1285,7 @@ export function buildReviewSummary(review: ReviewLike, inlineComments: PublishCo
 			? "Approve"
 			: "Comment";
 	const lines = [`**Verdict:** ${verdict}`];
+	if (publicationNotice?.trim()) lines.push("", publicationNotice.trim());
 	if (inlineComments.length > 0) {
 		lines.push("", "See the inline review comments for the primary findings.");
 	}
@@ -1503,6 +1508,7 @@ function buildLosslessReviewPayload(input: {
 	changedFiles?: readonly ChangedFileLike[];
 	bodyPreamble?: string;
 	bodyOverride?: string;
+	publicationNotice?: string;
 	diagnostics?: readonly string[];
 	event?: ReviewEventType;
 }): { payload?: PullReviewPayload; diagnostics: string[]; errors: string[] } {
@@ -1520,6 +1526,9 @@ function buildLosslessReviewPayload(input: {
 	if (input.bodyPreamble && containsReservedReviewMarker(input.bodyPreamble)) {
 		return { diagnostics, errors: ["publication preamble contains a reserved pi-pr-review marker"] };
 	}
+	if (input.publicationNotice && containsReservedReviewMarker(input.publicationNotice)) {
+		return { diagnostics, errors: ["publication notice contains a reserved pi-pr-review marker"] };
+	}
 	const selected = input.allowInlineComments
 		? selectInlineComments(input.review, input.changedFiles ?? [])
 		: { comments: [], diagnostics: [], errors: [] };
@@ -1529,7 +1538,10 @@ function buildLosslessReviewPayload(input: {
 	// Inline-placement diagnostics remain available to the host notification,
 	// while every affected finding is retained under Other Notes. Do not expose
 	// transport diagnostics as if they were review findings.
-	let content = input.bodyOverride?.trim() || buildReviewSummary(input.review, selected.comments);
+	let content = input.bodyOverride?.trim() || buildReviewSummary(input.review, selected.comments, input.publicationNotice);
+	if (input.bodyOverride?.trim() && input.publicationNotice?.trim()) {
+		content = `${content}\n\n${input.publicationNotice.trim()}`;
+	}
 	if (input.bodyPreamble?.trim()) content = `${input.bodyPreamble.trim()}\n\n${content}`;
 	const marker = canonicalReviewMarker(markerHeadSha);
 	const bodyError = validateReviewBody(content);
@@ -2148,8 +2160,10 @@ export async function publishPullReview(input: {
 	approveMaxPriorityLevel?: ApproveMaxPriorityLevel;
 	expectedRepository?: RepositoryBinding;
 	review: ReviewLike;
-	/** Host-sanitized original synthesis retained when structured extraction is partial or absent. */
+	/** Compatibility-only body override. The extension lifecycle keeps retained synthesis private. */
 	publicationBody?: string;
+	/** Short host-owned warning included in the concise summary. */
+	publicationNotice?: string;
 	/** Prevent uncertain/raw synthesis from selecting APPROVE or inline anchors. */
 	forceBodyOnly?: boolean;
 	/** Prevent partially trusted synthesis from selecting a merge-relevant event. */
@@ -2166,6 +2180,7 @@ export async function publishPullReview(input: {
 		expectedRepository,
 		review,
 		publicationBody,
+		publicationNotice,
 		forceBodyOnly = false,
 		forceComment = false,
 	} = input;
@@ -2266,6 +2281,7 @@ export async function publishPullReview(input: {
 			allowInlineComments,
 			changedFiles,
 			...(publicationBody ? { bodyOverride: publicationBody } : {}),
+			...(publicationNotice ? { publicationNotice } : {}),
 			...(isApprove ? { event: APPROVE_EVENT } : {}),
 			...(changedFileLookupFailed ? { diagnostics: [CHANGED_FILE_LOOKUP_DIAGNOSTIC] } : {}),
 			...(headPlan.stale

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pi-pr-review",
-  "version": "1.17.0",
+  "version": "1.17.2",
   "description": "Parallel AI code review for GitHub pull requests in the Pi coding agent, with model-agnostic tiered subagents, structured findings, optional verification, and host-gated COMMENT or qualified APPROVE publishing.",
   "keywords": [
     "pi-package",