diff --git a/frontend/src/mastra/rectification-v9-tools.ts b/frontend/src/mastra/rectification-v9-tools.ts
index 29e07e7d..b23a0f34 100644
--- a/frontend/src/mastra/rectification-v9-tools.ts
+++ b/frontend/src/mastra/rectification-v9-tools.ts
@@ -114,7 +114,7 @@ import {
   publicEvidenceItemStatus,
   resolveEvidenceQuote,
 } from "@/lib/rectification-agentic/v9/evidence-quote";
-import { projectTurnDecision } from "@/lib/rectification-agentic/v9/turn-decision";
+import { modelVisibleInference, projectTurnDecision } from "@/lib/rectification-agentic/v9/turn-decision";
 import {
   contrastPacketFromLatestResult,
   decideFromDossier,
@@ -624,8 +624,16 @@ function agentVisibleLatestProjection(
     representative_time_pre_inference: _preTime,
     house_table_pre_inference: _preHouse,
     natal_recast_pre_inference: _preNatal,
+    // R1 (TASK-rectification-grounding-20260927): the engine's own width is
+    // audit-only (BUG-593); the report width is skill_verification_report.
+    engine_indistinguishable_width_minutes: _engineWidth,
     ...rest
   } = projection;
+  // R1: the verification Markdown is kept once, in skill_verification_report;
+  // range_delivery carried a byte-identical copy.
+  const rangeDelivery = rest.range_delivery && typeof rest.range_delivery === "object" && !Array.isArray(rest.range_delivery)
+    ? (({ verification_markdown: _markdown, ...delivery }) => delivery)(rest.range_delivery as Record<string, unknown>)
+    : rest.range_delivery;
   const currentQuestion = extras.openQuestion
     ? extras.openQuestion.unrenderable === true
       ? {
@@ -647,13 +655,15 @@ function agentVisibleLatestProjection(
         probe_id: extras.openQuestion.probe_id,
       }
     : null;
+  // BUG-1057: the same slim candidate list the read-case shows.
   const compactInference = inference && typeof inference === "object" && !Array.isArray(inference)
-    ? inference as Record<string, unknown>
+    ? modelVisibleInference(inference as Record<string, unknown>)
     : null;
   const renderableChoice = extras.openQuestion?.kind === "choice"
     && extras.openQuestion.unrenderable !== true;
   return {
     ...rest,
+    ...(rangeDelivery !== undefined ? { range_delivery: rangeDelivery } : {}),
     current_question: currentQuestion,
     current_probe: null,
     inference_state: renderableChoice && compactInference
@@ -675,6 +685,17 @@ function agentVisibleLatestProjection(
   };
 }
 
+/**
+ * What rectification-offer-candidates hands the model (R1): the compare
+ * stripping applied to the offer payload. No open question rides on an offer.
+ */
+export function offerModelProjection(payload: Record<string, unknown>): Record<string, unknown> {
+  return agentVisibleLatestProjection(payload, {
+    openQuestion: null,
+    focusStatus: "not_applicable",
+  });
+}
+
 function followupPlanForParsed(
   parsed: DossierForTools,
   latest: DossierForTools["latestResult"],
@@ -2202,7 +2223,11 @@ export function createRectificationV9Tools(ctx: RectificationV9Context) {
           inputFingerprint,
           resultFingerprint: hashResult(payload),
         });
-        return payload;
+        // R1 (BUG-593 side path): the model gets the same stripped projection
+        // compare returns — no *_pre_inference, contrast packet, probes,
+        // duplicate report Markdown or engine width. The receipt fingerprint
+        // above stays over the full payload.
+        return offerModelProjection(payload);
       } catch (error) {
         await failReceipt("rectification-offer-candidates", "candidates.updated", error, { inputFingerprint });
         throw error;
diff --git a/frontend/tests/rectification-grounding-projection-20260927.test.ts b/frontend/tests/rectification-grounding-projection-20260927.test.ts
new file mode 100644
index 00000000..5378af76
--- /dev/null
+++ b/frontend/tests/rectification-grounding-projection-20260927.test.ts
@@ -0,0 +1,142 @@
+/**
+ * TASK-rectification-grounding-20260927 T5 (R1, BUG-593 side path): what
+ * compare and offer hand the model. Fixture: a real local engine response for
+ * a public AA chart (9 candidates), AGENTS §7.4.
+ *
+ * - offer used to return the full projection (~100 KB): `*_pre_inference`,
+ *   the contrast packet (31 KB) and the probe lists. It now gets the compare
+ *   stripping.
+ * - the verification Markdown is kept once (skill_verification_report);
+ *   range_delivery carried a byte-identical copy.
+ * - `engine_indistinguishable_width_minutes` is audit-only (BUG-593) and no
+ *   longer reaches the model.
+ * - batch returns the post-rescore range and whether this turn delivers.
+ * The full projection used by the case API and the stored receipts is
+ * unchanged.
+ */
+import assert from "node:assert/strict";
+import { readFileSync } from "node:fs";
+import test from "node:test";
+import {
+  CASE_ID,
+  TURN_ID,
+  USER_ID,
+  dossierFixture,
+  fakeAccounting,
+  receiptHandlers,
+} from "./rectification-v9-test-support.ts";
+import {
+  AA_EVIDENCE_ROWS,
+  AA_RANGE,
+  AA_TURNS,
+  buildGoldenLatest,
+  setFocusHandler,
+  stubGoldenFetch,
+} from "./rectification-grounding-support.ts";
+import {
+  batchRangeAfterRescore,
+  createRectificationV9Tools,
+  latestResultToolProjection,
+  offerModelProjection,
+} from "../src/mastra/rectification-v9-tools.ts";
+import { parseV9CaseDossier } from "../src/lib/rectification-agentic/v9/tool-service.ts";
+import { decideFromDossier } from "../src/lib/rectification-agentic/v9/decision-from-dossier.ts";
+
+const bytes = (value: unknown) => Buffer.byteLength(JSON.stringify(value), "utf8");
+
+const MODEL_HIDDEN_KEYS = [
+  "candidate_contrast_packet",
+  "discriminating_event_probes",
+  "event_clarification_probes",
+  "evidence_collection_probes",
+  "candidate_contrast_opportunities",
+  "dasha_agreement_pre_inference",
+  "representative_time_pre_inference",
+  "house_table_pre_inference",
+  "natal_recast_pre_inference",
+  "engine_indistinguishable_width_minutes",
+] as const;
+
+function assertModelView(view: Record<string, unknown>) {
+  for (const key of MODEL_HIDDEN_KEYS) assert.equal(key in view, false, key);
+  const report = view.skill_verification_report as { markdown?: string } | undefined;
+  assert.ok(report?.markdown && report.markdown.length > 200, "report Markdown kept once");
+  const delivery = view.range_delivery as Record<string, unknown> | undefined;
+  if (delivery) assert.equal("verification_markdown" in delivery, false);
+  const needle = JSON.stringify(report.markdown.slice(0, 80)).slice(1, -1);
+  assert.equal(JSON.stringify(view).split(needle).length - 1, 1, "Markdown appears exactly once");
+  const inference = view.inference_state as { candidates?: Array<Record<string, unknown>> } | null;
+  for (const candidate of inference?.candidates ?? []) {
+    assert.deepEqual(Object.keys(candidate), ["time", "score", "status", "cluster_range"]);
+  }
+}
+
+test("R1: offer's model view drops pre-inference, packet, probes, duplicate Markdown and engine width", async () => {
+  const { latest } = await buildGoldenLatest();
+  const parsed = parseV9CaseDossier(dossierFixture({ evidence: AA_EVIDENCE_ROWS, latestResult: latest, turns: AA_TURNS as never, candidateRange: AA_RANGE as never }))!;
+  const decision = decideFromDossier(parsed as never);
+  const full = latestResultToolProjection(parsed.latestResult as never, decision as never) as Record<string, unknown>;
+  // The full projection (case API, receipts) keeps its audit fields.
+  assert.ok("candidate_contrast_packet" in full);
+  assert.ok("engine_indistinguishable_width_minutes" in full);
+  const view = offerModelProjection({ ...full, candidate_snapshot: { is_current: true } });
+  assertModelView(view);
+  // Conclusion fields the delivery turn needs are still there.
+  for (const key of ["candidates", "representative_time", "credible_range", "session_outcome", "can_adopt", "event_fit_rate", "range_delivery"]) {
+    assert.ok(key in view, key);
+  }
+  console.log(JSON.stringify({ scope: "R1 offer model view", before_bytes: bytes(full), after_bytes: bytes(view) }));
+  assert.ok(bytes(view) < bytes(full) / 3);
+});
+
+test("R1: compare's model view (real tool, golden engine response) keeps the Markdown once and no engine width", async () => {
+  const { latest, compute } = await buildGoldenLatest();
+  const restore = stubGoldenFetch();
+  try {
+    const accounting = fakeAccounting({
+      ...receiptHandlers,
+      get_agentic_rectification_case_dossier: () => dossierFixture({
+        evidence: AA_EVIDENCE_ROWS,
+        latestResult: latest,
+        turns: AA_TURNS as never,
+        candidateRange: AA_RANGE as never,
+      }),
+      get_agentic_rectification_case_compute: () => compute,
+      set_agentic_rectification_conversation_focus: setFocusHandler,
+      refresh_agentic_rectification_vedastro_validation: () => ({ result_id: latest.result_id, decision_receipt: latest.decision_receipt }),
+    }, { fallback: () => null });
+    const tools = createRectificationV9Tools({
+      userId: USER_ID,
+      caseId: CASE_ID,
+      turnId: TURN_ID,
+      userMessage: "2000 年拿了奖",
+      accounting: accounting.client as never,
+    }) as unknown as Record<string, { execute(input: unknown): Promise<Record<string, unknown>> }>;
+    await tools["rectification-read-case"].execute({ caseId: CASE_ID });
+    const compare = await tools["rectification-compare-candidates"].execute({ caseId: CASE_ID });
+    assertModelView(compare);
+    console.log(JSON.stringify({ scope: "R1 compare model view", after_bytes: bytes(compare) }));
+  } finally {
+    restore();
+  }
+});
+
+test("T1/R3: batch range_after_rescore names the post-rescore range and whether this turn delivers", async () => {
+  const { latest } = await buildGoldenLatest();
+  const parsed = parseV9CaseDossier(dossierFixture({ evidence: AA_EVIDENCE_ROWS, latestResult: latest, candidateRange: AA_RANGE as never }))!;
+  const decision = decideFromDossier(parsed as never);
+  const withQuestion = batchRangeAfterRescore(decision as never, latest.decision_receipt, { question_id: "q" });
+  assert.deepEqual(withQuestion?.credible_range, decision.credibleRange);
+  assert.equal(withQuestion?.representative_time, decision.representativeTime ?? null);
+  assert.equal(withQuestion?.delivers_range_this_turn, false);
+  const deliverable = { ...decision, canAdopt: true, sessionOutcome: "adopt_representative" };
+  assert.equal(batchRangeAfterRescore(deliverable as never, latest.decision_receipt, null)?.delivers_range_this_turn, true);
+  assert.equal(batchRangeAfterRescore(null, latest.decision_receipt, null), null);
+  const tools = readFileSync(new URL("../src/mastra/rectification-v9-tools.ts", import.meta.url), "utf8");
+  // The receipt keeps hashing the pre-existing shape; only the returned value grows.
+  const batchReceipt = tools.indexOf("executedMethods: [...rescore.executedMethods],");
+  const batchReturn = tools.indexOf("range_after_rescore: rescore.rangeAfter");
+  assert.ok(batchReceipt > 0 && batchReturn > batchReceipt);
+  assert.match(tools.slice(batchReceipt - 120, batchReceipt), /resultFingerprint: hashResult\(projection\)/);
+  assert.match(tools, /return offerModelProjection\(payload\);/);
+});