diff --git a/tests/runner-e2e/OPERATIONS.md b/tests/runner-e2e/OPERATIONS.md
index 1dc7b6c3258..7d0015d3a6a 100644
--- a/tests/runner-e2e/OPERATIONS.md
+++ b/tests/runner-e2e/OPERATIONS.md
@@ -156,16 +156,17 @@ keys. The server environment strips the gateway token and upstream credentials.
 
 Every upstream attempt reserves its maximum estimated charge in an fsynced
 journal before dispatch. Requests, retries, and cancelled calls retain distinct
-IDs. Braintrust gateway requests propagate the reserved request ID, campaign,
+IDs. Braintrust gateway requests include the reserved request ID, campaign,
 execution, attempt and phase as allowlisted `paperclip_eval_*` metadata in the
-SDK-compatible parent context. The exact context is retained in the private
-receipt before dispatch, so a request canceled before response headers can
-still be looked up by `metadata.paperclip_eval_request_id` in project logs.
-This context contains no provider credential, prompt or response, and does not
-create an extra model call or a synthetic parent span. Direct provider routes
-receive no Braintrust context. Older project-only receipts cannot acquire this
-identity retroactively; time or session matches remain candidate evidence.
-A matching logged span does not itself prove complete usage or final billing.
+SDK-compatible parent context. The private receipt saves that context before
+dispatch. A live hosted probe did not retain the propagated metadata, so this
+header alone does not prove lookup by request ID after a pre-header cancellation.
+The returned `x-bt-span-id`, when present, identifies the Braintrust log row's
+`id`; the stored `span_id` and `root_span_id` can differ. The private receipt's
+historical field name `gatewaySpanId` continues to hold that returned row ID.
+Direct provider routes receive no Braintrust context. Older receipts cannot
+acquire a missing gateway row ID from a time or session match.
+A matching logged row does not itself prove complete usage or final billing.
 Unknown usage keeps the reservation held. One shared writer serializes
 admission across cells, so concurrent requests cannot each spend the same remaining
 allowance. Campaign deadline, request count, and in-flight count are also enforced.
@@ -735,6 +736,24 @@ Braintrust can double-count native/log observations; use the deduplicated gatewa
 ledger as the estimate authority. Private indexing receipts contain only counts,
 identities and verification status, without trace contents or credentials.
 
+Scheduled finalization separately verifies every non-infrastructure request in
+the campaign journal against its private gateway receipt. It queries Braintrust
+by row `id`, checks the expected provider and model, and retains the actual span
+and trace IDs in `gateway-indexing-receipt.json`. Setup, failed and canceled
+requests remain in the denominator. Missing receipts, missing row IDs, duplicate
+row attribution, conflicting records and query errors fail the gateway coverage
+gate. A receipt or journal change during inspection also prevents verification.
+The file reports whether campaign metadata was retained, independently of the
+receipt-to-row match. It contains no prompts, responses or credentials.
+
+Each lookup batches at most 100 rows, makes at most three rounds, and has a
+45-second deadline. It performs no inference, writes no vendor records, changes
+no journal amounts and releases no holds. The findings include
+`gatewayIndexVerification`; a missing match is `gateway_indexing_incomplete`.
+This verifies stored identity and coverage, not provider billing or automatic
+nesting of gateway spans under native attempt traces. Those remain separate
+qualification requirements.
+
 ## Inspection and limits
 
 Retained deterministic grades can be imported into a private Braintrust experiment
diff --git a/tests/runner-e2e/braintrust-gateway-indexing.test.ts b/tests/runner-e2e/braintrust-gateway-indexing.test.ts
new file mode 100644
index 00000000000..cbc7096b04a
--- /dev/null
+++ b/tests/runner-e2e/braintrust-gateway-indexing.test.ts
@@ -0,0 +1,174 @@
+import { randomUUID } from "node:crypto";
+import { mkdir, mkdtemp, readFile, rm, writeFile } from "node:fs/promises";
+import { tmpdir } from "node:os";
+import path from "node:path";
+import { afterEach, expect, it, vi } from "vitest";
+import { verifyBraintrustGatewayIndexing } from "./braintrust-gateway-indexing.js";
+import { openEvalCostLedger, readEvalCostLedger } from "./cost-ledger.js";
+import { forwardEvalRequest, type EvalGatewayReceipt, type EvalGatewayRoute } from "./eval-gateway.js";
+
+const directories: string[] = [];
+afterEach(async () => { for (const directory of directories.splice(0)) await rm(directory, { recursive: true, force: true }); });
+const config = { projectId: "11111111-1111-4111-8111-111111111111", apiKey: "private-key-canary" };
+async function fixture(count = 1, headers = true) {
+  const directory = await mkdtemp(path.join(tmpdir(), "gateway-index-test-")); directories.push(directory);
+  const ledgerPath = path.join(directory, "ledger.jsonl"), deadlineAt = Date.now() + 60_000;
+  const limits = { campaignId: "campaign", maxUsd: "1", deadlineAt: new Date(deadlineAt).toISOString(), maxRequests: 20, maxConcurrent: 1 };
+  const ledger = await openEvalCostLedger(ledgerPath, limits);
+  await mkdir(`${ledgerPath}.requests`);
+  const route: EvalGatewayRoute = { ...config, provider: "openai", mode: "braintrust", endpointName: "OPENAI_API_KEY",
+    maximumInputTokens: 1000, maximumOutputTokens: 100, timeoutMs: 2000,
+    price: { provider: "openai", model: "fixture", version: "fixture-1", sourceUrl: "https://provider.invalid/prices", inputUsdPerMillion: "2",
+      cachedReadUsdPerMillion: "1", outputUsdPerMillion: "10", cacheWrite30mUsdPerMillion: "2", cacheWrite12hUsdPerMillion: "2" } };
+  const rows: Array<Record<string, unknown>> = [], receipts: EvalGatewayReceipt[] = [];
+  for (let i = 0; i < count; i++) {
+    const rowId = randomUUID();
+    receipts.push(await forwardEvalRequest({ ledger, identity: { executionId: "hello", attemptId: "attempt", phase: i === 0 ? "setup" : "agent" }, route,
+      payload: { model: "fixture", input: "private-prompt-canary" }, deadlineAt,
+      onHeaders: () => {}, onChunk: async () => {},
+      persist: async receipt => { await writeFile(path.join(`${ledgerPath}.requests`, `${receipt.requestId}.json`), JSON.stringify(receipt)); },
+      transport: async () => {
+        if (!headers) throw new Error("private-upstream-error-canary");
+        return Response.json({ object: "response", status: "completed", id: `provider-response-${i}`, model: "fixture",
+          usage: { input_tokens: 10, input_tokens_details: { cache_write_tokens: 0, cached_tokens: 2 }, output_tokens: 5 } },
+        { headers: { "x-bt-used-endpoint": "OPENAI_API_KEY", "x-bt-span-id": rowId } });
+      },
+    }));
+    rows.push({ id: rowId, span_id: randomUUID(), root_span_id: randomUUID(), provider: "openai", model: "fixture" });
+  }
+  await ledger.close();
+  return { input: { ledgerPath, campaignId: "campaign", expected: [{ executionId: "hello", provider: "openai" as const, model: "fixture" }] },
+    ledgerPath, rows, receipts, limits, receiptFile: (index = 0) => path.join(`${ledgerPath}.requests`, `${receipts[index]!.requestId}.json`) };
+}
+
+it("joins setup and agent requests by returned row ID without adding indexed costs to the journal", async () => {
+  const f = await fixture(2), before = await readFile(f.ledgerPath, "utf8");
+  const transport = vi.fn<typeof fetch>(async () => Response.json({ data: f.rows }));
+  const result = await verifyBraintrustGatewayIndexing(config, f.input, transport);
+  expect(result).toMatchObject({ status: "verified", expectedRequests: 2, verifiedRequests: 2 });
+  expect(result.requests.map(row => row.phase)).toEqual(["setup", "agent"]);
+  expect(result.requests[0]).toMatchObject({ gatewayRowId: f.rows[0]!.id, spanId: f.rows[0]!.span_id,
+    traceId: f.rows[0]!.root_span_id, campaignMetadataPresent: false });
+  const sql = JSON.parse(String(transport.mock.calls[0]![1]!.body)).query;
+  expect(sql).toContain(`WHERE id IN ('${f.rows[0]!.id}'`);
+  expect(sql).not.toContain("WHERE span_id");
+  expect(sql).not.toMatch(/[, ]metadata[, ]|input|output|prompt|completion/);
+  expect(transport).toHaveBeenCalledTimes(1);
+  expect(await readFile(f.ledgerPath, "utf8")).toBe(before);
+  expect(JSON.stringify(result)).not.toMatch(/costUsd|private-|apiKey/);
+});
+
+it("retains a pre-header failure as missing coverage and preserves its hold", async () => {
+  const f = await fixture(1, false), transport = vi.fn<typeof fetch>();
+  const before = (await readEvalCostLedger(f.ledgerPath)).state.snapshot();
+  const result = await verifyBraintrustGatewayIndexing(config, f.input, transport);
+  expect(result).toMatchObject({ status: "incomplete", expectedRequests: 1, verifiedRequests: 0 });
+  expect(result.requests[0]?.reason).toBe("missing_gateway_row_id");
+  expect(transport).not.toHaveBeenCalled();
+  expect((await readEvalCostLedger(f.ledgerPath)).state.snapshot()).toEqual(before);
+  expect(before.heldUsd).not.toBe("0.000000000");
+});
+
+it.each(["missing", "malformed", "oversized", "wrong-request", "wrong-route", "wrong-model"])("reports a %s private receipt", async fault => {
+  const f = await fixture(), transport = vi.fn<typeof fetch>();
+  if (fault === "missing") await rm(f.receiptFile());
+  else if (fault === "malformed") await writeFile(f.receiptFile(), "private-error-canary");
+  else if (fault === "oversized") await writeFile(f.receiptFile(), "x".repeat(262_145));
+  else await writeFile(f.receiptFile(), JSON.stringify({ ...f.receipts[0],
+    ...(fault === "wrong-request" ? { requestId: randomUUID() } : fault === "wrong-route" ? { route: "direct" }
+      : { usage: { ...f.receipts[0]!.usage, model: "other" } }) }));
+  const result = await verifyBraintrustGatewayIndexing(config, f.input, transport);
+  expect(result.status).not.toBe("verified");
+  expect(transport).not.toHaveBeenCalled();
+  expect(JSON.stringify(result)).not.toContain("private-");
+});
+
+it.each(["wrong-provider", "wrong-model", "invalid-trace", "duplicate", "foreign-row"])("rejects %s indexed evidence", async fault => {
+  const f = await fixture();
+  if (fault === "wrong-provider") f.rows[0]!.provider = "anthropic";
+  if (fault === "wrong-model") f.rows[0]!.model = "other";
+  if (fault === "invalid-trace") f.rows[0]!.root_span_id = "private-token-canary";
+  if (fault === "duplicate") f.rows.push(f.rows[0]!);
+  if (fault === "foreign-row") f.rows[0]!.id = randomUUID();
+  const result = await verifyBraintrustGatewayIndexing(config, f.input, async () => Response.json({ data: f.rows }), { rounds: 1 });
+  expect(result.status).not.toBe("verified");
+  expect(result.verifiedRequests).toBe(0);
+  expect(JSON.stringify(result)).not.toContain("private-");
+});
+
+it.each(["request_id", "campaign_id", "execution_id", "attempt_id", "phase"])("rejects a conflicting %s even when other campaign metadata is absent", async field => {
+  const f = await fixture(), before = await readFile(f.ledgerPath, "utf8");
+  f.rows[0]![field] = "different-work";
+  const transport = vi.fn<typeof fetch>(async () => Response.json({ data: f.rows }));
+  const result = await verifyBraintrustGatewayIndexing(config, f.input, transport);
+  expect(result).toMatchObject({ status: "incomplete", expectedRequests: 1, verifiedRequests: 0 });
+  expect(result.requests[0]?.reason).toBe("indexed_identity_conflict");
+  expect(transport).toHaveBeenCalledTimes(1);
+  expect(await readFile(f.ledgerPath, "utf8")).toBe(before);
+});
+
+it.each(["absent", "partial", "complete"])("allows %s campaign metadata when no identity contradicts the receipt", async coverage => {
+  const f = await fixture();
+  Object.assign(f.rows[0]!, { request_id: null, campaign_id: null, execution_id: null, attempt_id: null, phase: null });
+  if (coverage !== "absent") f.rows[0]!.request_id = f.receipts[0]!.requestId;
+  if (coverage === "complete") Object.assign(f.rows[0]!, { campaign_id: "campaign", execution_id: "hello", attempt_id: "attempt", phase: "setup" });
+  const result = await verifyBraintrustGatewayIndexing(config, f.input, async () => Response.json({ data: f.rows }));
+  expect(result).toMatchObject({ status: "verified", expectedRequests: 1, verifiedRequests: 1 });
+  expect(result.requests[0]?.campaignMetadataPresent).toBe(coverage === "complete");
+});
+
+it("refuses to attribute one gateway row to two charged requests", async () => {
+  const f = await fixture(2), transport = vi.fn<typeof fetch>();
+  await writeFile(f.receiptFile(1), JSON.stringify({ ...f.receipts[1], gatewaySpanId: f.receipts[0]!.gatewaySpanId }));
+  const result = await verifyBraintrustGatewayIndexing(config, f.input, transport);
+  expect(result.requests.map(row => row.reason)).toEqual(["duplicate_gateway_row_id", "duplicate_gateway_row_id"]);
+  expect(result.status).toBe("incomplete");
+  expect(transport).not.toHaveBeenCalled();
+});
+
+it("queries delayed rows again without querying rows already found", async () => {
+  const f = await fixture(2), pause = vi.fn(async () => {});
+  const transport = vi.fn<typeof fetch>().mockResolvedValueOnce(Response.json({ data: [f.rows[0]] }))
+    .mockResolvedValueOnce(Response.json({ data: [f.rows[1]] }));
+  const result = await verifyBraintrustGatewayIndexing(config, f.input, transport, { pause });
+  expect(result.status).toBe("verified");
+  expect(transport).toHaveBeenCalledTimes(2);
+  const second = JSON.parse(String(transport.mock.calls[1]![1]!.body)).query;
+  expect(second).not.toContain(f.rows[0]!.id);
+  expect(second).toContain(f.rows[1]!.id);
+});
+
+it.each(["rejected", "malformed", "oversized", "cancelled"])("keeps %s lookup failures visible without response bodies", async fault => {
+  const f = await fixture();
+  const transport = vi.fn<typeof fetch>(async () => fault === "rejected" ? new Response("private-canary", { status: 401 })
+    : fault === "oversized" ? new Response("private-canary".repeat(100_000)) : new Response("private-canary"));
+  const result = await verifyBraintrustGatewayIndexing(config, f.input, transport, { rounds: 1,
+    ...(fault === "cancelled" ? { signal: AbortSignal.abort() } : {}) });
+  expect(result.status).toBe("unavailable");
+  expect(JSON.stringify(result)).not.toContain("private-canary");
+  if (fault === "cancelled") expect(transport).not.toHaveBeenCalled();
+});
+
+it.each(["receipt", "journal"])("rejects evidence when the %s changes during lookup", async kind => {
+  const f = await fixture();
+  const transport = vi.fn<typeof fetch>(async () => {
+    if (kind === "receipt") await writeFile(f.receiptFile(), JSON.stringify({ ...f.receipts[0], gatewaySpanId: randomUUID() }));
+    else {
+      const ledger = await openEvalCostLedger(f.ledgerPath, f.limits);
+      await ledger.reserve(randomUUID(), { executionId: "hello", attemptId: "another", phase: "setup" }, "0.01");
+      await ledger.close();
+    }
+    return Response.json({ data: f.rows });
+  });
+  const result = await verifyBraintrustGatewayIndexing(config, f.input, transport);
+  expect(result.status).toBe("unavailable");
+  expect(result.verifiedRequests).toBe(0);
+  expect(result.requests[0]?.reason).toBe(kind === "receipt" ? "receipt_changed" : "ledger_changed");
+});
+
+it("rejects a mismatched campaign or selection before lookup", async () => {
+  const f = await fixture(), transport = vi.fn<typeof fetch>();
+  await expect(verifyBraintrustGatewayIndexing(config, { ...f.input, campaignId: "other" }, transport)).rejects.toThrow("campaign mismatch");
+  await expect(verifyBraintrustGatewayIndexing(config, { ...f.input, expected: [{ ...f.input.expected[0]!, executionId: "other" }] }, transport)).rejects.toThrow("selection mismatch");
+  expect(transport).not.toHaveBeenCalled();
+});
diff --git a/tests/runner-e2e/braintrust-gateway-indexing.ts b/tests/runner-e2e/braintrust-gateway-indexing.ts
new file mode 100644
index 00000000000..ddb05282212
--- /dev/null
+++ b/tests/runner-e2e/braintrust-gateway-indexing.ts
@@ -0,0 +1,124 @@
+import { createHash } from "node:crypto";
+import { readFile, stat } from "node:fs/promises";
+import path from "node:path";
+import { setTimeout as delay } from "node:timers/promises";
+import type { EvalBraintrustConfig } from "./braintrust.js";
+import { queryBraintrustRows } from "./braintrust-query.js";
+import { readEvalCostLedger, validEvalLedgerId } from "./cost-ledger.js";
+
+type Status = "verified" | "incomplete" | "unavailable";
+interface ExpectedExecution { executionId: string; provider: "openai" | "anthropic"; model: string }
+interface RequestIndex {
+  requestId: string; executionId: string; attemptId: string; phase: string;
+  status: Status; reason?: string; receiptSha256?: string;
+  gatewayRowId?: string; spanId?: string; traceId?: string;
+  campaignMetadataPresent?: boolean;
+}
+export interface BraintrustGatewayIndexReceipt {
+  version: 1; projectId: string; campaignId: string; checkedAt: string;
+  ledgerCheckpoint: string; status: Status; expectedRequests: number; verifiedRequests: number;
+  requests: RequestIndex[];
+}
+const uuid = (value: unknown): value is string => typeof value === "string" && /^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$/i.test(value);
+const spanIdentity = (value: unknown): value is string => uuid(value) || typeof value === "string" && /^(?:[0-9a-f]{16}|[0-9a-f]{32})$/i.test(value) && !/^0+$/.test(value);
+const digest = (bytes: Buffer) => createHash("sha256").update(bytes).digest("hex");
+const object = (value: unknown): Record<string, unknown> => value && typeof value === "object" && !Array.isArray(value) ? value as Record<string, unknown> : {};
+async function readReceipt(file: string) {
+  if ((await stat(file)).size > 262_144) throw new Error("Oversized gateway receipt");
+  const bytes = await readFile(file);
+  if (bytes.length > 262_144) throw new Error("Oversized gateway receipt");
+  return { bytes, value: object(JSON.parse(bytes.toString("utf8"))) };
+}
+
+export async function verifyBraintrustGatewayIndexing(config: EvalBraintrustConfig, input: {
+  ledgerPath: string; campaignId: string; expected: ExpectedExecution[];
+}, transport: typeof fetch = fetch, options: { rounds?: number; pause?: () => Promise<void>; signal?: AbortSignal } = {}): Promise<BraintrustGatewayIndexReceipt> {
+  if (!uuid(config.projectId) || !config.apiKey?.trim() || /[\r\n]/.test(config.apiKey)
+      || !path.isAbsolute(input.ledgerPath) || !validEvalLedgerId(input.campaignId)
+      || !input.expected.length || input.expected.length > 1000 || new Set(input.expected.map(row => row.executionId)).size !== input.expected.length
+      || input.expected.some(row => !validEvalLedgerId(row.executionId) || !validEvalLedgerId(row.model) || !["openai", "anthropic"].includes(row.provider))) {
+    throw new Error("Invalid gateway index scope");
+  }
+  const rounds = options.rounds ?? 3;
+  if (!Number.isInteger(rounds) || rounds < 1 || rounds > 3) throw new Error("Invalid gateway index verification limit");
+  const ledger = await readEvalCostLedger(input.ledgerPath);
+  if (ledger.state.limits.campaignId !== input.campaignId) throw new Error("Gateway index campaign mismatch");
+  const expected = new Map(input.expected.map(row => [row.executionId, row]));
+  const modelRequests = [...ledger.state.requests].filter(([, request]) => request.identity.phase !== "infrastructure");
+  if (modelRequests.length > 1000 || modelRequests.some(([id, request]) => !uuid(id) || !expected.has(request.identity.executionId))) {
+    throw new Error("Gateway index request selection mismatch");
+  }
+  const receipt: BraintrustGatewayIndexReceipt = { version: 1, projectId: config.projectId, campaignId: input.campaignId,
+    checkedAt: new Date().toISOString(), ledgerCheckpoint: ledger.previous, status: "incomplete",
+    expectedRequests: modelRequests.length, verifiedRequests: 0, requests: [] };
+  for (const [requestId, request] of modelRequests) {
+    const row: RequestIndex = { requestId, ...request.identity, status: "incomplete" };
+    receipt.requests.push(row);
+    if (!request.finished) { row.reason = "request_in_flight"; continue; }
+    try {
+      const { bytes, value } = await readReceipt(path.join(`${input.ledgerPath}.requests`, `${requestId}.json`));
+      row.receiptSha256 = digest(bytes);
+      const planned = expected.get(request.identity.executionId)!;
+      if (value.requestId !== requestId || value.route !== "braintrust"
+          || request.observations.some(observation => observation.route !== "braintrust" || observation.provider !== planned.provider
+            || observation.model !== undefined && observation.model !== planned.model)
+          || object(value.usage).provider !== undefined && object(value.usage).provider !== planned.provider
+          || object(value.usage).model !== undefined && object(value.usage).model !== planned.model) {
+        row.reason = "receipt_identity_conflict"; continue;
+      }
+      if (!uuid(value.gatewaySpanId)) { row.reason = "missing_gateway_row_id"; continue; }
+      row.gatewayRowId = value.gatewaySpanId;
+      row.reason = "row_not_indexed";
+    } catch { row.status = "unavailable"; row.reason = "receipt_unavailable"; }
+  }
+  const rowIds = new Map<string, RequestIndex[]>();
+  for (const row of receipt.requests) if (row.gatewayRowId) rowIds.set(row.gatewayRowId, [...(rowIds.get(row.gatewayRowId) ?? []), row]);
+  for (const rows of rowIds.values()) if (rows.length !== 1) for (const row of rows) row.reason = "duplicate_gateway_row_id";
+  const signal = AbortSignal.any([AbortSignal.timeout(45_000), ...(options.signal ? [options.signal] : [])]);
+  const pause = options.pause ?? (() => delay(2_000, undefined, { signal }));
+  for (let round = 0; round < rounds && !signal.aborted; round++) {
+    const pending = receipt.requests.filter(row => row.reason === "row_not_indexed" || row.reason === "query_unavailable");
+    if (!pending.length) break;
+    for (let offset = 0; offset < pending.length && !signal.aborted; offset += 100) {
+      const batch = pending.slice(offset, offset + 100);
+      try {
+        const rows = await queryBraintrustRows(config, `SELECT id, span_id, root_span_id, metadata.provider AS provider, metadata.model AS model, metadata.paperclip_eval_request_id AS request_id, metadata.paperclip_eval_campaign_id AS campaign_id, metadata.paperclip_eval_execution_id AS execution_id, metadata.paperclip_eval_attempt_id AS attempt_id, metadata.paperclip_eval_phase AS phase FROM project_logs('${config.projectId}') WHERE id IN (${batch.map(row => `'${row.gatewayRowId}'`).join(",")}) LIMIT ${batch.length + 1}`, transport, signal);
+        if (rows.length > batch.length || rows.some(row => !batch.some(expected => expected.gatewayRowId === row.id))) throw new Error("Unexpected gateway index rows");
+        for (const row of batch) {
+          const matches = rows.filter(value => value.id === row.gatewayRowId);
+          row.status = "incomplete"; row.reason = "row_not_indexed";
+          if (!matches.length) continue;
+          const found = matches[0]!, planned = expected.get(row.executionId)!;
+          const identity = { request_id: row.requestId, campaign_id: input.campaignId,
+            execution_id: row.executionId, attempt_id: row.attemptId, phase: row.phase };
+          if (matches.length !== 1 || !spanIdentity(found.span_id) || !spanIdentity(found.root_span_id)
+              || found.model !== planned.model || found.provider !== planned.provider
+              || Object.entries(identity).some(([field, value]) => found[field] != null && found[field] !== value)) {
+            row.reason = "indexed_identity_conflict"; continue;
+          }
+          row.status = "verified"; delete row.reason;
+          row.spanId = found.span_id; row.traceId = found.root_span_id;
+          row.campaignMetadataPresent = Object.entries(identity).every(([field, value]) => found[field] === value);
+        }
+      } catch { for (const row of batch) { row.status = "unavailable"; row.reason = "query_unavailable"; } }
+    }
+    if (round + 1 < rounds && receipt.requests.some(row => ["row_not_indexed", "query_unavailable"].includes(row.reason ?? "")) && !signal.aborted) {
+      try { await pause(); } catch { break; }
+    }
+  }
+  for (const row of receipt.requests) if (row.receiptSha256) {
+    try {
+      const { bytes } = await readReceipt(path.join(`${input.ledgerPath}.requests`, `${row.requestId}.json`));
+      if (digest(bytes) !== row.receiptSha256) throw new Error("Gateway receipt changed");
+    } catch { row.status = "unavailable"; row.reason = "receipt_changed"; }
+  }
+  const after = await readEvalCostLedger(input.ledgerPath);
+  if (after.previous !== ledger.previous || after.sequence !== ledger.sequence) {
+    for (const row of receipt.requests) { row.status = "unavailable"; row.reason = "ledger_changed"; }
+  }
+  receipt.checkedAt = new Date().toISOString();
+  receipt.verifiedRequests = receipt.requests.filter(row => row.status === "verified").length;
+  receipt.status = receipt.expectedRequests > 0 && receipt.verifiedRequests === receipt.expectedRequests ? "verified"
+    : signal.aborted || receipt.requests.some(row => row.status === "unavailable") ? "unavailable" : "incomplete";
+  return receipt;
+}
diff --git a/tests/runner-e2e/braintrust-indexing.ts b/tests/runner-e2e/braintrust-indexing.ts
index 8ca5d9f11b4..f5b5f6abbbe 100644
--- a/tests/runner-e2e/braintrust-indexing.ts
+++ b/tests/runner-e2e/braintrust-indexing.ts
@@ -1,4 +1,5 @@
 import { setTimeout as delay } from "node:timers/promises";
+import { queryBraintrustRows as query } from "./braintrust-query.js";
 import { buildBraintrustExperiment, experimentEvidenceHash, type BraintrustExperimentReceipt } from "./braintrust-experiment.js";
 import { exportBraintrustAttemptParent, type EvalBraintrustConfig } from "./braintrust.js";
 import type { RunnerE2ECampaign, RunnerE2EResult } from "./types.js";
@@ -13,25 +14,6 @@ export interface BraintrustIndexReceipt {
   experiment: { id: string | null; status: IndexStatus; expectedRows: number; indexedRows?: number };
 }
 
-async function query(config: EvalBraintrustConfig, sql: string, transport: typeof fetch, signal: AbortSignal): Promise<Record<string, unknown>[]> {
-  const response = await transport("https://api.braintrust.dev/btql", {
-    method: "POST", redirect: "error", signal: AbortSignal.any([signal, AbortSignal.timeout(10_000)]),
-    headers: { authorization: `Bearer ${config.apiKey}`, "content-type": "application/json" },
-    body: JSON.stringify({ query: sql, fmt: "json" }),
-  });
-  const reader = response.body?.getReader(), chunks: Uint8Array[] = []; let size = 0;
-  try {
-    if (!response.ok) throw new Error("query_rejected");
-    while (reader) {
-      const next = await reader.read(); if (next.done) break;
-      size += next.value.length; if (size > 1_048_576) throw new Error("query_oversized");
-      chunks.push(next.value);
-    }
-    const payload = JSON.parse(Buffer.concat(chunks).toString("utf8"));
-    if (!payload || !Array.isArray(payload.data) || payload.data.some((row: unknown) => !row || typeof row !== "object" || Array.isArray(row))) throw new Error("invalid_query_result");
-    return payload.data;
-  } finally { await reader?.cancel().catch(() => {}); }
-}
 
 export function braintrustTraceIndexQuery(projectId: string, result: RunnerE2EResult): string {
   const trace = result.tracing;
diff --git a/tests/runner-e2e/braintrust-query.ts b/tests/runner-e2e/braintrust-query.ts
new file mode 100644
index 00000000000..198d6925c60
--- /dev/null
+++ b/tests/runner-e2e/braintrust-query.ts
@@ -0,0 +1,22 @@
+import type { EvalBraintrustConfig } from "./braintrust.js";
+
+export async function queryBraintrustRows(config: EvalBraintrustConfig, sql: string, transport: typeof fetch, signal: AbortSignal): Promise<Record<string, unknown>[]> {
+  const response = await transport("https://api.braintrust.dev/btql", {
+    method: "POST", redirect: "error", signal: AbortSignal.any([signal, AbortSignal.timeout(10_000)]),
+    headers: { authorization: `Bearer ${config.apiKey}`, "content-type": "application/json" },
+    body: JSON.stringify({ query: sql, fmt: "json" }),
+  });
+  const reader = response.body?.getReader(), chunks: Uint8Array[] = []; let size = 0;
+  try {
+    if (!response.ok) throw new Error("query_rejected");
+    while (reader) {
+      const next = await reader.read(); if (next.done) break;
+      size += next.value.length; if (size > 1_048_576) throw new Error("query_oversized");
+      chunks.push(next.value);
+    }
+    const payload = JSON.parse(Buffer.concat(chunks).toString("utf8"));
+    if (!payload || !Array.isArray(payload.data) || payload.data.some((row: unknown) => !row || typeof row !== "object" || Array.isArray(row))) throw new Error("invalid_query_result");
+    return payload.data;
+  } finally { await reader?.cancel().catch(() => {}); }
+}
+
diff --git a/tests/runner-e2e/scheduled-finalization.ts b/tests/runner-e2e/scheduled-finalization.ts
index 83dfe9bbdcd..5d93e4fe8be 100644
--- a/tests/runner-e2e/scheduled-finalization.ts
+++ b/tests/runner-e2e/scheduled-finalization.ts
@@ -8,6 +8,8 @@ import { openExperimentIngestion } from "./experiment-ingestion.js";
 import { collectScheduledIngestion } from "./ingestion-usage.js";
 import { importBraintrustExperiment, type BraintrustExperimentReceipt } from "./braintrust-experiment.js";
 import { verifyBraintrustIndexing } from "./braintrust-indexing.js";
+import { verifyBraintrustGatewayIndexing } from "./braintrust-gateway-indexing.js";
+import { runnerMatrix } from "./catalog.js";
 import { feedbackReviewTemplate, validateCampaignFeedbackReview, type CampaignFeedbackReview } from "./feedback-dispositions.js";
 import { expectedFixtureFeedback } from "./feedback-persistence.js";
 import { feedbackReview, summarizeCampaignFeedback, type FeedbackEvidence } from "./feedback.js";
@@ -110,8 +112,24 @@ export async function finalizeScheduledCampaign(env: NodeJS.ProcessEnv) {
     await writeFile(path.join(directory, "indexing-receipt.json"), `${JSON.stringify(receipt, null, 2)}\n`, { mode: 0o600 });
   } catch {}
   if (indexing !== "verified") findings.push("trace_or_experiment_indexing_incomplete");
+  let gatewayIndexing = "unavailable";
+  try {
+    if (!config || config.projectId !== manifest.projectId || manifest.route !== "braintrust") throw new Error("Missing gateway index scope");
+    const expected = manifest.executionIds.map(executionId => {
+      const execution = runnerMatrix.find(row => row.id === executionId);
+      if (!execution) throw new Error("Missing gateway execution");
+      const provider = execution.profile.credential === "ANTHROPIC_API_KEY" ? "anthropic" as const : "openai" as const;
+      if (!manifest.models.some(model => model.provider === provider && model.price.model === execution.profile.model)) throw new Error("Missing gateway model");
+      return { executionId, provider, model: execution.profile.model };
+    });
+    const receipt = await verifyBraintrustGatewayIndexing(config, { ledgerPath: manifest.ledgerPath,
+      campaignId: manifest.limits.campaignId, expected });
+    gatewayIndexing = receipt.status;
+    await writeFile(path.join(directory, "gateway-indexing-receipt.json"), `${JSON.stringify(receipt, null, 2)}\n`, { mode: 0o600 });
+  } catch {}
+  if (gatewayIndexing !== "verified") findings.push("gateway_indexing_incomplete");
   await writeFile(path.join(directory, "findings.json"), `${JSON.stringify({ version: 1, findings,
-    nativeIndexVerification: indexing, invoiceReconciliation: "pending",
+    nativeIndexVerification: indexing, gatewayIndexVerification: gatewayIndexing, invoiceReconciliation: "pending",
     accountingStatus: campaign.operations ? evalAccountingStatus(campaign.operations) : "unavailable",
     feedbackReviewMethod: "Only exact synthetic fixture bodies with all corroborating graders are automatically dismissed; all other feedback requires review" }, null, 2)}\n`, { mode: 0o600 });
   for (const finding of findings) console.error(`::error title=Eval operations::${finding}`);
diff --git a/tests/runner-e2e/scheduled-gateway-indexing.test.ts b/tests/runner-e2e/scheduled-gateway-indexing.test.ts
new file mode 100644
index 00000000000..f322d3e81f1
--- /dev/null
+++ b/tests/runner-e2e/scheduled-gateway-indexing.test.ts
@@ -0,0 +1,55 @@
+import { mkdir, mkdtemp, readFile, rm, writeFile } from "node:fs/promises";
+import { tmpdir } from "node:os";
+import path from "node:path";
+import { afterEach, beforeEach, expect, it, vi } from "vitest";
+import { finalizeScheduledCampaign } from "./scheduled-finalization.js";
+import { verifyBraintrustGatewayIndexing } from "./braintrust-gateway-indexing.js";
+import { runnerMatrix } from "./catalog.js";
+import { buildRunnerCampaign } from "./history.js";
+import { emptyFeedbackSummary } from "./feedback.js";
+import type { RunnerE2EResult } from "./types.js";
+
+vi.mock("./braintrust-gateway-indexing.js", () => ({ verifyBraintrustGatewayIndexing: vi.fn() }));
+vi.mock("./braintrust-indexing.js", () => ({ verifyBraintrustIndexing: vi.fn(async () => ({ status: "verified" })) }));
+vi.mock("./ingestion-usage.js", () => ({ collectScheduledIngestion: vi.fn(async () => ({ status: "complete" })) }));
+vi.mock("./experiment-ingestion.js", () => ({ openExperimentIngestion: vi.fn(async () => ({ transport: vi.fn(), close: async () => {} })) }));
+vi.mock("./braintrust-experiment.js", async importOriginal => ({ ...await importOriginal<typeof import("./braintrust-experiment.js")>(), importBraintrustExperiment: vi.fn() }));
+const directories: string[] = [];
+beforeEach(() => { vi.clearAllMocks(); });
+afterEach(async () => { for (const directory of directories.splice(0)) await rm(directory, { recursive: true, force: true }); });
+
+it.each(["verified", "incomplete", "unavailable"] as const)("retains %s gateway coverage independently of native indexing", async status => {
+  const workspace = await mkdtemp(path.join(tmpdir(), "scheduled-gateway-index-")); directories.push(workspace);
+  const directory = path.join(workspace, "tests/runner-e2e/results/scheduled"), manifestPath = path.join(directory, "manifest.json");
+  const campaignId = "scheduled-gateway-fixture", projectId = "11111111-1111-4111-8111-111111111111";
+  const cell = runnerMatrix.find(row => row.id === "core-compatibility.runner-acpx-claude.local.message-marker")!;
+  const root = path.join(workspace, "tests/runner-e2e/results", campaignId), resultPath = path.join(root, ...cell.id.split("."), "attempt-1");
+  await mkdir(directory, { recursive: true }); await mkdir(resultPath, { recursive: true });
+  const result: RunnerE2EResult = { schema: "paperclip.runner-e2e.result/v2", executionId: cell.id,
+    suiteId: cell.suite.id, suiteDefinitionHash: cell.suiteDefinitionHash,
+    accountingIdentity: { campaignId, attemptId: "one" }, source: { sha: "a".repeat(40), ref: null, workflowRunUrl: null },
+    attempt: 1, status: "passed", cleanup: "passed", profileId: cell.profile.id, caseId: cell.task.id,
+    environmentId: cell.environment.id, provider: cell.profile.provider, model: cell.profile.model, runtimeMode: cell.profile.expectedRuntimeMode,
+    startedAt: "2026-10-24T00:00:00.000Z", finishedAt: "2026-10-24T00:00:01.000Z", durationMs: 1000,
+    feedback: { ...emptyFeedbackSummary(), collection: "complete", invocationCollection: "complete" },
+    tracing: { projectId, traceId: "a".repeat(32), spanId: "b".repeat(16), status: "accepted", verification: "unverified", nativeConfiguration: "configured",
+      finalization: { spanId: "c".repeat(16), status: "accepted", outcome: "passed", verification: "unverified" } } };
+  const campaign = buildRunnerCampaign({ campaignId, generatedAt: result.finishedAt, expected: [cell.id], results: [result], attempts: [result] });
+  const ledgerPath = path.join(directory, "ledger.jsonl");
+  await writeFile(manifestPath, JSON.stringify({ limits: { campaignId }, ledgerPath, executionIds: [cell.id], projectId, route: "braintrust",
+    models: [{ provider: "anthropic", price: { model: cell.profile.model } }] }));
+  await writeFile(path.join(root, "campaign.json"), JSON.stringify(campaign));
+  await writeFile(path.join(resultPath, "result.json"), JSON.stringify(result));
+  const indexReceipt = { version: 1 as const, projectId, campaignId, checkedAt: result.finishedAt,
+    ledgerCheckpoint: "d".repeat(64), status, expectedRequests: 2, verifiedRequests: status === "verified" ? 2 : 1, requests: [] };
+  vi.mocked(verifyBraintrustGatewayIndexing).mockResolvedValue(indexReceipt);
+  const findings = await finalizeScheduledCampaign({ GITHUB_WORKSPACE: workspace, PAPERCLIP_E2E_OPERATIONS_MANIFEST: manifestPath,
+    PAPERCLIP_E2E_BRAINTRUST_PROJECT_ID: projectId, BRAINTRUST_API_KEY: "private-key-canary" });
+  expect(findings.includes("gateway_indexing_incomplete")).toBe(status !== "verified");
+  expect(findings).not.toContain("trace_or_experiment_indexing_incomplete");
+  expect(JSON.parse(await readFile(path.join(directory, "gateway-indexing-receipt.json"), "utf8"))).toEqual(indexReceipt);
+  expect(JSON.parse(await readFile(path.join(directory, "findings.json"), "utf8"))).toMatchObject({ nativeIndexVerification: "verified", gatewayIndexVerification: status });
+  expect(verifyBraintrustGatewayIndexing).toHaveBeenCalledWith(expect.objectContaining({ projectId }), {
+    ledgerPath, campaignId, expected: [{ executionId: cell.id, provider: "anthropic", model: cell.profile.model }],
+  });
+});