diff --git a/docs/packet-captures.md b/docs/packet-captures.md
index 32cedb11b..8db097386 100644
--- a/docs/packet-captures.md
+++ b/docs/packet-captures.md
@@ -54,6 +54,22 @@ resource limit; use selected fields, a display filter or an input-packet limit
 when dissection output is too large. Each call checks the recording's current
 digest, so pages from different source digests must not be combined.
 
+For a question about a known input prefix, select `packet_limit` for that
+prefix, the fields needed for the answer, and a matching output window. For
+example, a question about the first twelve PostgreSQL TCP packets within
+original frames 1–2,000 can use fields `tcp.srcport`, `tcp.dstport`, `tcp.flags`,
+filter `tcp.port == 5432`, `packet_limit: 2000` and `match_limit: 12`.
+The matching count then describes that input prefix. A filter containing
+`frame.number <= 2000` without an input limit still reads the entire capture;
+returning twelve rows does not limit producer output. These controls answer a
+bounded question and must not be used to infer whole-capture absence.
+
+Discover unfamiliar field names with `inspect_packet_fields`, using an exact
+protocol and a narrow query. If the installed names are already established,
+direct field selection avoids an additional glossary probe. Count discovery
+and any retries when comparing investigation cost; a small page alone does
+not establish a faster workflow.
+
 Additional selections:
 
 - `display_filter`: Wireshark display-filter syntax; omitted means no filter.
diff --git a/experiments/analysis-efficiency/README.md b/experiments/analysis-efficiency/README.md
index cdb6c3c9b..9a14b764a 100644
--- a/experiments/analysis-efficiency/README.md
+++ b/experiments/analysis-efficiency/README.md
@@ -72,6 +72,12 @@ checkpoint commits retain their original receipts; no history was rewritten.
 
 ## Optimization priorities and gradual migration
 
+The [larger/non-HTTP packet checkpoint](packet-field-workflows.md) records
+same-question SDK and actual Codex comparisons using existing input limits,
+field/filter selection, discovery and output windows. Lower delivered JSON did
+not establish stable total-token or speed gains; failed plans and negative
+controls remain visible.
+
 Agent token consumption and end-to-end task time are the primary objectives.
 Memory and CPU consumption are also measured and controlled. Correctness,
 Evidence provenance, caller choices and existing-user compatibility are
diff --git a/experiments/analysis-efficiency/agent-trial.py b/experiments/analysis-efficiency/agent-trial.py
index edb848a28..db994cea1 100644
--- a/experiments/analysis-efficiency/agent-trial.py
+++ b/experiments/analysis-efficiency/agent-trial.py
@@ -181,6 +181,7 @@ def main():
                  'inspect_android_class', 'inspect_android_method',
                  'inspect_firmware_regions', 'inspect_managed_artifact',
                  'inspect_sqlite_database', 'inspect_pcap_capture',
+                 'inspect_packet_capture', 'inspect_packet_fields',
                  'inspect_managed_members', 'inspect_managed_native_boundaries',
                  'inspect_artifact', 'inspect_wasm_artifact',
                  'inspect_evm_interface', 'get_evidence_bundle', 'export_evidence_bundle',
diff --git a/experiments/analysis-efficiency/compare.mjs b/experiments/analysis-efficiency/compare.mjs
index 4e33e5ead..01b9f97af 100644
--- a/experiments/analysis-efficiency/compare.mjs
+++ b/experiments/analysis-efficiency/compare.mjs
@@ -2,6 +2,7 @@ import { isDeepStrictEqual } from "node:util";
 import { readFile, writeFile } from "node:fs/promises";
 import { resolve } from "node:path";
 import { parseArgs } from "node:util";
+import { PACKET_ARTIFACTS } from "./packet-entry.mjs";
 
 const { values } = parseArgs({
   options: {
@@ -33,6 +34,8 @@ const sameSettings = [
   "provider_environment_policy_sha256",
   "measure_source_sha256",
   "native_workflow_source_sha256",
+  "packet_workflow_source_sha256",
+  "packet_oracle_source_sha256",
   "package_lock_sha256",
   "node_version",
   "sdk_client_version",
@@ -76,6 +79,7 @@ for (const [side, answer] of [
 // coverage. Repeated equivalent deliveries must also match every observation
 // field except the receipt and parent IDs, which identify their own invocations.
 if (
+  Object.hasOwn(PACKET_ARTIFACTS, left.artifact) ||
   ["wireshark-http.cap", "wireshark-http_redirects.pcapng"].includes(
     left.artifact,
   )
@@ -94,7 +98,11 @@ if (
       path: ["delivery_observations"],
       value: answer.delivery_observations,
     });
-    if (left.delivery === right.delivery) {
+    if (
+      left.delivery === right.delivery &&
+      (!Object.hasOwn(PACKET_ARTIFACTS, left.artifact) ||
+        left.packet_selection === right.packet_selection)
+    ) {
       answer.delivery_observations = answer.delivery_observations.map(
         (observation) =>
           Object.fromEntries(
diff --git a/experiments/analysis-efficiency/corpus.json b/experiments/analysis-efficiency/corpus.json
index 7896c558d..e595dde34 100644
--- a/experiments/analysis-efficiency/corpus.json
+++ b/experiments/analysis-efficiency/corpus.json
@@ -291,6 +291,58 @@
         "upstream_expected_digest": null,
         "derivation": "Unmodified capture published by official Wireshark wiki. Local digest pins downloaded bytes; no independently published upstream digest is claimed."
       }
+    },
+    {
+      "id": "wireshark-pgsql-jdbc.pcap",
+      "file": "pgsql-jdbc.pcap",
+      "bytes": 4906316,
+      "sha256": "6e68b8b792a0f4336092f98856abc7de8c54b55a29112854d28d676f9093c16b",
+      "provenance": {
+        "project": "Wireshark public sample captures",
+        "source_page": "https://wiki.wireshark.org/SampleCaptures",
+        "url": "https://wiki.wireshark.org/uploads/__moin_import__/attachments/SampleCaptures/pgsql-jdbc.pcap.gz",
+        "resolved_url": "https://gitlab.com/wireshark/wireshark/-/wikis/uploads/__moin_import__/attachments/SampleCaptures/pgsql-jdbc.pcap.gz",
+        "upstream_expected_digest": null,
+        "license": "not stated in inspected upstream catalogue",
+        "derivation": "Unmodified official saved sample; decompressed gzip where indicated. Local digest pins acquired bytes, not an upstream-published digest.",
+        "compressed_source": {
+          "name": "pgsql-jdbc.pcap.gz",
+          "bytes": 590093,
+          "sha256": "15b51bb4cc197753c1ac6b7dc3cf5697f22a81e3bb5d2e1d24d15de469c3a06b"
+        }
+      }
+    },
+    {
+      "id": "wireshark-dns.cap",
+      "file": "dns.cap",
+      "bytes": 4338,
+      "sha256": "041eeb6f98bb398f1ee8b09651b5b5a84f6a62639f95bf226f9e7b77355d9f28",
+      "provenance": {
+        "project": "Wireshark public sample captures",
+        "source_page": "https://wiki.wireshark.org/SampleCaptures",
+        "url": "https://wiki.wireshark.org/uploads/__moin_import__/attachments/SampleCaptures/dns.cap",
+        "resolved_url": "https://gitlab.com/wireshark/wireshark/-/wikis/uploads/__moin_import__/attachments/SampleCaptures/dns.cap",
+        "upstream_expected_digest": null,
+        "license": "not stated in inspected upstream catalogue",
+        "derivation": "Unmodified official saved sample; decompressed gzip where indicated. Local digest pins acquired bytes, not an upstream-published digest.",
+        "compressed_source": null
+      }
+    },
+    {
+      "id": "wireshark-dhcp.pcap",
+      "file": "dhcp.pcap",
+      "bytes": 1400,
+      "sha256": "2471b5420bdac826eecf8f61a2bbb4a3eb20dbfab7c02ff2be502f349f368214",
+      "provenance": {
+        "project": "Wireshark public sample captures",
+        "source_page": "https://wiki.wireshark.org/SampleCaptures",
+        "url": "https://wiki.wireshark.org/uploads/__moin_import__/attachments/SampleCaptures/dhcp.pcap",
+        "resolved_url": "https://gitlab.com/wireshark/wireshark/-/wikis/uploads/__moin_import__/attachments/SampleCaptures/dhcp.pcap",
+        "upstream_expected_digest": null,
+        "license": "not stated in inspected upstream catalogue",
+        "derivation": "Unmodified official saved sample; decompressed gzip where indicated. Local digest pins acquired bytes, not an upstream-published digest.",
+        "compressed_source": null
+      }
     }
   ]
 }
diff --git a/experiments/analysis-efficiency/measure.mjs b/experiments/analysis-efficiency/measure.mjs
index 60dce4f80..07a39e3c1 100644
--- a/experiments/analysis-efficiency/measure.mjs
+++ b/experiments/analysis-efficiency/measure.mjs
@@ -12,6 +12,7 @@ import {
 import { Client } from "@modelcontextprotocol/client";
 import { StdioClientTransport } from "@modelcontextprotocol/client/stdio";
 import { Ajv2020 } from "ajv/dist/2020.js";
+import { PACKET_ARTIFACTS, measurePacketEntry } from "./packet-entry.mjs";
 
 const root = resolve(dirname(fileURLToPath(import.meta.url)), "../..");
 const { values } = parseArgs({
@@ -29,6 +30,7 @@ const { values } = parseArgs({
     "managed-question": { type: "string", default: "inventory" },
     schemas: { type: "string", default: "default" },
     "sqlite-selection": { type: "string", default: "all" },
+    "packet-selection": { type: "string", default: "baseline" },
   },
 });
 if (
@@ -45,6 +47,15 @@ if (
   !["inventory", "selected"].includes(values["javascript-question"]) ||
   !["default", "selected"].includes(values.schemas) ||
   !["all", "columns"].includes(values["sqlite-selection"]) ||
+  !["baseline", "focused", "discovery", "pages"].includes(
+    values["packet-selection"],
+  ) ||
+  (!Object.hasOwn(PACKET_ARTIFACTS, values.artifact) &&
+    values["packet-selection"] !== "baseline") ||
+  (Object.hasOwn(PACKET_ARTIFACTS, values.artifact) &&
+    values.delivery !== "complete") ||
+  (values["packet-selection"] === "pages" &&
+    values.artifact !== "wireshark-pgsql-jdbc.pcap") ||
   (values.artifact !== "qgis-world-map.gpkg" &&
     values["sqlite-selection"] !== "all") ||
   (values.artifact === "qgis-world-map.gpkg" &&
@@ -142,6 +153,12 @@ const fingerprints = {
   native_workflow_source_sha256: await fileSha(
     fileURLToPath(new URL("./native-entry.mjs", import.meta.url)),
   ),
+  packet_workflow_source_sha256: await fileSha(
+    fileURLToPath(new URL("./packet-entry.mjs", import.meta.url)),
+  ),
+  packet_oracle_source_sha256: await fileSha(
+    fileURLToPath(new URL("./packet-oracle.py", import.meta.url)),
+  ),
   runtime_files_sha256: sha(JSON.stringify(runtimeFiles)),
   runtime_dispatcher_sha256: await fileSha(
     resolve(values["runtime-root"], "scripts/rea.mjs"),
@@ -722,6 +739,12 @@ const inspectPcapTarget = async () => {
 };
 
 const workflows = {
+  ...Object.fromEntries(
+    Object.keys(PACKET_ARTIFACTS).map((artifact) => [
+      artifact,
+      () => measurePacketEntry({ values, call, inputSha, catalogDefinitions }),
+    ]),
+  ),
   "wireshark-http.cap": inspectPcapTarget,
   "wireshark-http_redirects.pcapng": inspectPcapTarget,
   "qgis-world-map.gpkg": async () => {
@@ -1160,6 +1183,9 @@ const report = {
     values.artifact === "qgis-world-map.gpkg"
       ? values["sqlite-selection"]
       : null,
+  packet_selection: Object.hasOwn(PACKET_ARTIFACTS, values.artifact)
+    ? values["packet-selection"]
+    : null,
   provider_settings_sha256: sha(JSON.stringify(selectedProviders)),
   provider_search_path_sha256: sha(
     selectedProviders.PATH ?? process.env.PATH ?? "",
diff --git a/experiments/analysis-efficiency/packet-entry.mjs b/experiments/analysis-efficiency/packet-entry.mjs
new file mode 100644
index 000000000..5c6722a74
--- /dev/null
+++ b/experiments/analysis-efficiency/packet-entry.mjs
@@ -0,0 +1,316 @@
+import { isDeepStrictEqual } from "node:util";
+import { resolve } from "node:path";
+import { pathToFileURL } from "node:url";
+import { fileURLToPath } from "node:url";
+import { execFile } from "node:child_process";
+import { promisify } from "node:util";
+
+export const verifyPacketPage = (first, next, oracle) => {
+  if (
+    first.artifact.sha256 !== oracle.sha256 ||
+    next.artifact.sha256 !== oracle.sha256 ||
+    first.output_window.matching_packets !== oracle.matching_packets ||
+    next.output_window.matching_packets !== oracle.matching_packets ||
+    !isDeepStrictEqual(
+      packetFacts("postgres", first).rows,
+      oracle.rows.slice(0, 12),
+    ) ||
+    !isDeepStrictEqual(
+      packetFacts("postgres", next).rows,
+      oracle.rows.slice(12, 24),
+    )
+  )
+    throw new Error(
+      "Packet pages differ from independently parsed original matches",
+    );
+};
+
+export const PACKET_ARTIFACTS = {
+  "wireshark-pgsql-jdbc.pcap": {
+    kind: "postgres",
+    fields: ["tcp.srcport", "tcp.dstport", "tcp.flags"],
+    extra: ["tcp.len"],
+    protocol: "tcp",
+    query: "tcp.flags",
+    discovered: "tcp.flags",
+    filter: "tcp.port == 5432",
+  },
+  "wireshark-dns.cap": {
+    kind: "dns",
+    fields: [
+      "dns.id",
+      "dns.flags.response",
+      "dns.flags.rcode",
+      "dns.qry.name",
+      "dns.qry.type",
+    ],
+    extra: ["dns.resp.name", "dns.resp.ttl", "udp.srcport", "udp.dstport"],
+    protocol: "dns",
+    query: "dns.flags.rcode",
+    discovered: "dns.flags.rcode",
+    filter: "dns.flags.response == 1 && dns.flags.rcode != 0",
+  },
+  "wireshark-dhcp.pcap": {
+    kind: "dhcp",
+    fields: [
+      "dhcp.id",
+      "dhcp.option.dhcp",
+      "dhcp.ip.your",
+      "dhcp.option.dhcp_server_id",
+    ],
+    extra: [
+      "dhcp.type",
+      "dhcp.ip.client",
+      "dhcp.hw.mac_addr",
+      "udp.srcport",
+      "udp.dstport",
+    ],
+    protocol: "dhcp",
+    query: "Option 53:",
+    discovered: "dhcp.option.dhcp",
+    filter: "dhcp",
+  },
+};
+
+export const packetRequest = (artifact, path, selection) => {
+  const descriptor = PACKET_ARTIFACTS[artifact];
+  if (!descriptor) throw new Error("Unknown pinned packet question");
+  const baseline = selection === "baseline";
+  return {
+    path,
+    fields: baseline
+      ? [...descriptor.fields, ...descriptor.extra]
+      : descriptor.fields,
+    display_filter:
+      descriptor.kind === "postgres" && baseline
+        ? `${descriptor.filter} && frame.number <= 2000`
+        : descriptor.kind === "dns" && baseline
+          ? "dns"
+          : descriptor.filter,
+    ...(descriptor.kind === "postgres" && !baseline
+      ? { packet_limit: 2000, match_limit: 12 }
+      : {}),
+  };
+};
+
+const number = (row, field) => {
+  const value = row[field];
+  if (
+    !Array.isArray(value) ||
+    value.length !== 1 ||
+    !/^(?:[0-9]+|0x[0-9a-f]+)$/iu.test(value[0])
+  )
+    throw new Error(`Expected one recorded integer for ${field}`);
+  const parsed = Number(value[0]);
+  if (!Number.isSafeInteger(parsed))
+    throw new Error(`Unsafe recorded integer for ${field}`);
+  return parsed;
+};
+
+export const packetFacts = (kind, report) => {
+  let packets = report.packets;
+  if (kind === "postgres") packets = packets.slice(0, 12);
+  if (kind === "dns")
+    packets = packets.filter(
+      (row) =>
+        number(row, "dns.flags.response") === 1 &&
+        number(row, "dns.flags.rcode") !== 0,
+    );
+  return {
+    kind,
+    matching_packets:
+      kind === "dns" ? packets.length : report.output_window.matching_packets,
+    rows: packets.map((row) => {
+      const frame = number(row, "frame.number");
+      if (kind === "postgres")
+        return {
+          frame,
+          source_port: number(row, "tcp.srcport"),
+          destination_port: number(row, "tcp.dstport"),
+          flags: number(row, "tcp.flags"),
+        };
+      if (kind === "dns") {
+        const names = row["dns.qry.name"];
+        const types = row["dns.qry.type"];
+        if (
+          !Array.isArray(names) ||
+          names.length === 0 ||
+          !Array.isArray(types) ||
+          types.length !== names.length
+        )
+          throw new Error(
+            "The fixed DNS question requires recorded names and types; missing fields are unknown",
+          );
+        return {
+          frame,
+          id: number(row, "dns.id"),
+          rcode: number(row, "dns.flags.rcode"),
+          names,
+          types: types.map((value) => number({ type: [value] }, "type")),
+        };
+      }
+      const server = row["dhcp.option.dhcp_server_id"];
+      const address = row["dhcp.ip.your"];
+      if (
+        !Array.isArray(address) ||
+        address.length !== 1 ||
+        (server !== undefined &&
+          (!Array.isArray(server) || server.length !== 1))
+      )
+        throw new Error("Expected recorded DHCP address scalar");
+      return {
+        frame,
+        id: number(row, "dhcp.id"),
+        message_type: number(row, "dhcp.option.dhcp"),
+        your_address: address[0],
+        ...(server === undefined ? {} : { server_identifier: server[0] }),
+      };
+    }),
+  };
+};
+
+export const measurePacketEntry = async ({
+  values,
+  call,
+  inputSha,
+  catalogDefinitions,
+}) => {
+  const descriptor = PACKET_ARTIFACTS[values.artifact];
+  const { parseEvidence } = await import(
+    pathToFileURL(resolve(values["runtime-root"], "dist/domain/evidence.js"))
+      .href
+  );
+  const { inspectPacketCaptureInputSchema, packetCaptureSchema } = await import(
+    pathToFileURL(
+      resolve(values["runtime-root"], "dist/domain/captures/packetCapture.js"),
+    ).href
+  );
+  const { inspectPacketFieldsInputSchema, packetFieldsCatalogSchema } =
+    await import(
+      pathToFileURL(
+        resolve(values["runtime-root"], "dist/domain/captures/packetFields.js"),
+      ).href
+    );
+  const observations = [];
+  const selection = values["packet-selection"];
+  const checked = (evidence, operation, input, schema) => {
+    parseEvidence(evidence);
+    const report = schema.parse(evidence.normalized_result);
+    if (
+      evidence.operation !== operation ||
+      !isDeepStrictEqual(evidence.parameters, input) ||
+      !isDeepStrictEqual(report.selection, input)
+    )
+      throw new Error(
+        "Packet Evidence operation or choices differs from the request",
+      );
+    if (
+      operation === "inspect_packet_capture" &&
+      (evidence.subject?.local_path !== values.path ||
+        evidence.subject?.digest?.sha256 !== inputSha ||
+        report.artifact.path !== values.path ||
+        report.artifact.sha256 !== inputSha)
+    )
+      throw new Error("Packet Evidence identifies different capture bytes");
+    if (
+      operation === "inspect_packet_fields" &&
+      (evidence.subject?.local_path !== report.tool.path ||
+        evidence.subject?.digest?.sha256 !== report.tool.sha256)
+    )
+      throw new Error("Field discovery identifies a different executable");
+    observations.push({
+      evidence_id: evidence.evidence_id,
+      operation,
+      selection: report.selection,
+      output_window: report.output_window,
+      read_scope: report.read_scope ?? null,
+      limitations: evidence.limitations,
+    });
+    return report;
+  };
+  if (selection === "discovery") {
+    const input = inspectPacketFieldsInputSchema.parse({
+      kind: "fields",
+      protocol: descriptor.protocol,
+      query: descriptor.query,
+      limit: 1,
+    });
+    const found = checked(
+      await call("inspect_packet_fields", input),
+      "inspect_packet_fields",
+      input,
+      packetFieldsCatalogSchema,
+    );
+    if (
+      !found.records.some((row) => row.abbreviation === descriptor.discovered)
+    )
+      throw new Error(
+        "Selected installation does not advertise the required protocol field",
+      );
+  }
+  const input = inspectPacketCaptureInputSchema.parse(
+    packetRequest(values.artifact, values.path, selection),
+  );
+  for (const name of [
+    "inspect_packet_capture",
+    ...(selection === "discovery" ? ["inspect_packet_fields"] : []),
+  ])
+    if (!catalogDefinitions.has(name))
+      throw new Error(`Missing live packet contract ${name}`);
+  const report = checked(
+    await call("inspect_packet_capture", input),
+    "inspect_packet_capture",
+    input,
+    packetCaptureSchema,
+  );
+  const facts = packetFacts(descriptor.kind, report);
+  if (selection === "pages") {
+    const { stdout } = await promisify(execFile)(
+      "python3",
+      [
+        fileURLToPath(new URL("./packet-oracle.py", import.meta.url)),
+        "postgres",
+        values.path,
+        "--row-limit",
+        "24",
+      ],
+      { timeout: 30_000, maxBuffer: 128 * 1024 },
+    );
+    const oracle = JSON.parse(stdout);
+    const nextInput = inspectPacketCaptureInputSchema.parse({
+      ...input,
+      match_offset: 12,
+    });
+    const next = checked(
+      await call("inspect_packet_capture", nextInput),
+      "inspect_packet_capture",
+      nextInput,
+      packetCaptureSchema,
+    );
+    const repeated = checked(
+      await call("inspect_packet_capture", input),
+      "inspect_packet_capture",
+      input,
+      packetCaptureSchema,
+    );
+    verifyPacketPage(report, next, oracle);
+    if (
+      !isDeepStrictEqual(report.packets, repeated.packets) ||
+      !isDeepStrictEqual(report.output_window, repeated.output_window) ||
+      next.artifact.sha256 !== report.artifact.sha256 ||
+      next.output_window.matching_packets !==
+        report.output_window.matching_packets ||
+      Number(next.packets[0]?.["frame.number"]?.[0]) <=
+        Number(report.packets.at(-1)?.["frame.number"]?.[0])
+    )
+      throw new Error(
+        "Packet page order, digest or repeated observations changed",
+      );
+  }
+  return {
+    artifact_sha256: inputSha,
+    facts,
+    delivery_observations: observations,
+    evidence_ids: [...new Set(observations.map((row) => row.evidence_id))],
+  };
+};
diff --git a/experiments/analysis-efficiency/packet-field-workflows.md b/experiments/analysis-efficiency/packet-field-workflows.md
new file mode 100644
index 000000000..8f893c30c
--- /dev/null
+++ b/experiments/analysis-efficiency/packet-field-workflows.md
@@ -0,0 +1,226 @@
+# Larger and non-HTTP packet investigation costs
+
+This closes the bounded experiment in [#3130](https://github.com/morluto/rea/issues/3130).
+Existing field, filter and input/output-window controls substantially reduce
+delivered JSON for the selected questions. Actual Codex total input and elapsed
+time are less consistent. Requiring field discovery for a known field adds cost.
+These results support documenting caller-selected controls, while retaining the
+current API, defaults and uncached provider behavior.
+
+## Runtime and public corpus
+
+Measurements use immutable main
+`ec67aebed02dbd20fa5c7703fcd183ad0b1c343c`, Node 24.18.0, the locked MCP client and
+server, Linux, TShark/capinfos 4.0.17 and util-linux prlimit. The SDK captures live
+advertised contracts and passes selected definitions to actual public MCP calls.
+Runtime-file, dispatcher, lockfile, experiment-source and explicit provider
+configuration digests remain with the private trial records. Subsequent main
+integration checks do not relabel these measurements.
+
+Three additional saved recordings come from the
+[official Wireshark catalogue](https://wiki.wireshark.org/SampleCaptures).
+`corpus.json` pins their acquisition URLs, exact bytes and SHA-256 values.
+The PostgreSQL recording is described upstream as a JDBC test session; it is
+not a representative production workload. Capture-specific licenses were not
+stated in the inspected catalogue. No captures or raw trial records are
+redistributed in Git.
+
+| Recording            |                        Acquired/analyzed size | Recorded frames | Fixed question                                                                                                                                         |
+| -------------------- | --------------------------------------------: | --------------: | ------------------------------------------------------------------------------------------------------------------------------------------------------ |
+| `pgsql-jdbc.pcap.gz` | 590,093 compressed / 4,906,316 expanded bytes |          18,472 | Within original frames 1–2,000, report the first twelve TCP packets using port 5432, their source/destination ports and flags, and the matching count. |
+| `dns.cap`            |                                   4,338 bytes |              38 | Report nonzero-RCODE DNS responses, transaction IDs and question names/types.                                                                          |
+| `dhcp.pcap`          |                                   1,400 bytes |               4 | Report message types, transaction IDs, offered address and present server identifiers in frame order.                                                  |
+
+The larger application recording and two small non-HTTP controls expand the
+question/protocol coverage. Large DNS/DHCP, PCAPNG non-HTTP, USB, Bluetooth,
+Windows and macOS provider performance remain unverified.
+
+`packet-oracle.py` independently reads classic PCAP, Ethernet/IPv4 and selected
+TCP/UDP headers. It does not consume TShark output. Its narrow supported scope
+fails on truncation, IP fragmentation, DNS compression cycles, duplicate DHCP
+options or DHCP overload. TCP header facts follow
+[RFC 9293](https://www.rfc-editor.org/rfc/rfc9293.html), DNS names/header fields
+[RFC 1035](https://www.rfc-editor.org/rfc/rfc1035.html), and DHCP options
+[RFC 2132](https://www.rfc-editor.org/rfc/rfc2132.html).
+SQL interpretation, TCP reassembly and network completeness are outside the
+oracle. The question and oracle were fixed before measured cohorts.
+
+The oracle establishes 1,337 selected PostgreSQL TCP packets in the input prefix
+and six DNS error responses. DHCP records contain Discover/Offer with transaction
+ID 15645 and Request/ACK with ID 15646. An absent server identifier stays omitted;
+the recording does not establish a single transaction containing all four steps.
+
+## Equivalent questions and SDK measurements
+
+Each arm uses fresh actual MCP sessions; heavy trials run sequentially.
+Two interleaved repetitions cover each arm and each recording. Discovery, catalog,
+RPC, schema checking, encoding and cleanup costs are retained. JSON bytes describe
+serialized SDK replies, not raw transport bytes or model tokens.
+
+The PostgreSQL baseline selects `tcp.srcport`, `tcp.dstport`, `tcp.flags` and
+`tcp.len`, retains the default fifty-row output window and reads to EOF with
+`tcp.port == 5432 && frame.number <= 2000`. The focused arm selects the three
+required fields, uses `packet_limit: 2000`, filter `tcp.port == 5432` and
+`match_limit: 12`. Both establish the same requested prefix facts and count.
+Their source read scopes differ explicitly: to EOF versus an input limit.
+Neither scope establishes whole-network completeness.
+
+DNS focuses the nine-field baseline on five requested fields and the error-response
+filter. DHCP selects four required fields from the nine-field baseline; all four
+records are still read. The discovery arm adds one narrow installed-field probe
+before the focused request. It tests the cost of validating an already specified
+field, rather than proving a benefit for a previously unknown field or guessing
+that an installed registration appears in traffic.
+
+| Question          | Baseline reply bytes | Focused reply bytes | Focused change | Discovery + focused bytes | Discovery change |
+| ----------------- | -------------------: | ------------------: | -------------: | ------------------------: | ---------------: |
+| PostgreSQL prefix |               23,054 |              13,304 |        −42.29% |                    21,096 |           −8.49% |
+| DNS errors        |               29,002 |              13,424 |        −53.71% |                    21,226 |          −26.81% |
+| DHCP sequence     |               13,770 |              12,020 |        −12.71% |                    19,840 |          +44.08% |
+
+Both repetitions produce the same reply size in each cell and pass the independent
+fact comparison. Authentication uses the selected runtime's `parseEvidence` and
+named input/output schemas, checking operation, normalized caller choices,
+artifact or executable identity, original frame order and omission semantics.
+
+| Question          | Baseline tool-call time, ms | Focused tool-call time, ms | Discovery + focused tool-call time, ms |
+| ----------------- | --------------------------- | -------------------------- | -------------------------------------- |
+| PostgreSQL prefix | 1,670 / 2,255               | 1,708 / 933                | 3,549 / 2,674                          |
+| DNS errors        | 892 / 786                   | 889 / 778                  | 2,506 / 2,608                          |
+| DHCP sequence     | 763 / 759                   | 755 / 731                  | 2,502 / 2,502                          |
+
+Whole SDK workflow times, including connection/catalog, vary from 4.23–8.79 seconds
+for these arms. The PostgreSQL focused tool time improves in one repetition and
+regresses slightly in the other. Discovery consistently adds tool work. Two
+repetitions are descriptive observations, not a statistical speed guarantee.
+
+A separate first-page, next-page and repeated-first-page sequence costs
+922 / 804 / 1,244 ms and returns 39,918 reply bytes in total. Each call repeats
+dissection, reports the same 1,337 matching count and capture digest, and preserves
+the original frame order. Identical repeated observations retain the same
+content-addressed Evidence ID; distinct pages have distinct IDs. Page delivery
+does not establish capture or network completeness. This narrow repeat does not
+justify introducing a generic cache or weakening provider identity/cleanup rules.
+
+The two original HTTP controls were also replayed with two SDK pairs each on the
+new pinned runtime. Useful summary/prefix facts remain equivalent. For `http.cap`,
+complete delivery returns 7,601 bytes and summary plus a two-URI retained page
+17,540 bytes (+130.76%); for `http_redirects.pcapng`, 7,535 versus 17,240 bytes
+(+128.80%). These remain negative controls. The original model observations in
+[the earlier checkpoint](pcap-small-captures.md) keep their original runtime pin.
+
+## Actual Codex CLI observations
+
+Six actual Codex CLI 0.161.0 runs use requested alias `gpt-5.6-terra`, reasoning
+`high`, a disposable configuration and the pinned REA MCP runtime. Resolved
+server-side model identity is not inferred. Both tool names are enabled in every
+arm, exact-name discovery guidance is identical, shell/browser/other-agent tools
+are disabled and copied authentication is removed after each run.
+The shared question and generated plans contain no oracle answers.
+
+Order is baseline 1, focused 1, discovery 1, focused 2, baseline 2, discovery 2.
+The rubric checks authenticated actual tool results, exact header-derived facts,
+citations and the planned calls. Answer correctness and adherence to the plan
+are reported separately. Limitation wording is reviewed in transcripts; a string
+or string array is accepted because the prompt does not require one representation.
+
+| Trial       | Total input tokens | Cached input tokens | Output tokens | Elapsed seconds | Actual MCP calls | Facts / plan                       |
+| ----------- | -----------------: | ------------------: | ------------: | --------------: | ---------------: | ---------------------------------- |
+| Baseline 1  |             40,649 |              27,904 |         1,574 |          27.989 |                1 | pass / pass                        |
+| Focused 1   |             36,642 |              19,968 |           943 |          38.542 |                1 | pass / pass                        |
+| Discovery 1 |             52,802 |              33,024 |         1,089 |          44.398 |                2 | pass / pass                        |
+| Focused 2   |             40,912 |              22,016 |           726 |          32.520 |                1 | pass / pass                        |
+| Baseline 2  |             40,181 |              19,968 |           896 |          40.299 |                1 | pass / pass                        |
+| Discovery 2 |             48,754 |              30,976 |         1,780 |          76.081 |                1 | pass / **fail: discovery omitted** |
+
+The two eligible baseline/focused pairs show total-input changes of −9.86% and
++1.82%, and elapsed changes of +37.70% and −19.30%. Thus reduced SDK reply size
+does not establish a stable total-token or speed improvement in these runs.
+The first complete discovery comparison adds 29.90% input and 58.63% elapsed time.
+The second discovery attempt is retained as a workflow failure and cannot serve
+as a successful two-call discovery comparison. Its cost is not discarded or
+counted as a passing trial. No model-side recovery or retries occurred in these
+six attempts.
+
+An earlier configuration preflight rejected the packet tool under Codex's
+`approval_policy: never`: 36,651 input tokens, 549 output tokens, 21.603 seconds,
+one rejected call and no packet observations. The experiment runner now explicitly
+preapproves the two authorized passive packet tools, consistently across all arms.
+That failed preflight remains separate from the corrected cohort and its cost is
+reported here; CLI exit zero alone is never a factual or workflow pass.
+
+## Resource and compatibility scope
+
+All engine/model workloads use an owned Linux resource scope with one CPU,
+3 GiB early-stop threshold, 4 GiB hard memory ceiling, no swap and a 192-task cap.
+Node's heap is additionally limited to 512 MiB. The nineteen non-HTTP SDK trials
+have a maximum observed payload-exit memory peak of 860,909,568 bytes; the six model
+attempts peak at 940,806,144 bytes. Payload-exit counters and cleanup were verified
+with the observer still alive and report no memory-limit/OOM events in that phase.
+Final whole-scope counters remain unavailable, so no stronger whole-scope claim
+is made. CPU usage and observed process ownership remain in each private receipt.
+
+A full PostgreSQL probe with four selected fields and no input prefix exceeds the
+existing 4 MiB producer-output ceiling even with the default fifty-row window.
+It correctly fails instead of delivering partial success. The SDK error retains
+large truncated diagnostics (5,790,979 serialized bytes); this separate product
+concern is recorded privately for later repair, outside #3130. Paging alone cannot
+prevent producer expansion. The successful prefix experiment does not prove that
+the full recording can be inspected with those selections.
+
+Only experiment sources, corpus provenance, a question template and guidance
+change. Production schemas, provider limits, CLI/MCP defaults, Evidence identity
+and session/resource ownership stay as before. Eight parser regressions retain
+malformed/truncated-input and omission checks. Existing clients need no migration.
+
+## Reproduction
+
+Acquire the exact corpus outside the repository and verify the pinned hashes.
+Use the pinned Node/npm toolchain, locked dependencies, `npm run build:cached` and
+an external resource guard before invoking actual engines/models. Supply explicit
+provider settings in a private JSON file. An example SDK arm is:
+
+```sh
+node experiments/analysis-efficiency/measure.mjs \
+  --runtime-root /absolute/pinned/runtime --revision PINNED_SHA \
+  --artifact wireshark-pgsql-jdbc.pcap --path /private/pgsql-jdbc.pcap \
+  --providers /private/providers.json --schemas selected \
+  --packet-selection focused --output /private/new-trial
+```
+
+Use `baseline`, `focused`, `discovery`, or the PostgreSQL-only `pages` diagnostic.
+Run `packet-oracle.py postgres PATH` independently and compare `answer.json` facts.
+Generate each model prompt with `packet-model-prompt.mjs --selection ARM`, then use
+the existing `agent-trial.py` with those two enabled tools and context capture.
+Run `review-packet.mjs` against the actual trial, oracle, runtime and selected arm.
+Its nonzero exit includes a plan failure even when all requested facts are correct.
+Retain failed attempts, actual usage, digests, coverage, cleanup and resource
+receipts alongside passing observations. No raw results need to be committed.
+
+## Review verification
+
+The comparator now recognizes the three non-HTTP questions: observations remain
+attached to each run, while different plans compare requested facts. Repeated
+identical plans still compare stable observation scope. Capture grading authenticates
+Evidence against the actual call and assesses plan differences separately; an
+actual 24-row SDK window replay retains a factual pass and a plan failure. This
+is a grading replay, not an additional model trial. The stricter header oracle
+preserves all original sample facts and rejects truncated IPv4/TCP options and
+unsupported Ethernet protocols. The page diagnostic now derives matches 13–24
+independently before accepting the second page; skipped, reversed and changed-byte
+pages are rejected. Original cohort measurements retain their recorded source
+digests and runtime pin; review checks do not relabel them.
+
+Capture grading keeps authenticated facts by Evidence ID and evaluates the cited
+observations. Four replays using actual first-page and second-page SDK Evidence
+verify that later or reordered pages and repeated reads preserve a correct cited
+answer, while citing only the wrong page fails factual grading. Extra calls still
+fail the fixed workflow plan. Regrading all six original model records preserves
+their outcomes and recorded costs; these replays are not new model trials.
+
+Failed or malformed calls remain workflow failures and never become authenticated
+Evidence. Nine failure-injection and answer-shape replays using an actual successful
+capture verify that later recovery preserves supported facts, while failed-only
+attempts and missing or unobserved citations fail factual grading. Missing
+limitations and extra answer keys fail the fixed plan without changing correct
+recorded facts. Successful original-model outcomes remain unchanged.
diff --git a/experiments/analysis-efficiency/packet-model-prompt.mjs b/experiments/analysis-efficiency/packet-model-prompt.mjs
new file mode 100644
index 000000000..f49348fbd
--- /dev/null
+++ b/experiments/analysis-efficiency/packet-model-prompt.mjs
@@ -0,0 +1,40 @@
+import { readFile } from "node:fs/promises";
+import { parseArgs } from "node:util";
+import { packetRequest } from "./packet-entry.mjs";
+
+const { values } = parseArgs({ options: { selection: { type: "string" } } });
+if (!["baseline", "focused", "discovery"].includes(values.selection))
+  throw new Error("Select baseline, focused or discovery");
+const question = await readFile(
+  new URL("./questions/packet-postgres.txt", import.meta.url),
+  "utf8",
+);
+const plan = [
+  ...(values.selection === "discovery"
+    ? [
+        {
+          tool: "inspect_packet_fields",
+          arguments: {
+            kind: "fields",
+            protocol: "tcp",
+            query: "tcp.flags",
+            limit: 1,
+          },
+        },
+      ]
+    : []),
+  {
+    tool: "inspect_packet_capture",
+    arguments: packetRequest(
+      "wireshark-pgsql-jdbc.pcap",
+      "{{artifact_path}}",
+      values.selection,
+    ),
+  },
+];
+process.stdout.write(
+  question +
+    "\nExecute this fixed public REA MCP plan in order. These are input selections, not expected answers. Preserve omitted defaults. All calls count toward the comparison.\n" +
+    JSON.stringify(plan, null, 2) +
+    "\n",
+);
diff --git a/experiments/analysis-efficiency/packet-oracle.py b/experiments/analysis-efficiency/packet-oracle.py
new file mode 100644
index 000000000..db5797b5f
--- /dev/null
+++ b/experiments/analysis-efficiency/packet-oracle.py
@@ -0,0 +1,180 @@
+"""Independent, narrow oracle for the three pinned Ethernet/IPv4 PCAPs.
+
+Reads header bytes, not TShark output. No PCAPNG, IP fragmentation or TCP
+reassembly support is claimed. DNS compression follows RFC 1035; DHCP options
+53/54 follow RFC 2132. Unsupported selected traffic fails rather than guessing.
+"""
+import argparse
+import hashlib
+import json
+import struct
+from pathlib import Path
+
+
+def require(condition, reason):
+    if not condition:
+        raise ValueError(reason)
+
+
+def packets(data):
+    require(len(data) >= 24, 'truncated PCAP header')
+    magic = data[:4]
+    require(magic in (b'\xd4\xc3\xb2\xa1', b'\xa1\xb2\xc3\xd4'),
+            'oracle requires microsecond classic PCAP')
+    endian = '<' if magic == b'\xd4\xc3\xb2\xa1' else '>'
+    require(struct.unpack_from(endian + 'HH', data, 4) == (2, 4),
+            'unsupported PCAP version')
+    require(struct.unpack_from(endian + 'I', data, 20)[0] == 1,
+            'oracle requires Ethernet')
+    offset, frame = 24, 0
+    while offset < len(data):
+        require(offset + 16 <= len(data), 'truncated PCAP record header')
+        captured, original = struct.unpack_from(endian + 'II', data, offset + 8)
+        offset += 16
+        require(captured <= original and offset + captured <= len(data),
+                'invalid captured packet length')
+        frame += 1
+        yield frame, data[offset:offset + captured]
+        offset += captured
+
+
+def transport(packet):
+    require(len(packet) >= 14, 'truncated Ethernet header')
+    if packet[12:14] == b'\x08\x06':
+        return None  # ARP cannot contain the selected TCP/UDP messages.
+    require(packet[12:14] == b'\x08\x00', 'unsupported Ethernet protocol')
+    ip = packet[14:]
+    require(len(ip) >= 20 and ip[0] >> 4 == 4, 'invalid IPv4 header')
+    header = (ip[0] & 15) * 4
+    total = struct.unpack_from('!H', ip, 2)[0]
+    require(20 <= header <= len(ip) and total >= header, 'invalid IPv4 lengths')
+    require(total <= len(ip), 'truncated IPv4 packet')
+    require(struct.unpack_from('!H', ip, 6)[0] & 0x3fff == 0,
+            'fragmentation is outside oracle scope')
+    return ip[9], ip[header:min(total, len(ip))]
+
+
+def dns_name(message, offset):
+    labels, visited, end = [], set(), None
+    while True:
+        require(offset < len(message) and offset not in visited,
+                'invalid DNS name or compression cycle')
+        visited.add(offset)
+        size = message[offset]
+        if size & 0xc0 == 0xc0:
+            require(offset + 1 < len(message), 'truncated DNS compression pointer')
+            if end is None:
+                end = offset + 2
+            offset = ((size & 0x3f) << 8) | message[offset + 1]
+            continue
+        require(size & 0xc0 == 0 and offset + 1 + size <= len(message),
+                'invalid DNS label')
+        offset += 1
+        if size == 0:
+            return '.'.join(labels), end if end is not None else offset
+        labels.append(message[offset:offset + size].decode('ascii'))
+        require(sum(len(label) + 1 for label in labels) <= 255,
+                'DNS name exceeds protocol length')
+        offset += size
+
+
+def dhcp_options(payload):
+    require(len(payload) >= 240 and payload[236:240] == b'\x63\x82\x53\x63',
+            'invalid DHCP fixed header or cookie')
+    options, offset = {}, 240
+    while offset < len(payload):
+        kind = payload[offset]
+        offset += 1
+        if kind == 255:
+            return options
+        if kind == 0:
+            continue
+        require(offset < len(payload), 'missing DHCP option length')
+        size = payload[offset]
+        offset += 1
+        require(offset + size <= len(payload), 'truncated DHCP option')
+        require(kind not in options, 'duplicate DHCP options outside oracle scope')
+        options[kind] = payload[offset:offset + size]
+        offset += size
+    raise ValueError('DHCP end option missing')
+
+
+def analyze(data, kind, row_limit=12):
+    rows, matching, count = [], 0, 0
+    for frame, packet in packets(data):
+        count = frame
+        decoded = transport(packet)
+        if decoded is None:
+            continue
+        protocol, segment = decoded
+        if kind == 'postgres' and protocol == 6 and frame <= 2000:
+            require(len(segment) >= 20, 'truncated TCP header')
+            tcp_header = (segment[12] >> 4) * 4
+            require(20 <= tcp_header <= len(segment), 'invalid or truncated TCP header/options')
+            source, destination = struct.unpack_from('!HH', segment)
+            if 5432 not in (source, destination):
+                continue
+            matching += 1
+            if len(rows) < row_limit:
+                rows.append({'frame': frame, 'source_port': source,
+                             'destination_port': destination,
+                             'flags': struct.unpack_from('!H', segment, 12)[0] & 0xfff})
+        elif kind in ('dns', 'dhcp') and protocol == 17:
+            require(len(segment) >= 8, 'truncated UDP header')
+            source, destination, length = struct.unpack_from('!HHH', segment)
+            selected = 53 in (source, destination) if kind == 'dns' else (
+                source, destination) in ((67, 68), (68, 67))
+            if not selected:
+                continue
+            require(8 <= length <= len(segment), 'truncated selected UDP message')
+            payload = segment[8:length]
+            if kind == 'dns':
+                require(len(payload) >= 12, 'truncated DNS header')
+                identifier, flags, questions = struct.unpack_from('!HHH', payload)
+                if not flags & 0x8000 or flags & 15 == 0:
+                    continue
+                matching += 1
+                offset, names, types = 12, [], []
+                for _ in range(questions):
+                    name, offset = dns_name(payload, offset)
+                    require(offset + 4 <= len(payload), 'truncated DNS question')
+                    types.append(struct.unpack_from('!H', payload, offset)[0])
+                    names.append(name)
+                    offset += 4
+                rows.append({'frame': frame, 'id': identifier, 'rcode': flags & 15,
+                             'names': names, 'types': types})
+            else:
+                options = dhcp_options(payload)
+                require(53 in options and len(options[53]) == 1,
+                        'missing DHCP message type')
+                server = options.get(54)
+                require(server is None or len(server) == 4, 'invalid DHCP server ID')
+                require(52 not in options, 'DHCP overload outside oracle scope')
+                matching += 1
+                row = {'frame': frame, 'id': struct.unpack_from('!I', payload, 4)[0],
+                       'message_type': options[53][0],
+                       'your_address': '.'.join(map(str, payload[16:20]))}
+                if server is not None:
+                    row['server_identifier'] = '.'.join(map(str, server))
+                rows.append(row)
+    return {'kind': kind, 'capture_frames': count, 'matching_packets': matching,
+            'question_input_frames': min(2000, count) if kind == 'postgres' else count,
+            'rows': rows, 'scope': 'selected recorded headers; no network completeness claim'}
+
+
+def main():
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument('kind', choices=['postgres', 'dns', 'dhcp'])
+    parser.add_argument('path', type=Path)
+    parser.add_argument('--row-limit', type=int, choices=[12, 24], default=12,
+                        help='PostgreSQL rows; 24 independently checks the second page')
+    args = parser.parse_args()
+    with args.path.open('rb') as source:
+        data = source.read(32 * 1024 * 1024 + 1)
+    require(len(data) <= 32 * 1024 * 1024, 'oracle input budget exceeded')
+    print(json.dumps({'sha256': hashlib.sha256(data).hexdigest(),
+                      **analyze(data, args.kind, args.row_limit)}, indent=2))
+
+
+if __name__ == '__main__':
+    main()
diff --git a/experiments/analysis-efficiency/questions/packet-postgres.txt b/experiments/analysis-efficiency/questions/packet-postgres.txt
new file mode 100644
index 000000000..82914ba3e
--- /dev/null
+++ b/experiments/analysis-efficiency/questions/packet-postgres.txt
@@ -0,0 +1,3 @@
+Inspect the selected saved PostgreSQL capture {{artifact_path}} with REA. Within original capture frames 1 through 2000, return the first 12 TCP packets with source or destination port 5432 in original frame order. For each return integer frame, source_port, destination_port and flags (the numeric TCP flags word, not prose). Also report the matching packet count within this 2000-frame question scope.
+
+Do not claim whole-capture or network completeness. Return a single JSON object with exactly kind ("postgres"), matching_packets, rows, evidence_ids and limitations. Use actual successful REA Evidence IDs. Include a concise limitation explaining the recorded input scope; no extra claims or tool calls are needed.
diff --git a/experiments/analysis-efficiency/review-packet.mjs b/experiments/analysis-efficiency/review-packet.mjs
new file mode 100644
index 000000000..e7e3b0def
--- /dev/null
+++ b/experiments/analysis-efficiency/review-packet.mjs
@@ -0,0 +1,247 @@
+import { readFile, writeFile } from "node:fs/promises";
+import { resolve } from "node:path";
+import { pathToFileURL } from "node:url";
+import { isDeepStrictEqual, parseArgs } from "node:util";
+import { packetFacts, packetRequest } from "./packet-entry.mjs";
+
+const { values } = parseArgs({
+  options: {
+    trial: { type: "string" },
+    oracle: { type: "string" },
+    "runtime-root": { type: "string" },
+    selection: { type: "string" },
+  },
+});
+if (
+  !values.trial ||
+  !values.oracle ||
+  !values["runtime-root"] ||
+  !["baseline", "focused", "discovery"].includes(values.selection)
+)
+  throw new Error(
+    "Use --trial DIR --oracle JSON --runtime-root DIR --selection baseline|focused|discovery",
+  );
+
+const root = values["runtime-root"];
+const { parseEvidence } = await import(
+  pathToFileURL(resolve(root, "dist/domain/evidence.js")).href
+);
+const { inspectPacketCaptureInputSchema, packetCaptureSchema } = await import(
+  pathToFileURL(resolve(root, "dist/domain/captures/packetCapture.js")).href
+);
+const { inspectPacketFieldsInputSchema, packetFieldsCatalogSchema } =
+  await import(
+    pathToFileURL(resolve(root, "dist/domain/captures/packetFields.js")).href
+  );
+const oracle = JSON.parse(await readFile(values.oracle, "utf8"));
+const settings = JSON.parse(
+  await readFile(resolve(values.trial, "trial-settings.json"), "utf8"),
+);
+const summary = JSON.parse(
+  await readFile(resolve(values.trial, "summary.json"), "utf8"),
+);
+const expected = Object.fromEntries(
+  ["kind", "matching_packets", "rows"].map((key) => [key, oracle[key]]),
+);
+const calls = (await readFile(resolve(values.trial, "events.jsonl"), "utf8"))
+  .split("\n")
+  .filter(Boolean)
+  .map(JSON.parse)
+  .filter(
+    (event) =>
+      event.type === "item.completed" && event.item?.type === "mcp_tool_call",
+  )
+  .map((event) => event.item);
+const reasons = [];
+const workflowReasons = [];
+let answer;
+const captures = new Map();
+try {
+  if (oracle.kind !== "postgres")
+    throw new Error("This model rubric is the fixed PostgreSQL question only");
+  const expectedCalls =
+    values.selection === "discovery"
+      ? ["inspect_packet_fields", "inspect_packet_capture"]
+      : ["inspect_packet_capture"];
+  if (
+    !isDeepStrictEqual(
+      calls.map((call) => call.tool),
+      expectedCalls,
+    ) ||
+    calls.some(
+      (call) =>
+        call.server !== "rea" || call.status !== "completed" || call.error,
+    )
+  )
+    workflowReasons.push(
+      "Completed public REA call sequence differs from the fixed plan",
+    );
+  const ids = new Set();
+  const conflictingIds = new Set();
+  for (const call of calls) {
+    if (call.server !== "rea" || call.status !== "completed" || call.error)
+      continue;
+    try {
+      const result = call.result;
+      if (result?.isError)
+        throw new Error("Tool failure is not observed packet evidence");
+      const text =
+        result?.content?.filter((item) => item.type === "text") ?? [];
+      if (!result?.structuredContent && text.length !== 1)
+        throw new Error("No unambiguous captured Evidence envelope");
+      const evidence = result.structuredContent ?? JSON.parse(text[0].text);
+      parseEvidence(evidence);
+      if (conflictingIds.has(evidence.evidence_id))
+        throw new Error("Previously conflicting Evidence identity");
+      if (call.tool === "inspect_packet_fields") {
+        const input = inspectPacketFieldsInputSchema.parse({
+          kind: "fields",
+          protocol: "tcp",
+          query: "tcp.flags",
+          limit: 1,
+        });
+        const report = packetFieldsCatalogSchema.parse(
+          evidence.normalized_result,
+        );
+        if (
+          !isDeepStrictEqual(
+            inspectPacketFieldsInputSchema.parse(call.arguments),
+            input,
+          ) ||
+          !isDeepStrictEqual(evidence.parameters, input) ||
+          !isDeepStrictEqual(report.selection, input) ||
+          evidence.operation !== call.tool ||
+          !report.records.some((row) => row.abbreviation === "tcp.flags") ||
+          evidence.subject?.local_path !== report.tool.path ||
+          evidence.subject?.digest?.sha256 !== report.tool.sha256
+        )
+          throw new Error(
+            "Discovery Evidence does not establish the requested installed field",
+          );
+      } else if (call.tool === "inspect_packet_capture") {
+        const planned = inspectPacketCaptureInputSchema.parse(
+          packetRequest(
+            "wireshark-pgsql-jdbc.pcap",
+            settings.selected_artifact_path,
+            values.selection,
+          ),
+        );
+        const input = inspectPacketCaptureInputSchema.parse(call.arguments);
+        if (!isDeepStrictEqual(input, planned))
+          workflowReasons.push("Capture selection differs from the fixed plan");
+        const report = packetCaptureSchema.parse(evidence.normalized_result);
+        if (
+          !isDeepStrictEqual(evidence.parameters, input) ||
+          !isDeepStrictEqual(report.selection, input) ||
+          evidence.operation !== call.tool ||
+          evidence.subject?.local_path !== input.path ||
+          evidence.subject?.digest?.sha256 !== oracle.sha256 ||
+          report.artifact.path !== input.path ||
+          report.artifact.sha256 !== oracle.sha256
+        )
+          throw new Error(
+            "Capture Evidence differs from the fixed input/byte identity",
+          );
+        const observed = packetFacts("postgres", report);
+        if (
+          captures.has(evidence.evidence_id) &&
+          !isDeepStrictEqual(captures.get(evidence.evidence_id), observed)
+        ) {
+          captures.delete(evidence.evidence_id);
+          ids.delete(evidence.evidence_id);
+          conflictingIds.add(evidence.evidence_id);
+          throw new Error("Repeated Evidence identity has inconsistent facts");
+        }
+        captures.set(evidence.evidence_id, observed);
+      } else {
+        throw new Error("Call is not a packet observation tool");
+      }
+      ids.add(evidence.evidence_id);
+    } catch (error) {
+      workflowReasons.push(
+        `Call ${call.tool} did not provide authenticated packet observations: ${String(error)}`,
+      );
+    }
+  }
+  answer = JSON.parse(
+    (await readFile(resolve(values.trial, "answer.txt"), "utf8")).trim(),
+  );
+  if (
+    !isDeepStrictEqual(
+      Object.keys(answer).sort(),
+      [
+        "kind",
+        "matching_packets",
+        "rows",
+        "evidence_ids",
+        "limitations",
+      ].sort(),
+    )
+  )
+    workflowReasons.push("Final answer shape differs from the fixed plan");
+  if (
+    !isDeepStrictEqual(
+      Object.fromEntries(
+        Object.keys(expected).map((key) => [key, answer[key]]),
+      ),
+      expected,
+    )
+  )
+    throw new Error("Final answer is not the exact requested recorded facts");
+  if (
+    !Array.isArray(answer.evidence_ids) ||
+    answer.evidence_ids.some((id) => !ids.has(id)) ||
+    new Set(answer.evidence_ids).size !== answer.evidence_ids.length
+  )
+    throw new Error("Final citations lack the authenticated packet Evidence");
+  const supportingFacts = answer.evidence_ids
+    .map((id) => captures.get(id))
+    .find((observed) => isDeepStrictEqual(observed, expected));
+  if (!supportingFacts)
+    throw new Error(
+      "Cited observations differ from independent raw header oracle",
+    );
+  const limitations =
+    typeof answer.limitations === "string"
+      ? [answer.limitations]
+      : answer.limitations;
+  if (
+    !Array.isArray(limitations) ||
+    limitations.length === 0 ||
+    limitations.some((item) => typeof item !== "string" || !item.trim())
+  )
+    workflowReasons.push("Final answer omits investigation limitations");
+  if (
+    summary.exit_code !== 0 ||
+    summary.timeout ||
+    summary.selected_artifact_bytes_unchanged !== true ||
+    settings.selected_artifact_sha256_before !== oracle.sha256 ||
+    summary.selected_artifact_sha256_after !== oracle.sha256
+  )
+    workflowReasons.push(
+      "Trial did not complete with the selected artifact unchanged",
+    );
+} catch (error) {
+  reasons.push(String(error));
+}
+const review = {
+  factual_correctness: reasons.length ? "failed" : "passed",
+  scope:
+    "Fixed first-2000-frame PostgreSQL question; authenticated tool observations and independent raw header oracle",
+  workflow_correctness:
+    reasons.length || workflowReasons.length ? "failed" : "passed",
+  reasons,
+  workflow_reasons: workflowReasons,
+  completed_calls: calls.length,
+  wall_seconds: summary.wall_seconds,
+  usage: summary.usage,
+  selected_artifact_sha256: oracle.sha256,
+  limitation_prose:
+    "Presence checked; wording requires transcript review, not heuristic factual certification",
+};
+await writeFile(
+  resolve(values.trial, "packet-review.json"),
+  JSON.stringify(review, null, 2) + "\n",
+);
+console.log(JSON.stringify(review));
+if (reasons.length || workflowReasons.length) process.exitCode = 1;
diff --git a/experiments/analysis-efficiency/test_packet_oracle.py b/experiments/analysis-efficiency/test_packet_oracle.py
new file mode 100644
index 000000000..6727b3b7e
--- /dev/null
+++ b/experiments/analysis-efficiency/test_packet_oracle.py
@@ -0,0 +1,67 @@
+import importlib.util
+import struct
+import unittest
+from pathlib import Path
+
+spec = importlib.util.spec_from_file_location('packet_oracle', Path(__file__).with_name('packet-oracle.py'))
+oracle = importlib.util.module_from_spec(spec)
+spec.loader.exec_module(oracle)
+
+
+class PacketOracleTest(unittest.TestCase):
+    def test_compression_pointer_preserves_question_cursor(self):
+        message = b'\x03www\x07example\x03com\0' + b'\xc0\0' + b'\0\x01\0\x01'
+        self.assertEqual(oracle.dns_name(message, 17), ('www.example.com', 19))
+
+    def test_compression_cycle_and_missing_pointer_byte_fail(self):
+        for message in (b'\xc0\0', b'\xc0', b'\xc0\x02\xc0\0'):
+            with self.subTest(message=message), self.assertRaises(ValueError):
+                oracle.dns_name(message, 0)
+
+    def test_truncated_pcap_record_is_not_exhaustion(self):
+        header = struct.pack('<IHHIIII', 0xa1b2c3d4, 2, 4, 0, 0, 65535, 1)
+        with self.assertRaises(ValueError):
+            list(oracle.packets(header + b'\0'))
+        record = struct.pack('<IIII', 0, 0, 20, 20)
+        with self.assertRaises(ValueError):
+            list(oracle.packets(header + record + b'\0' * 19))
+
+    def test_missing_dhcp_server_id_stays_omitted(self):
+        payload = b'\0' * 236 + b'\x63\x82\x53\x63\x35\x01\x01\xff'
+        self.assertEqual(oracle.dhcp_options(payload), {53: b'\x01'})
+
+    def test_dhcp_duplicates_truncation_and_missing_end_fail(self):
+        prefix = b'\0' * 236 + b'\x63\x82\x53\x63'
+        for suffix in (b'\x35\x01\x01\x35\x01\x02\xff', b'\x36\x04\0', b'\x35\x01\x01'):
+            with self.subTest(suffix=suffix), self.assertRaises(ValueError):
+                oracle.dhcp_options(prefix + suffix)
+
+    def test_fragmented_selected_traffic_cannot_pass_header_oracle(self):
+        ip = bytearray(20)
+        ip[0], ip[9] = 0x45, 17
+        struct.pack_into('!H', ip, 2, 20)
+        struct.pack_into('!H', ip, 6, 0x2000)
+        with self.assertRaises(ValueError):
+            oracle.transport(b'\0' * 12 + b'\x08\0' + ip)
+
+    def test_declared_ip_length_and_tcp_options_cannot_be_truncated(self):
+        ip = bytearray(40)
+        ip[0], ip[9], ip[32] = 0x45, 6, 0xf0
+        struct.pack_into('!H', ip, 2, 40)
+        struct.pack_into('!HH', ip, 20, 12345, 5432)
+        packet = b'\0' * 12 + b'\x08\0' + ip
+        header = struct.pack('<IHHIIII', 0xa1b2c3d4, 2, 4, 0, 0, 65535, 1)
+        record = struct.pack('<IIII', 0, 0, len(packet), len(packet))
+        with self.assertRaisesRegex(ValueError, 'TCP header'):
+            oracle.analyze(header + record + packet, 'postgres')
+        struct.pack_into('!H', ip, 2, 60)
+        with self.assertRaisesRegex(ValueError, 'truncated IPv4'):
+            oracle.transport(b'\0' * 12 + b'\x08\0' + ip)
+
+    def test_unknown_link_protocol_does_not_prove_tcp_absence(self):
+        with self.assertRaisesRegex(ValueError, 'unsupported Ethernet'):
+            oracle.transport(b'\0' * 12 + b'\x86\xdd' + b'\0' * 40)
+
+
+if __name__ == '__main__':
+    unittest.main()
diff --git a/tests/evaluation/analysisEfficiencyComparison.test.ts b/tests/evaluation/analysisEfficiencyComparison.test.ts
index 57a1ffb8c..924ba1612 100644
--- a/tests/evaluation/analysisEfficiencyComparison.test.ts
+++ b/tests/evaluation/analysisEfficiencyComparison.test.ts
@@ -30,11 +30,17 @@ type Fixture = {
     delivery: Delivery;
     provider_search_path_sha256: string;
     calls: unknown[];
+    packet_selection?: string;
   };
   answer: {
     capture: { coverage: string; capture_start: null };
     http_request_uris_first2: string[];
     delivery_observations: Observation[];
+    facts?: {
+      kind: string;
+      matching_packets: number;
+      rows: { frame: number }[];
+    };
   };
 };
 const reportSchema = z.object({
@@ -123,6 +129,39 @@ function compare(left: Fixture, right: Fixture) {
   }
 }
 
+for (const artifact of [
+  "wireshark-pgsql-jdbc.pcap",
+  "wireshark-dns.cap",
+  "wireshark-dhcp.pcap",
+]) {
+  test(`${artifact}: compare requested facts across packet plans and preserve observations`, () => {
+    const left = fixture("complete", artifact);
+    left.summary.packet_selection = "baseline";
+    left.answer.facts = {
+      kind: "packet",
+      matching_packets: 1337,
+      rows: [{ frame: 9 }],
+    };
+    const right = structuredClone(left);
+    right.summary.packet_selection = "focused";
+    right.answer.delivery_observations[0]!.evidence_id = "ev_focused";
+    right.answer.delivery_observations[0]!.coverage =
+      "up-to-input-packet-limit";
+    const equivalent = compare(left, right);
+    assert.equal(equivalent.status, 0);
+    assert.equal(equivalent.report.facts_match, true);
+    assert.deepEqual(
+      equivalent.report.preserved_run_specific_values.map((row) => row.value),
+      [left.answer.delivery_observations, right.answer.delivery_observations],
+    );
+    right.answer.facts!.matching_packets++;
+    assert.equal(compare(left, right).report.facts_match, false);
+    right.answer.facts!.matching_packets--;
+    right.summary.packet_selection = "baseline";
+    assert.equal(compare(left, right).report.facts_match, false);
+  });
+}
+
 for (const artifact of [
   "wireshark-http.cap",
   "wireshark-http_redirects.pcapng",
diff --git a/tests/evaluation/packetWorkflowSemantics.test.ts b/tests/evaluation/packetWorkflowSemantics.test.ts
new file mode 100644
index 000000000..aac1f5976
--- /dev/null
+++ b/tests/evaluation/packetWorkflowSemantics.test.ts
@@ -0,0 +1,32 @@
+import assert from "node:assert/strict";
+import { spawnSync } from "node:child_process";
+import { test } from "vitest";
+
+test("packet page verification rejects skipped matches with unchanged counts and monotone frames", () => {
+  const moduleUrl = new URL(
+    "../../experiments/analysis-efficiency/packet-entry.mjs",
+    import.meta.url,
+  ).href;
+  const source = `
+    import assert from 'node:assert/strict';
+    import {verifyPacketPage} from ${JSON.stringify(moduleUrl)};
+    const rows = Array.from({length:36}, (_,i)=>({frame:i+1, source_port:1000,destination_port:5432,flags:16}));
+    const sha256 = 'a'.repeat(64);
+    const report = (selected)=>({artifact:{sha256},output_window:{matching_packets:36},packets:selected.map(row=>({
+      'frame.number':[String(row.frame)],'tcp.srcport':['1000'],'tcp.dstport':['5432'],'tcp.flags':['0x0010']
+    }))});
+    const oracle={sha256,matching_packets:36,rows:rows.slice(0,24)};
+    const first=report(rows.slice(0,12));
+    verifyPacketPage(first,report(rows.slice(12,24)),oracle);
+    assert.throws(()=>verifyPacketPage(first,report(rows.slice(24,36)),oracle),/independently parsed/);
+    assert.throws(()=>verifyPacketPage(first,report(rows.slice(12,24).reverse()),oracle),/independently parsed/);
+    assert.throws(()=>verifyPacketPage(first,{...report(rows.slice(12,24)),artifact:{sha256:'b'.repeat(64)}},oracle),/independently parsed/);
+  `;
+  // Semantic checker only; synthetic rows do not establish provider behavior.
+  const result = spawnSync(process.execPath, ["--input-type=module", "-"], {
+    input: source,
+    encoding: "utf8",
+  });
+  assert.ifError(result.error);
+  assert.equal(result.status, 0, result.stderr);
+});