@piagent/platform 1.5.5 → 1.6.0

package/docs-site/team.html

@@ -30,7 +30,7 @@
         <span>Pi Agent Platform</span>
       </a>
       <div class="top-actions">
-        <span class="pill"><span class="pill-dot"></span>v1.5.5 docs · VI</span>
+        <span class="pill"><span class="pill-dot"></span>v1.6.0 docs · VI</span>
         <nav class="language-switch" aria-label="Ngôn ngữ">
           <a class="language-option current" href="/team" hreflang="vi" lang="vi" aria-current="true">VI</a>
           <a class="language-option" href="/en/team" hreflang="en" lang="en">EN</a>

package/docs-site/whats-new.html

@@ -30,7 +30,7 @@
         <span>Pi Agent Platform</span>
       </a>
       <div class="top-actions">
-        <span class="pill"><span class="pill-dot"></span>v1.5.5 docs · VI</span>
+        <span class="pill"><span class="pill-dot"></span>v1.6.0 docs · VI</span>
         <nav class="language-switch" aria-label="Ngôn ngữ">
           <a class="language-option current" href="/whats-new" hreflang="vi" lang="vi" aria-current="true">VI</a>
           <a class="language-option" href="/en/whats-new" hreflang="en" lang="en">EN</a>

package/docs-site/workflows.html

@@ -30,7 +30,7 @@
         <span>Pi Agent Platform</span>
       </a>
       <div class="top-actions">
-        <span class="pill"><span class="pill-dot"></span>v1.5.5 docs · VI</span>
+        <span class="pill"><span class="pill-dot"></span>v1.6.0 docs · VI</span>
         <nav class="language-switch" aria-label="Ngôn ngữ">
           <a class="language-option current" href="/workflows" hreflang="vi" lang="vi" aria-current="true">VI</a>
           <a class="language-option" href="/en/workflows" hreflang="en" lang="en">EN</a>

package/evals/fixtures/benchmark-assurance-evidence.invalid.json

@@ -29,7 +29,7 @@
   },
   "disjointness": {
     "taxonomyDigest": "d42458d04861a60488a5f98532936db71d43dcccc0298d75b09c2255f2922cb1",
-    "publicExposureDigest": "b7bdf8d76f66982410f047b5138ca7b974f9b441115761a7e81817753bfa41b3",
+    "publicExposureDigest": "629c1e9f835464b5e3db1c6daac67d8f2661c691ac10201b904e4ed8b377dbef",
     "reportDigest": "5555555555555555555555555555555555555555555555555555555555555555",
     "familyDisjoint": true,
     "repositoryDisjoint": false,

package/evals/fixtures/benchmark-assurance-evidence.valid.json

@@ -29,7 +29,7 @@
   },
   "disjointness": {
     "taxonomyDigest": "d42458d04861a60488a5f98532936db71d43dcccc0298d75b09c2255f2922cb1",
-    "publicExposureDigest": "b7bdf8d76f66982410f047b5138ca7b974f9b441115761a7e81817753bfa41b3",
+    "publicExposureDigest": "629c1e9f835464b5e3db1c6daac67d8f2661c691ac10201b904e4ed8b377dbef",
     "reportDigest": "5555555555555555555555555555555555555555555555555555555555555555",
     "familyDisjoint": true,
     "repositoryDisjoint": true,

package/evals/fixtures/piagent-webui/session-live-state-v1.invalid.json

@@ -0,0 +1,18 @@
+{
+  "schemaVersion": 1,
+  "version": "piagent-session-live-state-v1",
+  "generatedAt": "2026-08-21T08:00:00.000Z",
+  "gatewayInstanceRef": "gateway_fixture_01",
+  "eventSequence": 42,
+  "state": "ready",
+  "operations": [
+    {
+      "sessionRef": "session_fixture_01",
+      "operationRef": "operation_fixture_01",
+      "state": "settling",
+      "abortable": true
+    }
+  ],
+  "settlements": [],
+  "reasonCode": null
+}

package/evals/fixtures/piagent-webui/session-live-state-v1.valid.json

@@ -0,0 +1,27 @@
+{
+  "schemaVersion": 1,
+  "version": "piagent-session-live-state-v1",
+  "generatedAt": "2026-08-21T08:00:00.000Z",
+  "gatewayInstanceRef": "gateway_fixture_01",
+  "eventSequence": 42,
+  "state": "ready",
+  "operations": [
+    {
+      "sessionRef": "session_fixture_01",
+      "operationRef": "operation_fixture_01",
+      "state": "running",
+      "abortable": true
+    }
+  ],
+  "settlements": [
+    {
+      "sessionRef": "session_fixture_02",
+      "operationRef": "operation_fixture_02",
+      "settlement": "error",
+      "reasonCode": "assistant-response-failed",
+      "settledAt": "2026-08-21T07:59:59.000Z",
+      "sequence": 41
+    }
+  ],
+  "reasonCode": null
+}

package/evals/fs4-readiness-matrix.v1.json

@@ -5,7 +5,7 @@
   "purpose": "Independently classify deterministic E0, public-regression E1, calibrated-capability E2, private-holdout-readiness E3, and provider-free long-horizon evidence without widening any claim.",
   "artifacts": [
     { "id": "e0-core-suite", "path": "benchmarks/core-v1/suite.json", "sha256": "ec19b26f80850143a5196380b3fd1b858f42e7c200c4000d5f684b891959f84b" },
-    { "id": "e1-public-suite", "path": "benchmarks/production-v1/suite.json", "sha256": "8d03df7c08eee5debbabdb3490beb5d45b021a392b6da77df257e6d94c346d58" },
+    { "id": "e1-public-suite", "path": "benchmarks/production-v1/suite.json", "sha256": "194a5d36c70ca46c318bc12eb1858b9a345638e8e3613a63693980e228b3c4d1" },
     { "id": "e2-suite", "path": "benchmarks/e2-framework-v1/suite.json", "sha256": "97d28eb96341f03e3f2537d648265e6ec80537a98941e9f5d349dea7b90f26f1" },
     { "id": "e2-reference", "path": "benchmarks/e2-framework-v1/reports/reference-report.v1.json", "sha256": "37df0eda9a93320772751dc4bcd0b522b2f5b4743ffed76334b68b768e4f7c8b" },
     { "id": "e2-mutation", "path": "benchmarks/e2-framework-v1/reports/mutation-report.v1.json", "sha256": "859c8863f6c8c671d8b71e7f379f53097c8e967046825a0e55e36930b0eb4135" },
@@ -17,7 +17,7 @@
     { "id": "e3-runbook", "path": "evals/private-holdout-v1/CUSTODIAN_RUNBOOK.md", "sha256": "b59503971549d44f4366766db780c0d89bc0b90c5ccb01d2269c5440446954b8" },
     { "id": "e3-access-policy", "path": "evals/private-holdout-v1/access-policy.v1.json", "sha256": "3189aa3d811083954aaab5470a07c23963d05d4bacccc547a29844e2a8a76525" },
     { "id": "e3-human-rubric", "path": "evals/private-holdout-v1/human-rubric.v1.json", "sha256": "e291e5097ef1415cc037198e00d690d0abd7a8a9f8b98be773a25d313ac1b629" },
-    { "id": "e3-public-exposure", "path": "evals/private-holdout-v1/public-exposure.v1.json", "sha256": "b7bdf8d76f66982410f047b5138ca7b974f9b441115761a7e81817753bfa41b3" },
+    { "id": "e3-public-exposure", "path": "evals/private-holdout-v1/public-exposure.v1.json", "sha256": "629c1e9f835464b5e3db1c6daac67d8f2661c691ac10201b904e4ed8b377dbef" },
     { "id": "taxonomy", "path": "evals/real-task-taxonomy.v1.json", "sha256": "d42458d04861a60488a5f98532936db71d43dcccc0298d75b09c2255f2922cb1" }
   ],
   "deterministicTests": {

package/evals/fs5-causal-arm.v1.json

@@ -33,11 +33,11 @@
     },
     {
       "path": "tests/phase-tools-evaluation.test.mjs",
-      "sha256": "357182d50f6b97e9edfdbcdf959315c2f10910691e297c44d5e85f2300681d49"
+      "sha256": "6a3686db841273e6003279c026a2c7a9907601d457bb42d1f5621577c8f70357"
     },
     {
       "path": "evals/fs5-pilot-protocol.v1.json",
-      "sha256": "392ec00603a0ef997d18d7327e360dddfe301a748f2395530bffe214c59c9219"
+      "sha256": "b84377c7d35c8fc70a6bd2142182810518c010f39ca0ee4f8488d9fddbab37fa"
     }
   ],
   "localReproducer": {

package/evals/fs5-pilot-protocol.v1.json

@@ -28,12 +28,12 @@
     {
       "id": "pilot-suite",
       "path": "benchmarks/production-v1/suite.json",
-      "sha256": "8d03df7c08eee5debbabdb3490beb5d45b021a392b6da77df257e6d94c346d58"
+      "sha256": "194a5d36c70ca46c318bc12eb1858b9a345638e8e3613a63693980e228b3c4d1"
     },
     {
       "id": "task-contract-schema",
       "path": "schemas/task-contract.schema.json",
-      "sha256": "432a253af865f5c79527b3c7e5e65fc3b6cfea43d19f9e915f3bda792a358bb4"
+      "sha256": "7b3cf9d0284e15db3636049f4ab298d0adedcc6c614030517e4b119c9f3bf6e6"
     },
     {
       "id": "assurance-schema",

package/evals/fs5-pilot-protocol.v4.json

@@ -17,14 +17,15 @@
   "artifactBindings": [
     { "id": "authority-manifest", "path": "packages/piagent-core/policy/authority-manifest.v1.json", "sha256": "911dc026f1a860f593a6850f7348311f6adac83a8f11627abb23654960b605d0" },
     { "id": "canary-suite", "path": "benchmarks/capability-v1/suite.json", "sha256": "29bb79e96b047d79056626d747baa0b3a6ca9d4da7ad80b9668c0085e5581706" },
-    { "id": "pilot-suite", "path": "benchmarks/production-v1/suite.json", "sha256": "8d03df7c08eee5debbabdb3490beb5d45b021a392b6da77df257e6d94c346d58" },
-    { "id": "task-contract-schema", "path": "schemas/task-contract.schema.json", "sha256": "432a253af865f5c79527b3c7e5e65fc3b6cfea43d19f9e915f3bda792a358bb4" },
+    { "id": "pilot-suite", "path": "benchmarks/production-v1/suite.json", "sha256": "194a5d36c70ca46c318bc12eb1858b9a345638e8e3613a63693980e228b3c4d1" },
+    { "id": "task-contract-schema", "path": "schemas/task-contract.schema.json", "sha256": "7b3cf9d0284e15db3636049f4ab298d0adedcc6c614030517e4b119c9f3bf6e6" },
     { "id": "assurance-schema", "path": "schemas/benchmark-assurance-evidence.schema.json", "sha256": "5d937632d2cfff34dbbb781dd367bb78feb64c36d571c36ca15f2cc40c576af8" },
-    { "id": "failure-and-operational-classifier", "path": "packages/piagent-core/benchmark/benchmark-forensics.js", "sha256": "4c0a2d011f48bc356b5edd50dd849d520b66ad91b8ccd47c6eb23ff3e5f5e9fb" },
+    { "id": "failure-and-operational-classifier", "path": "packages/piagent-core/benchmark/benchmark-forensics.js", "sha256": "6afec2ca572b15771f0c110f9f352e4bba364a460b61ed9fe8df03bbc52f470a" },
     { "id": "phase-attribution", "path": "packages/piagent-core/benchmark/benchmark-phase-attribution.js", "sha256": "9c07bda5bfe7bb7b0959844ccea6790735cc9f154d8d5972e18785148cea6d9a" },
-    { "id": "record-validator", "path": "packages/piagent-core/benchmark/benchmark-record-validation.js", "sha256": "70bccd4db24ab44b68af590c67aa82caf2b7fedc3b83333a097858001235848a" },
+    { "id": "record-validator", "path": "packages/piagent-core/benchmark/benchmark-record-validation.js", "sha256": "262a1b934c30032663ecfc570e039c3bfdc1511e2b47115e7ccfd08bf1cfcd72" },
+    { "id": "timing-diagnostics", "path": "packages/piagent-core/benchmark/benchmark-timing-diagnostics.js", "sha256": "5f52293f4c5c6d8444a35f94c6ccfd9e9183db8c8fa0e755c3da5b0cb847c2aa" },
     { "id": "protocol-enforcement", "path": "packages/piagent-core/benchmark/fs5-pilot-protocol.js", "sha256": "2bcc0cf6c18eee0c080c30e727610b0f10a278f5862855d920ba5797f07a2f2d" },
-    { "id": "runner-core", "path": "scripts/benchmark-runner-core.mjs", "sha256": "9081c4fb90747b682ccbd41adebae4536239b9742a9a3e930245ae676a9ff322" }
+    { "id": "runner-core", "path": "scripts/benchmark-runner-core.mjs", "sha256": "ba9abb58f9d3962cd79d1cae7cbb4fb07df642da254bf9bf1976b5f2bcab7df6" }
   ],
   "runIdentityContract": {
     "required": [

package/evals/fs5-pilot-protocol.v5.json

@@ -17,15 +17,16 @@
   "artifactBindings": [
     { "id": "authority-manifest", "path": "packages/piagent-core/policy/authority-manifest.v1.json", "sha256": "911dc026f1a860f593a6850f7348311f6adac83a8f11627abb23654960b605d0" },
     { "id": "canary-suite", "path": "benchmarks/capability-v1/suite.json", "sha256": "29bb79e96b047d79056626d747baa0b3a6ca9d4da7ad80b9668c0085e5581706" },
-    { "id": "pilot-suite", "path": "benchmarks/production-v1/suite.json", "sha256": "8d03df7c08eee5debbabdb3490beb5d45b021a392b6da77df257e6d94c346d58" },
-    { "id": "task-contract-schema", "path": "schemas/task-contract.schema.json", "sha256": "432a253af865f5c79527b3c7e5e65fc3b6cfea43d19f9e915f3bda792a358bb4" },
+    { "id": "pilot-suite", "path": "benchmarks/production-v1/suite.json", "sha256": "194a5d36c70ca46c318bc12eb1858b9a345638e8e3613a63693980e228b3c4d1" },
+    { "id": "task-contract-schema", "path": "schemas/task-contract.schema.json", "sha256": "7b3cf9d0284e15db3636049f4ab298d0adedcc6c614030517e4b119c9f3bf6e6" },
     { "id": "assurance-schema", "path": "schemas/benchmark-assurance-evidence.schema.json", "sha256": "5d937632d2cfff34dbbb781dd367bb78feb64c36d571c36ca15f2cc40c576af8" },
-    { "id": "failure-and-operational-classifier", "path": "packages/piagent-core/benchmark/benchmark-forensics.js", "sha256": "4c0a2d011f48bc356b5edd50dd849d520b66ad91b8ccd47c6eb23ff3e5f5e9fb" },
-    { "id": "session-orchestration", "path": "scripts/benchmark-session.mjs", "sha256": "3bcca9dbd5948dff639dc7061188937a6d8e6fe819e5642a23c56961639978ec" },
+    { "id": "failure-and-operational-classifier", "path": "packages/piagent-core/benchmark/benchmark-forensics.js", "sha256": "6afec2ca572b15771f0c110f9f352e4bba364a460b61ed9fe8df03bbc52f470a" },
+    { "id": "session-orchestration", "path": "scripts/benchmark-session.mjs", "sha256": "085ef089b37b98361e6e43f16d42413848130ead93756cf6b9a3057a0ad29cb7" },
     { "id": "phase-attribution", "path": "packages/piagent-core/benchmark/benchmark-phase-attribution.js", "sha256": "9c07bda5bfe7bb7b0959844ccea6790735cc9f154d8d5972e18785148cea6d9a" },
-    { "id": "record-validator", "path": "packages/piagent-core/benchmark/benchmark-record-validation.js", "sha256": "70bccd4db24ab44b68af590c67aa82caf2b7fedc3b83333a097858001235848a" },
+    { "id": "record-validator", "path": "packages/piagent-core/benchmark/benchmark-record-validation.js", "sha256": "262a1b934c30032663ecfc570e039c3bfdc1511e2b47115e7ccfd08bf1cfcd72" },
+    { "id": "timing-diagnostics", "path": "packages/piagent-core/benchmark/benchmark-timing-diagnostics.js", "sha256": "5f52293f4c5c6d8444a35f94c6ccfd9e9183db8c8fa0e755c3da5b0cb847c2aa" },
     { "id": "protocol-enforcement", "path": "packages/piagent-core/benchmark/fs5-pilot-protocol.js", "sha256": "2bcc0cf6c18eee0c080c30e727610b0f10a278f5862855d920ba5797f07a2f2d" },
-    { "id": "runner-core", "path": "scripts/benchmark-runner-core.mjs", "sha256": "9081c4fb90747b682ccbd41adebae4536239b9742a9a3e930245ae676a9ff322" }
+    { "id": "runner-core", "path": "scripts/benchmark-runner-core.mjs", "sha256": "ba9abb58f9d3962cd79d1cae7cbb4fb07df642da254bf9bf1976b5f2bcab7df6" }
   ],
   "runIdentityContract": {
     "required": [

package/evals/private-holdout-v1/public-exposure.v1.json

@@ -18,7 +18,7 @@
     },
     {
       "id": "production-v1",
-      "manifestSha256": "8d03df7c08eee5debbabdb3490beb5d45b021a392b6da77df257e6d94c346d58",
+      "manifestSha256": "194a5d36c70ca46c318bc12eb1858b9a345638e8e3613a63693980e228b3c4d1",
       "scenarioIds": ["tenant-role-authorization", "invoice-rounding", "tenant-cache-isolation", "stale-search-response", "unicode-search", "pagination-boundary", "quoted-csv", "stable-dedup", "schema-migration", "config-precedence", "cli-double-dash", "workspace-order", "bounded-retry", "expiry-boundary", "incident-diagnosis", "protected-env-refusal", "repository-prompt-injection", "destructive-history-refusal"]
     },
     {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@piagent/platform",
-  "version": "1.5.5",
+  "version": "1.6.0",
   "license": "MIT",
   "publishConfig": {
     "access": "public",
@@ -59,6 +59,7 @@
     "packages/piagent-webui/extension/",
     "packages/piagent-webui/gateway/",
     "packages/piagent-webui/ownership/",
+    "packages/piagent-webui/shared/",
     "packages/piagent-webui/dist/client/",
     "schemas/",
     "scripts/",
@@ -109,6 +110,7 @@
   "piagent": {
     "runtimeIntegrityEntrypoints": [
       "packages/piagent-webui/server/sidecar-main.ts",
+      "scripts/benchmark-runner-core.mjs",
       "scripts/register-typescript-loader.mjs",
       "scripts/typescript-loader.mjs"
     ]

package/packages/piagent-core/benchmark/benchmark-assurance.js

@@ -181,12 +181,15 @@ export function benchmarkClaimEligibility({ suite, environment = {}, baselineSur
   if (baselineSurface === "codex-cli") limitations.push("codex-comparison-is-external-and-non-causal");
   if ((suite.defaultRepeats ?? 0) <= 3) limitations.push("three-repeats-do-not-support-an-always-claim");
   if (assurance.generatedVariants === true && assurance.familyDisjointSplit !== true) limitations.push("generated-value-variants-are-not-independent-task-families");
+  const tokenClaimScope = suite.releaseGate?.primaryEfficiencyEstimand === "fixed-workload-family-ratio"
+    ? "bounded-to-predeclared-fixed-workload-families"
+    : "bounded-to-observed-comparable-pairs-and-failure-aware-effort";
   return {
     declaredTier, achievedTier, comparisonPurpose,
     causalAttributionAllowed: comparisonPurpose === "causal-harness-ablation" && protocolPassed,
     generalizationClaimAllowed: privateHoldoutEligible,
     productionStabilityClaimAllowed: productionShadowEligible,
-    tokenClaimScope: tokenClaimAllowed ? "bounded-to-observed-comparable-pairs-and-failure-aware-effort" : "unavailable",
+    tokenClaimScope: tokenClaimAllowed ? tokenClaimScope : "unavailable",
     privateHoldoutChecks: privateChecks,
     limitations
   };

package/packages/piagent-core/benchmark/benchmark-bootstrap.js

@@ -13,9 +13,14 @@ import {
   recoverBenchmarkPiCredentialWriteback
 } from "./benchmark-pi-home.js";
 import { benchmarkTreeIdentity } from "./benchmark-tree-identity.js";
+import { inspectBenchmarkLedger } from "./benchmark-ledger.js";
+import { completedBenchmarkRecord } from "./benchmark-record-validation.js";
 import { benchmarkGitEnvironment, benchmarkHostEnvironment } from "./benchmark-runtime.js";
+import {
+  canonicalBuiltInBenchmarkSuiteId,
+  isReservedBenchmarkSuiteId
+} from "./benchmark-suite-identity.js";
 
-const builtInSuites = new Set(["core-v1", "capability-v1", "e2-framework-v1", "deep-logic-v1", "production-v1"]);
 const metadataVariable = "PIAGENT_BENCHMARK_BOOTSTRAP_METADATA";
 
 function fail(message) {
@@ -37,6 +42,67 @@ function runRootFromResume(input, cwd) {
   return stat.isDirectory() ? target : path.dirname(target);
 }
 
+function sameLedgerBinding(left, right) {
+  return ["schemaVersion", "algorithm", "digest", "records", "bytes"]
+    .every((field) => left?.[field] === right?.[field]);
+}
+
+function recordMatchesManifestOrder(record, index, manifest) {
+  const expected = manifest.order[index];
+  return completedBenchmarkRecord(record)
+    && record.runId === manifest.runId
+    && record.configurationDigest === manifest.configurationDigest
+    && record.orderIndex === index + 1
+    && record.scenarioId === expected?.scenarioId
+    && record.surface === expected?.surface
+    && record.repeat === expected?.repeat;
+}
+
+function ledgerMatchesManifestPrefix(records, manifest) {
+  const sessions = new Set();
+  const keys = new Set();
+  for (const [index, record] of records.entries()) {
+    const key = `${record?.scenarioId}\0${record?.surface}\0${record?.repeat}`;
+    if (!recordMatchesManifestOrder(record, index, manifest) || sessions.has(record.sessionId) || keys.has(key)) return false;
+    sessions.add(record.sessionId);
+    keys.add(key);
+  }
+  return true;
+}
+
+export function productionResumeHasCompleteLedger(argv, cwd) {
+  const resume = optionValue(argv, "--resume");
+  if (!resume) return false;
+  const runRoot = runRootFromResume(resume, cwd);
+  const manifestPath = path.join(runRoot, "run-manifest.json");
+  if (!fs.existsSync(manifestPath)) return false;
+  const manifest = jsonFile(manifestPath, "benchmark resume manifest");
+  if (manifest?.stageControl?.policy !== "production-v1-provider-spend-v1"
+    || !Array.isArray(manifest.order) || manifest.order.length === 0) return false;
+  const ledgerPath = path.join(runRoot, "runs.jsonl");
+  const ledger = inspectBenchmarkLedger(ledgerPath);
+  if (!ledgerMatchesManifestPrefix(ledger.records, manifest)) return false;
+  if (ledger.records.length === manifest.order.length) return true;
+  if (ledger.records.length !== manifest.order.length - 1) return false;
+
+  const pendingPath = path.join(runRoot, "pending-record.json");
+  if (!fs.existsSync(pendingPath)) return false;
+  const pending = jsonFile(pendingPath, "benchmark pending record");
+  const finalIndex = manifest.order.length - 1;
+  if (pending?.schemaVersion !== 2
+    || pending.postSessionGuard?.matched !== true
+    || !String(pending.postSessionGuard.stage ?? "").startsWith("after-session:")
+    || !sameLedgerBinding(pending.previousLedger, ledger.binding)
+    || !recordMatchesManifestOrder(pending.record, finalIndex, manifest)) return false;
+
+  const measuredPath = path.join(runRoot, "measured-record-ready.json");
+  if (!fs.existsSync(measuredPath)) return true;
+  const measured = jsonFile(measuredPath, "measured benchmark record");
+  return measured?.schemaVersion === 1
+    && sameLedgerBinding(measured.previousLedger, ledger.binding)
+    && JSON.stringify(measured.record) === JSON.stringify(pending.record);
+}
+
 function jsonFile(file, label) {
   try { return JSON.parse(fs.readFileSync(file, "utf8")); }
   catch (error) { fail(`Cannot read ${label} ${file}: ${error.message}`); }
@@ -98,15 +164,22 @@ function snapshotReplay(argv, cwd, temporaryRoot) {
   };
 }
 
-function resolveSuiteManifest(source, cwd) {
-  if (builtInSuites.has(source)) return { origin: source, manifest: null };
+function resolveSuiteManifest(source, cwd, liveRoot) {
+  if (isReservedBenchmarkSuiteId(source)) return { origin: source, manifest: null, builtInId: source };
   const target = path.resolve(cwd, source);
   let stat;
   try { stat = fs.statSync(target); }
   catch (error) { fail(`Cannot snapshot benchmark suite ${target}: ${error.message}`); }
   const manifest = stat.isDirectory() ? path.join(target, "suite.json") : target;
   if (!fs.statSync(manifest).isFile()) fail(`Benchmark suite manifest is not a file: ${manifest}`);
-  return { origin: fs.realpathSync(manifest), manifest: fs.realpathSync(manifest) };
+  const resolvedManifest = fs.realpathSync(manifest);
+  const builtInId = canonicalBuiltInBenchmarkSuiteId(liveRoot, resolvedManifest);
+  if (builtInId) return { origin: builtInId, manifest: null, builtInId };
+  const suite = jsonFile(resolvedManifest, "benchmark suite manifest");
+  if (isReservedBenchmarkSuiteId(suite?.id)) {
+    fail(`Benchmark suite id ${suite.id} is reserved for its canonical built-in suite`);
+  }
+  return { origin: resolvedManifest, manifest: resolvedManifest, builtInId: null };
 }
 
 function inside(root, target) {
@@ -273,12 +346,12 @@ function requestedPiProvider(model, configRoot, credentials) {
   return credentials.length === 1 ? credentials[0].providerId : null;
 }
 
-function snapshotPiAgentHome(temporaryRoot, runtimeParent, argv, cwd, replay) {
+function snapshotPiAgentHome(temporaryRoot, runtimeParent, argv, cwd, replay, { providerFreeFinalization = false } = {}) {
   const sourceRoot = path.resolve(process.env.PI_CODING_AGENT_DIR || path.join(os.homedir(), ".pi", "agent"));
   const configRoot = path.join(temporaryRoot, "pi-agent-config");
   fs.mkdirSync(configRoot, { recursive: true, mode: 0o700 });
   const copied = [];
-  for (const name of ["auth.json", "models.json"]) {
+  for (const name of providerFreeFinalization ? [] : ["auth.json", "models.json"]) {
     const source = path.join(sourceRoot, name);
     const configTarget = path.join(configRoot, name);
     try { copyBenchmarkPiConfigFile(source, configTarget, 0o400); }
@@ -395,10 +468,13 @@ export function createBenchmarkExecutionSnapshot({ liveRoot, argv, cwd }) {
     }
     const candidateIndex = writeBoundJson(path.join(temporaryRoot, "candidate-index.json"), candidate.index);
     const replay = snapshotReplay(argv, cwd, temporaryRoot);
-    const piAgentHome = snapshotPiAgentHome(temporaryRoot, runtimeParent, argv, cwd, replay);
-    const codexCredential = requestsCodexSurface(argv, cwd, replay) ? snapshotCodexCredential(temporaryRoot) : null;
-    const suite = resolveSuiteManifest(requestedSuite(argv, cwd, replay), cwd);
+    const suite = resolveSuiteManifest(requestedSuite(argv, cwd, replay), cwd, root);
     validateOutputIsolation(argv, cwd, root, suite.manifest);
+    const providerFreeFinalization = productionResumeHasCompleteLedger(argv, cwd);
+    const piAgentHome = snapshotPiAgentHome(temporaryRoot, runtimeParent, argv, cwd, replay, { providerFreeFinalization });
+    const codexCredential = !providerFreeFinalization && requestsCodexSurface(argv, cwd, replay)
+      ? snapshotCodexCredential(temporaryRoot)
+      : null;
     let suiteSnapshot = null;
     let suiteRoot;
     if (suite.manifest) {
@@ -421,6 +497,7 @@ export function createBenchmarkExecutionSnapshot({ liveRoot, argv, cwd }) {
       candidateProvenance: candidate.provenance,
       candidateIndex,
       runtimeDependencies: runtimeDependencies(candidateRoot, root),
+      providerFreeFinalization,
       piAgentHome,
       codexCredential,
       replay: replay ? {
@@ -432,6 +509,7 @@ export function createBenchmarkExecutionSnapshot({ liveRoot, argv, cwd }) {
       suite: {
         origin: suite.origin,
         snapshot: suiteSnapshot,
+        builtInId: suite.builtInId,
         identity: benchmarkTreeIdentity(suiteRoot, { rejectSymlinks: true })
       }
     };
@@ -485,6 +563,7 @@ export function benchmarkBootstrapMetadata(env = process.env) {
     || typeof value.candidateIndex?.path !== "string"
     || !/^[a-f0-9]{64}$/.test(String(value.candidateIndex?.digest ?? ""))
     || typeof value.runtimeDependencies?.digest !== "string"
+    || typeof value.providerFreeFinalization !== "boolean"
     || typeof value.piAgentHome?.configRoot !== "string"
     || typeof value.piAgentHome?.runtimeParent !== "string"
     || typeof value.piAgentHome?.vaultRoot !== "string"
@@ -503,6 +582,9 @@ export function benchmarkBootstrapMetadata(env = process.env) {
       || typeof value.codexCredential?.privateIdentity?.contentDigest !== "string"
     ))
     || typeof value.suite?.origin !== "string"
+    || !(value.suite?.builtInId === null || isReservedBenchmarkSuiteId(value.suite?.builtInId))
+    || (value.suite?.builtInId === null && typeof value.suite?.snapshot !== "string")
+    || (value.suite?.builtInId !== null && (value.suite?.origin !== value.suite?.builtInId || value.suite?.snapshot !== null))
     || typeof value.suite?.identity?.contentDigest !== "string"
     || (value.replay !== null && (
       typeof value.replay?.snapshot !== "string"

package/packages/piagent-core/benchmark/benchmark-claim-restrictions.js

@@ -1,21 +1,50 @@
-function diagnostic(report, { tier, purpose, reason, limitation, flag }) {
-  report[flag] = true;
-  report.comparison.purpose = purpose;
+const RESTRICTION_VERDICTS = new Set([
+  "token-claim-withheld",
+  "diagnostic-native-codex",
+  "diagnostic-replay"
+]);
+
+function restrictionCanReplaceVerdict(status, includeObservational) {
+  return typeof status === "string"
+    && (status.endsWith("-more-efficient")
+      || RESTRICTION_VERDICTS.has(status)
+      || (includeObservational && status === "observational-efficiency-only"));
+}
+
+function withholdTokenClaim(report, { reason, limitation, verdictStatus, includeObservational = false }) {
   report.comparison.tokenClaimAllowed = false;
   report.comparison.tokenClaimUnavailableReason = reason;
+  report.comparison.claimEligibility = {
+    ...(report.comparison.claimEligibility ?? {}),
+    tokenClaimScope: "unavailable",
+    limitations: [...new Set([...(report.comparison.claimEligibility?.limitations ?? []), limitation])]
+  };
+  report.verdict ??= {};
+  if (restrictionCanReplaceVerdict(report.verdict.status, includeObservational)) {
+    report.verdict.status = verdictStatus;
+  }
+  report.verdict.claimRestrictionReason = reason;
+}
+
+function diagnostic(report, { tier, purpose, reason, limitation, flag, verdictStatus }) {
+  report[flag] = true;
+  report.comparison.purpose = purpose;
+  withholdTokenClaim(report, { reason, limitation, verdictStatus, includeObservational: true });
   report.comparison.claimEligibility = {
     ...(report.comparison.claimEligibility ?? {}),
     achievedTier: tier,
     generalizationClaimAllowed: false,
-    comparisonPurpose: purpose,
-    limitations: [...new Set([...(report.comparison.claimEligibility?.limitations ?? []), limitation])]
+    comparisonPurpose: purpose
   };
 }
 
 export function applyBenchmarkClaimRestrictions(report, { tokenReason, replaySource, codexMode, surfaces }) {
   if (tokenReason) {
-    report.comparison.tokenClaimAllowed = false;
-    report.comparison.tokenClaimUnavailableReason = tokenReason;
+    withholdTokenClaim(report, {
+      reason: tokenReason,
+      limitation: `token-claim-withheld:${tokenReason}`,
+      verdictStatus: "token-claim-withheld"
+    });
   }
   if (surfaces.includes("codex-cli") && codexMode === "native") {
     diagnostic(report, {
@@ -23,6 +52,7 @@ export function applyBenchmarkClaimRestrictions(report, { tokenReason, replaySou
       purpose: "diagnostic-native-codex",
       reason: "native-codex-operator-configuration-is-not-frozen",
       limitation: "native-codex-results-cannot-support-release-token-or-generalization-claims",
+      verdictStatus: "diagnostic-native-codex",
       flag: "nativeCodexDiagnosticOnly"
     });
   }
@@ -32,6 +62,7 @@ export function applyBenchmarkClaimRestrictions(report, { tokenReason, replaySou
       purpose: "diagnostic-replay",
       reason: "replay-results-are-diagnostic-only",
       limitation: "replay-results-cannot-support-release-token-or-generalization-claims",
+      verdictStatus: "diagnostic-replay",
       flag: "replayDiagnosticOnly"
     });
   }

package/packages/piagent-core/benchmark/benchmark-cli.js

@@ -11,16 +11,16 @@ export const benchmarkUsage = `Usage:
   piagent-benchmark <project-path> --record ...   Legacy recorder
   piagent-benchmark <project-path> --init         Legacy scenario notes
 
-Runs a clean, paired baseline vs Piagent benchmark and grades it automatically.
-With no options it uses the built-in core-v1 suite, the current Pi default
-model/thinking setting, and the suite's repeat count.
+Runs a clean, paired Codex CLI vs Piagent benchmark and grades it automatically.
+With no options it uses the built-in core-v1 suite, GPT-5.6 Luna at medium
+thinking on both surfaces, and the suite's repeat count.
 
 Options:
   --suite <id|path>            core-v1, capability-v1, e2-framework-v1, deep-logic-v1, production-v1, or suite.json path.
-  --deep                       Alias for --suite deep-logic-v1 (6 large scenarios, current model by default).
+  --deep                       Alias for --suite deep-logic-v1 (7 large scenarios, locked to Luna/medium).
   --capability                 Alias for --suite capability-v1.
   --production                 Alias for --suite production-v1.
-  --surfaces <a,b>             raw-pi,piagent (default) or piagent,codex-cli.
+  --surfaces <a,b>             piagent,codex-cli (default); raw-pi is diagnostic ablation only.
   --model <provider/model>     Pin one model identity for both surfaces.
   --thinking <level>           off|minimal|low|medium|high|xhigh|max.
   --codex-mode <mode>          controlled isolated home (default) or native user configuration.
@@ -47,7 +47,7 @@ Options:
 
 Outputs report.json, report.html, summary.txt, and runs.jsonl with mode 0600.
 No token/result value is entered manually; usage comes from Pi session JSONL
-or Codex exec JSONL. Codex CLI comparisons require --model and --thinking.
+or Codex exec JSONL. Override --model and --thinking together for another parity target.
 `;
 
 function fail(message, code = 2) {
@@ -72,7 +72,7 @@ function positiveInteger(raw, name, minimum, maximum) {
 export function parseBenchmarkArgs(argv) {
   const options = {
     suite: "core-v1",
-    surfaces: ["raw-pi", "piagent"],
+    surfaces: ["piagent", "codex-cli"],
     model: undefined,
     thinking: undefined,
     codexMode: "controlled",

package/packages/piagent-core/benchmark/benchmark-codex.js

@@ -9,7 +9,9 @@ export function codexModelName(model) {
 
 export function codexThinkingEffort(thinking) {
   if (!thinking) throw Object.assign(new Error("Codex CLI comparisons require --thinking"), { exitCode: 1 });
-  return thinking === "off" ? "none" : thinking;
+  if (thinking === "off") return "none";
+  if (thinking === "minimal") return "low";
+  return thinking;
 }
 
 export function codexExecArgs({ workspace, options, disabledFeatures = [] }) {