@piagent/platform 1.6.0 → 1.6.1

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

@@ -1,4 +1,5 @@
 import { benchmarkClaimEligibility } from "./benchmark-assurance.js";
+import { integrateCodexRelativeEfficiencyReport } from "./benchmark-codex-relative-efficiency.js";
 import {
   comparableAttemptUsage,
   comparableDuration,
@@ -42,13 +43,13 @@ import {
   surfaceSummary,
   workflowContinuityEvidenceComplete
 } from "./benchmark-summary-support.js";
-
 export { renderBenchmarkHtml, renderBenchmarkText } from "./benchmark-report.js";
 export { benchmarkAssuranceEvidenceValidationErrors, benchmarkClaimEligibility } from "./benchmark-assurance.js";
 export { median } from "./benchmark-statistics.js";
 export { benchmarkSuiteValidationErrors, validateBenchmarkSuite } from "./benchmark-suite.js";
 export { aggregateSessionUsage, benchmarkTokenAccounting, createCodexExecJsonlCollector, parseCodexExecJsonl } from "./benchmark-usage.js";
 export { benchmarkPricingSnapshotValidationErrors, normalizeBenchmarkUsageCost } from "./benchmark-normalized-cost.js";
+export { CODEX_RELATIVE_EFFICIENCY_POLICY, PRODUCTION_SUBAGENT_BUDGET_POLICY, evaluateCodexRelativeEfficiency, summarizeBenchmarkSubagentBudget } from "./benchmark-codex-relative-efficiency.js";
 export { evaluateWorkflowEvidence } from "./benchmark-workflow.js";
 const SURFACE_LABELS = Object.freeze({
   "raw-pi": "Raw Pi",
@@ -737,7 +738,7 @@ export function summarizeBenchmark({
     protocolPassed: protocol.passed,
     tokenClaimAllowed
   });
-  return {
+  return integrateCodexRelativeEfficiencyReport({
     schemaVersion: 2,
     measurementSchemaVersion: BENCHMARK_MEASUREMENT_SCHEMA_VERSION,
     runId,
@@ -1029,5 +1030,5 @@ export function summarizeBenchmark({
       note: `Raw metrics and hidden verifier results are authoritative. Fixed-workload efficiency includes every predeclared paired attempt, including exact provider-started failed-attempt usage, without conditioning token measurement on task outcome; quality and continuity remain independent hard gates. Successful-pair efficiency uses matched ${benchmarkSurfaceLabel(candidateSurface)}/${benchmarkSurfaceLabel(baselineSurface)} resolved-outcome ratios; failure-aware effort divides every comparable attempt by resolved outcomes. Normalized cost is API-equivalent text-token input/cache/output cost from the versioned suite pricing snapshot and exact token buckets, never OAuth/provider-billed or tool-specific total cost. Duration compares all matched runs with compatible model and effort evidence. Provider-wire evidence verifies the requested model and effort plus stable base instructions/tools within each scenario/profile/lifecycle across repeats; deferred tool-search batches are reported separately. Confidence intervals cluster repeats by scenario family. Claim scope is ${claimEligibility.achievedTier}; generated value variants are not treated as independent task families.`
     },
     runs: reportRuns
-  };
+  }, suite);
 }

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

@@ -10,6 +10,7 @@ import {
 } from "./benchmark-candidate.js";
 import { assertBenchmarkLedgerBinding, inspectBenchmarkLedger } from "./benchmark-ledger.js";
 import { benchmarkPhaseAttribution } from "./benchmark-phase-attribution.js";
+import { causalRuntimeEvidence } from "./benchmark-runtime-causal.js";
 
 export { materializeBenchmarkCandidate } from "./benchmark-candidate.js";
 
@@ -709,8 +710,8 @@ export function benchmarkCausalContextReceipt(events, {
     ? "uncompacted"
     : compactedPrompts === prompts.length ? "compacted" : "mixed";
   return {
-    schemaVersion: 2,
-    evidenceSource: "context-telemetry-closed-aggregate-v2",
+    schemaVersion: 3,
+    evidenceSource: "context-telemetry-closed-aggregate-v3",
     applicability: "piagent",
     available,
     coverage: {
@@ -754,6 +755,7 @@ export function benchmarkCausalContextReceipt(events, {
         },
         definition: "matched-edit-recovery-context-receipt-v1"
       },
+      runtimeCausal: causalRuntimeEvidence(visible),
       compaction: { eventsObserved: compactions, state: compactions > 0 ? "observed" : "not-observed" },
       managedPrefix: { promptsObserved: prompts.length, compactedPrompts, state: managedPrefixState }
     } : null

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

@@ -59,7 +59,8 @@ function publicCommandIdentity(value) {
 export function benchmarkPreflightReceipt({
   packageVersion, source, candidateProvenance, suite, suiteDigest,
   runtimeDependencies, runtimeCommands, environmentPolicy, configurationDigest,
-  rootSeedDigest, options, runtime, hostReadinessPolicyDigest = null, hostReadiness = null
+  rootSeedDigest, options, runtime, hostReadinessPolicyDigest = null, hostReadiness = null,
+  providerFreeEvidence = null
 }) {
   return {
     schemaVersion: 1,
@@ -89,6 +90,7 @@ export function benchmarkPreflightReceipt({
       stopAfterFailedPair: options.stopAfterFailedPair
     },
     ...(hostReadiness ? { hostReadiness } : {}),
+    ...(providerFreeEvidence ? { providerFreeEvidence } : {}),
     runtime: {
       gitVersion: runtime.gitVersion,
       piVersion: runtime.piVersion,

package/packages/piagent-core/benchmark/benchmark-provider-free-evidence.js

@@ -0,0 +1,345 @@
+import crypto from "node:crypto";
+import fs from "node:fs";
+import os from "node:os";
+import path from "node:path";
+
+const HASH = /^[a-f0-9]{64}$/;
+const COMMIT = /^[a-f0-9]{40,64}$/;
+const LANE_DEFINITIONS = Object.freeze([
+  Object.freeze({
+    id: "runtime-conformance-v1",
+    configuration: "evals/runtime-conformance-v1/lane.json",
+    runner: "evals/runtime-conformance-v1/runner.mjs",
+    timeoutMilliseconds: 5 * 60_000
+  }),
+  Object.freeze({
+    id: "long-horizon-v1",
+    configuration: "evals/long-horizon-v1/lane.json",
+    runner: "evals/long-horizon-v1/runner.mjs",
+    timeoutMilliseconds: 40 * 60_000
+  }),
+  Object.freeze({
+    id: "webui-parity-v1",
+    configuration: "packages/piagent-webui/benchmark/parity-lane.v1.json",
+    runner: "packages/piagent-webui/benchmark/parity-benchmark.mjs",
+    timeoutMilliseconds: 20 * 60_000
+  })
+]);
+
+function digestBytes(bytes) {
+  return crypto.createHash("sha256").update(bytes).digest("hex");
+}
+
+function fileDigest(root, relative) {
+  return digestBytes(fs.readFileSync(path.join(root, relative)));
+}
+
+function receiptDigest(value) {
+  const copy = structuredClone(value);
+  delete copy.digest;
+  return digestBytes(JSON.stringify(copy));
+}
+
+function validCompletionTime(value) {
+  const milliseconds = typeof value === "string" ? Date.parse(value) : Number.NaN;
+  return Number.isFinite(milliseconds)
+    && new Date(milliseconds).toISOString() === value
+    && milliseconds <= Date.now() + 60_000;
+}
+
+function expectedLaneBindings(packageRoot) {
+  return LANE_DEFINITIONS.map((lane) => ({
+    id: lane.id,
+    configurationPath: lane.configuration,
+    configurationDigest: fileDigest(packageRoot, lane.configuration),
+    runnerPath: lane.runner,
+    runnerDigest: fileDigest(packageRoot, lane.runner)
+  }));
+}
+
+export function productionProviderFreeEvidenceBinding({
+  packageRoot,
+  source,
+  candidateProvenance,
+  configurationDigest
+}) {
+  if (source?.kind !== "git-working-tree" || source.dirty !== false || !COMMIT.test(String(source.commit ?? ""))) {
+    throw new Error("Production provider-free evidence requires an exact clean Git commit");
+  }
+  if (!HASH.test(String(candidateProvenance?.contentDigest ?? ""))) {
+    throw new Error("Production provider-free evidence requires an exact candidate tree digest");
+  }
+  if (!HASH.test(String(configurationDigest ?? ""))) {
+    throw new Error("Production provider-free evidence requires the frozen production configuration digest");
+  }
+  const binding = {
+    schemaVersion: 1,
+    source: {
+      kind: source.kind,
+      commit: source.commit,
+      clean: true,
+      treeDigest: candidateProvenance.contentDigest,
+      treeAlgorithm: candidateProvenance.algorithm
+    },
+    productionConfigurationDigest: configurationDigest,
+    lanes: expectedLaneBindings(packageRoot)
+  };
+  return { ...binding, digest: receiptDigest(binding) };
+}
+
+function lanePassed(id, result) {
+  if (id === "runtime-conformance-v1") {
+    return result?.summary?.passed === true
+      && result?.provider?.used === false
+      && result.provider.calls === 0
+      && result.provider.modelInputTokens === 0
+      && result.provider.modelOutputTokens === 0
+      && Object.values(result?.gates ?? {}).every((value) => value === true);
+  }
+  if (id === "long-horizon-v1") {
+    return result?.evidenceClass === "provider-free-long-horizon"
+      && result?.providerUsed === false
+      && result?.wallClockQualified === true
+      && result?.completedFromResume === true
+      && result?.context?.withinCeiling === true
+      && result?.stateGrowth?.withinCeiling === true
+      && result?.continuation?.enforcementSafe === true
+      && result?.verification?.stableCurrentTree === true;
+  }
+  return result?.benchmark === "webui-parity-v1"
+    && result?.passed === true
+    && result?.providerCalls === 0
+    && result?.modelTokens === 0
+    && result?.invariants?.uiStability === "deterministic-current-state"
+    && result?.invariants?.uiStabilitySuites === 7;
+}
+
+function laneSummary(id, result) {
+  if (id === "runtime-conformance-v1") return {
+    passed: result.summary.passed,
+    configuredCases: result.summary.configuredCases,
+    executedCases: result.summary.executedCases,
+    failedCases: result.summary.failedCases,
+    gates: result.gates
+  };
+  if (id === "long-horizon-v1") return {
+    evidenceClass: result.evidenceClass,
+    wallClockQualified: result.wallClockQualified,
+    completedFromResume: result.completedFromResume,
+    contextWithinCeiling: result.context.withinCeiling,
+    stateGrowthWithinCeiling: result.stateGrowth.withinCeiling,
+    continuationEnforcementSafe: result.continuation.enforcementSafe,
+    stableCurrentTree: result.verification.stableCurrentTree
+  };
+  return {
+    benchmark: result.benchmark,
+    passed: result.passed,
+    uiStability: result.invariants.uiStability,
+    uiStabilitySuites: result.invariants.uiStabilitySuites,
+    deterministicStabilityStepPassed: result.steps.some((step) => step.name === "deterministic-ui-stability" && step.passed === true)
+  };
+}
+
+function summaryPassed(id, summary) {
+  if (id === "runtime-conformance-v1") return summary?.passed === true
+    && summary.configuredCases === summary.executedCases && summary.failedCases === 0
+    && Object.values(summary.gates ?? {}).every((value) => value === true);
+  if (id === "long-horizon-v1") return summary?.evidenceClass === "provider-free-long-horizon"
+    && summary.wallClockQualified === true && summary.completedFromResume === true
+    && summary.contextWithinCeiling === true && summary.stateGrowthWithinCeiling === true
+    && summary.continuationEnforcementSafe === true && summary.stableCurrentTree === true;
+  return summary?.benchmark === "webui-parity-v1" && summary.passed === true
+    && summary.uiStability === "deterministic-current-state"
+    && summary.uiStabilitySuites === 7
+    && summary.deterministicStabilityStepPassed === true;
+}
+
+function privateWrite(file, value) {
+  fs.mkdirSync(path.dirname(file), { recursive: true, mode: 0o700 });
+  const temporary = `${file}.${process.pid}.${Date.now()}.tmp`;
+  fs.writeFileSync(temporary, value, { mode: 0o600 });
+  fs.renameSync(temporary, file);
+  try { fs.chmodSync(file, 0o600); } catch { /* Non-POSIX filesystem. */ }
+}
+
+export function productionProviderFreeEvidenceValidationErrors(receipt, expectedBinding) {
+  const errors = [];
+  if (receipt?.schemaVersion !== 1 || receipt?.kind !== "production-provider-free-evidence-v1") {
+    return ["missing-or-unsupported-provider-free-evidence"];
+  }
+  if (!validCompletionTime(receipt.completedAt)) errors.push("provider-free-completion-time-invalid-or-future");
+  const validExpectedBinding = expectedBinding?.schemaVersion === 1
+    && expectedBinding?.source?.kind === "git-working-tree"
+    && expectedBinding.source.clean === true
+    && COMMIT.test(String(expectedBinding.source.commit ?? ""))
+    && HASH.test(String(expectedBinding.source.treeDigest ?? ""))
+    && HASH.test(String(expectedBinding.productionConfigurationDigest ?? ""))
+    && Array.isArray(expectedBinding.lanes)
+    && expectedBinding.lanes.length === LANE_DEFINITIONS.length
+    && LANE_DEFINITIONS.every((definition, index) => {
+      const lane = expectedBinding.lanes[index];
+      return lane?.id === definition.id
+        && lane.configurationPath === definition.configuration
+        && lane.runnerPath === definition.runner
+        && HASH.test(String(lane.configurationDigest ?? ""))
+        && HASH.test(String(lane.runnerDigest ?? ""));
+    })
+    && expectedBinding.digest === receiptDigest(expectedBinding);
+  if (!validExpectedBinding) return ["invalid-provider-free-expected-binding"];
+  if (receipt?.binding?.digest !== expectedBinding?.digest
+    || JSON.stringify(receipt.binding) !== JSON.stringify(expectedBinding)) errors.push("provider-free-binding-mismatch");
+  if (!Array.isArray(receipt.lanes) || receipt.lanes.length !== LANE_DEFINITIONS.length) {
+    errors.push("provider-free-lane-count-mismatch");
+  } else {
+    for (const expected of expectedBinding.lanes) {
+      const lane = receipt.lanes.find((item) => item?.id === expected.id);
+      if (!lane) { errors.push(`provider-free-lane-missing:${expected.id}`); continue; }
+      if (lane.passed !== true || lane.providerUsed !== false || lane.providerCalls !== 0 || lane.modelTokens !== 0) {
+        errors.push(`provider-free-lane-failed:${expected.id}`);
+      }
+      if (!summaryPassed(expected.id, lane.summary)) errors.push(`provider-free-lane-summary-invalid:${expected.id}`);
+      if (lane.configurationDigest !== expected.configurationDigest
+        || lane.runnerDigest !== expected.runnerDigest
+        || !HASH.test(String(lane.resultDigest ?? ""))) errors.push(`provider-free-lane-binding-mismatch:${expected.id}`);
+    }
+  }
+  if (receipt.digest !== receiptDigest(receipt)) errors.push("provider-free-receipt-digest-mismatch");
+  return errors;
+}
+
+export function productionProviderFreeEvidenceContextValidationErrors(receipt, {
+  source,
+  candidateProvenance,
+  configurationDigest
+} = {}) {
+  const errors = productionProviderFreeEvidenceValidationErrors(receipt, receipt?.binding);
+  const binding = receipt?.binding;
+  if (binding?.source?.kind !== source?.kind
+    || binding?.source?.commit !== source?.commit
+    || binding?.source?.clean !== (source?.dirty === false)) errors.push("provider-free-source-binding-mismatch");
+  if (binding?.source?.treeDigest !== candidateProvenance?.contentDigest
+    || binding?.source?.treeAlgorithm !== candidateProvenance?.algorithm) errors.push("provider-free-tree-binding-mismatch");
+  if (binding?.productionConfigurationDigest !== configurationDigest) errors.push("provider-free-production-configuration-binding-mismatch");
+  return [...new Set(errors)];
+}
+
+export function assertProductionProviderFreeEvidence(receipt, expectedBinding, label = "production provider-free evidence") {
+  const errors = productionProviderFreeEvidenceValidationErrors(receipt, expectedBinding);
+  if (errors.length > 0) throw new Error(`${label} is missing or changed (${errors.join(", ")})`);
+  return receipt;
+}
+
+export async function collectProductionProviderFreeEvidence({
+  packageRoot,
+  liveRoot,
+  runCommand,
+  source,
+  candidateProvenance,
+  configurationDigest
+}) {
+  const binding = productionProviderFreeEvidenceBinding({ packageRoot, source, candidateProvenance, configurationDigest });
+  const cacheRoot = path.join(liveRoot, ".pi", "benchmarks", "provider-free-evidence", binding.digest);
+  const cachePath = path.join(cacheRoot, "receipt.json");
+  if (fs.existsSync(cachePath)) {
+    const cached = JSON.parse(fs.readFileSync(cachePath, "utf8"));
+    if (productionProviderFreeEvidenceValidationErrors(cached, binding).length === 0) return cached;
+  }
+
+  const assertLiveSource = async (stage) => {
+    const commit = await runCommand("git", ["-C", liveRoot, "rev-parse", "HEAD"], { cwd: liveRoot, timeoutMs: 15_000 });
+    const status = await runCommand("git", ["-C", liveRoot, "status", "--porcelain=v1", "--untracked-files=all"], { cwd: liveRoot, timeoutMs: 15_000 });
+    if (commit.code !== 0 || commit.stdout.trim() !== source.commit || status.code !== 0 || status.stdout.trim()) {
+      throw new Error(`Production provider-free evidence source changed at ${stage}`);
+    }
+    for (const lane of binding.lanes) {
+      if (fileDigest(liveRoot, lane.configurationPath) !== lane.configurationDigest
+        || fileDigest(liveRoot, lane.runnerPath) !== lane.runnerDigest) {
+        throw new Error(`Production provider-free lane assets changed at ${stage}: ${lane.id}`);
+      }
+    }
+  };
+
+  const temporaryRoot = fs.mkdtempSync(path.join(os.tmpdir(), "piagent-provider-free-evidence-"));
+  const lanes = [];
+  try {
+    await assertLiveSource("before-lanes");
+    for (const definition of LANE_DEFINITIONS) {
+      const output = path.join(temporaryRoot, `${definition.id}.json`);
+      const result = await runCommand(process.execPath, [path.join(liveRoot, definition.runner), "--output", output], {
+        cwd: liveRoot,
+        timeoutMs: definition.timeoutMilliseconds
+      });
+      if (result.code !== 0) {
+        throw new Error(`Required provider-free lane failed: ${definition.id} (${result.stderr.trim() || result.signal || `exit ${result.code}`})`);
+      }
+      const bytes = fs.readFileSync(output);
+      const parsed = JSON.parse(bytes.toString("utf8"));
+      if (!lanePassed(definition.id, parsed)) throw new Error(`Required provider-free lane returned an incomplete receipt: ${definition.id}`);
+      const expected = binding.lanes.find((lane) => lane.id === definition.id);
+      lanes.push({
+        schemaVersion: 1,
+        id: definition.id,
+        passed: true,
+        providerUsed: false,
+        providerCalls: 0,
+        modelTokens: 0,
+        configurationDigest: expected.configurationDigest,
+        runnerDigest: expected.runnerDigest,
+        resultDigest: digestBytes(bytes),
+        summary: laneSummary(definition.id, parsed)
+      });
+      await assertLiveSource(`after-${definition.id}`);
+    }
+    const receipt = {
+      schemaVersion: 1,
+      kind: "production-provider-free-evidence-v1",
+      completedAt: new Date().toISOString(),
+      binding,
+      lanes
+    };
+    receipt.digest = receiptDigest(receipt);
+    assertProductionProviderFreeEvidence(receipt, binding);
+    privateWrite(cachePath, `${JSON.stringify(receipt, null, 2)}\n`);
+    return receipt;
+  } finally {
+    fs.rmSync(temporaryRoot, { recursive: true, force: true });
+  }
+}
+
+export function productionProviderFreeLaneDefinitions() {
+  return LANE_DEFINITIONS.map((lane) => ({ ...lane }));
+}
+
+export async function prepareProductionProviderFreeEvidence({
+  required,
+  packageRoot,
+  bootstrapMetadata,
+  candidateProvenance,
+  configurationDigest,
+  runCommand,
+  resumedReceipt = undefined
+}) {
+  if (!required) return { receipt: null, binding: null };
+  const binding = productionProviderFreeEvidenceBinding({
+    packageRoot,
+    source: bootstrapMetadata.sourceIdentity,
+    candidateProvenance,
+    configurationDigest
+  });
+  const receipt = resumedReceipt !== undefined
+    ? assertProductionProviderFreeEvidence(resumedReceipt, binding, "resumed production provider-free evidence")
+    : await collectProductionProviderFreeEvidence({
+      packageRoot,
+      liveRoot: bootstrapMetadata.liveRoot,
+      runCommand,
+      source: bootstrapMetadata.sourceIdentity,
+      candidateProvenance,
+      configurationDigest
+    });
+  return { receipt, binding };
+}
+
+export function productionProviderFreeEvidenceError(receipt, binding, label) {
+  try { assertProductionProviderFreeEvidence(receipt, binding, label); return null; }
+  catch (error) { return error; }
+}

package/packages/piagent-core/benchmark/benchmark-record-validation.js

@@ -3,6 +3,7 @@ const causalCoverageLanesV1 = new Set([
   "telemetry-window", "session-lifecycle", "criterion-initial-pack", "pack-lifecycle", "direct-fallback-rereads", "managed-prefix"
 ]);
 const causalCoverageLanesV2 = new Set([...causalCoverageLanesV1, "edit-recovery-context"]);
+const causalCoverageLanesV3 = causalCoverageLanesV2;
 const causalMaximumAggregate = 1_000_000_000;
 
 function exactKeys(value, keys) {
@@ -53,9 +54,75 @@ function validEditRecoveryAggregate(value) {
     : value.injectedChars > 0 && value.injectedEstimatedTokens > 0;
 }
 
+function validRuntimeSection(value, evidenceSource) {
+  return exactKeys(value, ["coverageStatus", "observedEvents", "evidenceSource", "aggregate"])
+    && ["not-observed", "partial", "complete"].includes(value.coverageStatus)
+    && boundedInteger(value.observedEvents)
+    && value.evidenceSource === evidenceSource
+    && (value.coverageStatus === "not-observed"
+      ? value.observedEvents === 0 && value.aggregate === null
+      : value.observedEvents > 0 && (value.coverageStatus === "partial") === (value.aggregate === null));
+}
+
+function validAdaptiveContextRuntime(value) {
+  if (!validRuntimeSection(value, "context-telemetry")) return false;
+  if (value.coverageStatus !== "complete") return true;
+  const aggregate = value.aggregate;
+  const fields = ["projected", "noOp", "cancelled", "deterministicCompactions", "hostSummaryFallbacks",
+    "estimatedSavingsTokens", "minimumSavingsTokens", "minimumSavingsMet", "minimumSavingsNotMet",
+    "governorProviderCalls", "deterministicCompactionProviderCalls", "protocolIntegrity", "definition"];
+  if (!exactKeys(aggregate, fields)
+    || !fields.slice(0, -2).every((field) => boundedInteger(aggregate[field]))
+    || aggregate.definition !== "adaptive-context-runtime-receipt-v1"
+    || !exactKeys(aggregate.protocolIntegrity, ["checks", "intact", "failed", "orphanCalls", "orphanResults"])
+    || !Object.values(aggregate.protocolIntegrity).every(boundedInteger)) return false;
+  const eventCount = aggregate.projected + aggregate.noOp + aggregate.cancelled
+    + aggregate.deterministicCompactions + aggregate.hostSummaryFallbacks;
+  const savingsDecisions = aggregate.minimumSavingsMet + aggregate.minimumSavingsNotMet;
+  const protocol = aggregate.protocolIntegrity;
+  return eventCount === value.observedEvents
+    && savingsDecisions <= eventCount
+    && aggregate.deterministicCompactionProviderCalls === 0
+    && protocol.intact + protocol.failed === protocol.checks
+    && (protocol.failed > 0 || protocol.orphanCalls + protocol.orphanResults === 0);
+}
+
+function validEditFreshnessRuntime(value) {
+  if (!validRuntimeSection(value, "context-telemetry")) return false;
+  if (value.coverageStatus !== "complete") return true;
+  const aggregate = value.aggregate;
+  const fields = ["snapshotObservations", "readSnapshots", "mutationSnapshots", "staleDetections",
+    "enforcedStaleDetections", "staleBlocks", "rereadRecoveries", "mutationRecoveries",
+    "incompleteRecoveries", "definition"];
+  if (!exactKeys(aggregate, fields)
+    || !fields.slice(0, -1).every((field) => boundedInteger(aggregate[field]))
+    || aggregate.definition !== "edit-freshness-runtime-receipt-v1") return false;
+  return aggregate.snapshotObservations + aggregate.staleDetections === value.observedEvents
+    && aggregate.readSnapshots + aggregate.mutationSnapshots === aggregate.snapshotObservations
+    && aggregate.enforcedStaleDetections <= aggregate.staleDetections
+    && aggregate.staleBlocks === aggregate.enforcedStaleDetections
+    && aggregate.rereadRecoveries === aggregate.enforcedStaleDetections
+    && aggregate.mutationRecoveries <= aggregate.rereadRecoveries
+    && aggregate.incompleteRecoveries === 0;
+}
+
+function validDelegatedRuntimeSection(value) {
+  return validRuntimeSection(value, "provider-free-runtime-conformance")
+    && value.coverageStatus === "not-observed";
+}
+
+function validRuntimeCausalAggregate(value) {
+  return exactKeys(value, ["adaptiveContext", "operationLifecycle", "boundedEmission", "editFreshness"])
+    && validAdaptiveContextRuntime(value.adaptiveContext)
+    && validDelegatedRuntimeSection(value.operationLifecycle)
+    && validDelegatedRuntimeSection(value.boundedEmission)
+    && validEditFreshnessRuntime(value.editFreshness);
+}
+
 function validCausalAggregates(value, criterionExpected, schemaVersion) {
   const expectedKeys = ["packCounts", "estimatedTokens", "selectedItemEstimatedTokens", "selectedItemCounts", "criterionInitialPack", "directFallbackRereads", "compaction", "managedPrefix"];
-  if (schemaVersion === 2) expectedKeys.splice(6, 0, "editRecoveryContext");
+  if (schemaVersion >= 2) expectedKeys.splice(6, 0, "editRecoveryContext");
+  if (schemaVersion >= 3) expectedKeys.splice(7, 0, "runtimeCausal");
   if (!exactKeys(value, expectedKeys)
     || !validCausalTriplet(value.packCounts)
     || !validCausalTriplet(value.estimatedTokens)
@@ -74,7 +141,8 @@ function validCausalAggregates(value, criterionExpected, schemaVersion) {
     || !boundedInteger(value.managedPrefix.compactedPrompts)
     || value.managedPrefix.compactedPrompts > value.managedPrefix.promptsObserved
     || !["compacted", "uncompacted", "mixed"].includes(value.managedPrefix.state)
-    || (schemaVersion === 2 && !validEditRecoveryAggregate(value.editRecoveryContext))) return false;
+    || (schemaVersion >= 2 && !validEditRecoveryAggregate(value.editRecoveryContext))
+    || (schemaVersion >= 3 && !validRuntimeCausalAggregate(value.runtimeCausal))) return false;
   const compacted = value.managedPrefix.compactedPrompts;
   const prompts = value.managedPrefix.promptsObserved;
   const expectedState = compacted === 0 ? "uncompacted" : compacted === prompts ? "compacted" : "mixed";
@@ -85,7 +153,7 @@ function validCausalAggregates(value, criterionExpected, schemaVersion) {
 
 export function validBenchmarkCausalContextReceipt(receipt, surface) {
   if (!exactKeys(receipt, ["schemaVersion", "evidenceSource", "applicability", "available", "coverage", "aggregates"])
-    || ![1, 2].includes(receipt.schemaVersion) || typeof receipt.available !== "boolean"
+    || ![1, 2, 3].includes(receipt.schemaVersion) || typeof receipt.available !== "boolean"
     || !exactKeys(receipt.coverage, ["status", "telemetryTruncated", "telemetryIntegrityFailures", "recoverableTailBytes", "criterionExpected", "sessionEventsObserved", "observedLanes", "requiredLanes", "missingLanes"])) return false;
   const coverage = receipt.coverage;
   if (surface !== "piagent") return receipt.schemaVersion === 1 && receipt.evidenceSource === "not-applicable"
@@ -95,7 +163,8 @@ export function validBenchmarkCausalContextReceipt(receipt, surface) {
     && coverage.criterionExpected === false && coverage.sessionEventsObserved === 0
     && coverage.observedLanes === 0 && coverage.requiredLanes === 0
     && Array.isArray(coverage.missingLanes) && coverage.missingLanes.length === 0;
-  const coverageLanes = receipt.schemaVersion === 2 ? causalCoverageLanesV2 : causalCoverageLanesV1;
+  const coverageLanes = receipt.schemaVersion === 3 ? causalCoverageLanesV3
+    : receipt.schemaVersion === 2 ? causalCoverageLanesV2 : causalCoverageLanesV1;
   const evidenceSource = `context-telemetry-closed-aggregate-v${receipt.schemaVersion}`;
   if (receipt.evidenceSource !== evidenceSource || receipt.applicability !== "piagent"
     || typeof coverage.telemetryTruncated !== "boolean" || typeof coverage.criterionExpected !== "boolean"
@@ -120,7 +189,7 @@ export function summarizeBenchmarkCausalContextEvidence(runs, { required = false
     && run.causalContextReceipt.coverage.status === "complete"
     && run.causalContextReceipt.aggregates);
   const fullyCovered = piagentRuns.length > 0 && complete.length === piagentRuns.length;
-  const recoveryComplete = complete.filter((run) => run.causalContextReceipt.schemaVersion === 2);
+  const recoveryComplete = complete.filter((run) => run.causalContextReceipt.schemaVersion >= 2);
   const recoveryFullyCovered = fullyCovered && recoveryComplete.length === piagentRuns.length;
   const sum = (select) => complete.reduce((total, run) => total + select(run.causalContextReceipt.aggregates), 0);
   const sumRecovery = (select) => recoveryComplete.reduce((total, run) => total + select(run.causalContextReceipt.aggregates.editRecoveryContext), 0);

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

@@ -17,6 +17,10 @@ function displayPercent(value) {
   return Number.isFinite(value) ? `${display(value)}%` : "n/a";
 }
 
+function displayRatioPercent(value) {
+  return Number.isFinite(value) ? displayPercent(value * 100) : "n/a";
+}
+
 function topToolSummary(toolNames, limit = 8) {
   const entries = Object.entries(toolNames ?? {}).slice(0, limit);
   return entries.length ? entries.map(([name, count]) => `${name}:${count}`).join(", ") : "none";
@@ -107,6 +111,9 @@ export function renderBenchmarkText(report) {
   const durationConfidence = report.comparison.durationRatioConfidence95;
   const normalizedCost = report.comparison.normalizedCost;
   const normalizedCostConfidence = normalizedCost?.ratioConfidence95;
+  const codexRelative = report.comparison.codexRelativeEfficiency;
+  const codexRelativeTrafficConfidence = codexRelative?.totalTokenTraffic?.ratioConfidence95;
+  const codexRelativeCostConfidence = codexRelative?.apiEquivalentCost?.ratioConfidence95;
   const causalContext = report.comparison.causalContextEvidence;
   const causalAggregates = causalContext?.aggregates;
   const hostReadiness = report.environment.hostReadinessHistory;
@@ -164,6 +171,10 @@ export function renderBenchmarkText(report) {
     `All-successful-pairs descriptive ratio: ${display(report.comparison.allSuccessfulPairsFreshTokenRatio, 4)}`,
     `Normalized API-equivalent text-token cost: ${display(normalizedCost?.ratio, 4)} | 95% CI ${normalizedCostConfidence ? `${display(normalizedCostConfidence.lower, 4)}..${display(normalizedCostConfidence.upper, 4)}` : "n/a"} | delta ${displayPercent(normalizedCost?.deltaPercent)}`,
     `Normalized-cost source: ${normalizedCost?.pricingSnapshot?.id ?? "n/a"} | billed/provider-reported cost: no | exact/applicable pairs ${normalizedCost?.comparablePairs ?? 0}/${normalizedCost?.expectedPairs ?? 0}`,
+    `Codex-relative total token traffic: ${display(codexRelative?.totalTokenTraffic?.ratio, 4)} | 95% CI ${codexRelativeTrafficConfidence ? `${display(codexRelativeTrafficConfidence.lower, 4)}..${display(codexRelativeTrafficConfidence.upper, 4)}` : "n/a"} | required upper <= ${display(codexRelative?.policy?.maximumTotalTokenTrafficRatioUpper95, 2)}`,
+    `Codex-relative API-equivalent cost: ${display(codexRelative?.apiEquivalentCost?.ratio, 4)} | 95% CI ${codexRelativeCostConfidence ? `${display(codexRelativeCostConfidence.lower, 4)}..${display(codexRelativeCostConfidence.upper, 4)}` : "n/a"} | required upper <= ${display(codexRelative?.policy?.maximumApiEquivalentCostRatioUpper95, 2)}`,
+    `Codex-relative subagent budget: ${codexRelative?.subagents?.sessions ?? "n/a"} sessions | ${displayRatioPercent(codexRelative?.subagents?.trafficShare)} traffic share | max ${codexRelative?.policy?.maximumSubagentSessionsPerAttempt ?? "n/a"}/attempt and ${displayRatioPercent(codexRelative?.policy?.maximumSubagentTrafficShare)}`,
+    `Codex-relative efficiency gate: ${report.comparison.codexRelativeEfficiencyGate === null ? "n/a" : report.comparison.codexRelativeEfficiencyGate ? "pass" : `fail (${(codexRelative?.failures ?? []).join(", ") || "missing evidence"})`} | subscription-billed claim: no`,
     `Duration pair wins: ${candidateLabel} ${report.comparison.pairedDurationWins?.[candidateKey] ?? 0} | ${baselineLabel} ${report.comparison.pairedDurationWins?.[baselineKey] ?? 0} | ties ${report.comparison.pairedDurationWins?.ties ?? 0}`,
     `Median paired duration delta: ${display(report.comparison.medianPairedDurationDeltaSeconds, 2)}s (negative favors ${candidateLabel})`,
     `Duration ratio 95% CI: ${durationConfidence ? `${display(durationConfidence.lower, 4)}..${display(durationConfidence.upper, 4)}` : "n/a"}`,
@@ -261,6 +272,9 @@ export function renderBenchmarkHtml(report) {
   const durationConfidence = report.comparison.durationRatioConfidence95;
   const normalizedCost = report.comparison.normalizedCost;
   const normalizedCostConfidence = normalizedCost?.ratioConfidence95;
+  const codexRelative = report.comparison.codexRelativeEfficiency;
+  const codexRelativeTrafficConfidence = codexRelative?.totalTokenTraffic?.ratioConfidence95;
+  const codexRelativeCostConfidence = codexRelative?.apiEquivalentCost?.ratioConfidence95;
   const causalContext = report.comparison.causalContextEvidence;
   const causalAggregates = causalContext?.aggregates;
   const hostReadiness = report.environment.hostReadinessHistory;
@@ -271,7 +285,7 @@ export function renderBenchmarkHtml(report) {
 <html lang="en"><head><meta charset="utf-8"><meta name="viewport" content="width=device-width,initial-scale=1"><title>Piagent Benchmark ${htmlEscape(report.runId)}</title>
 <style>body{font:14px system-ui,sans-serif;color:#202124;max-width:1180px;margin:32px auto;padding:0 20px}h1{font-size:24px}h2{font-size:17px;margin-top:28px}.table-wrap{overflow-x:auto}table{border-collapse:collapse;width:100%;white-space:nowrap}th,td{border:1px solid #d8dadd;padding:8px;text-align:left}thead th{background:#f4f5f6}.metric{display:inline-block;margin:0 24px 8px 0}.note{color:#5f6368}</style></head><body>
 <h1>Piagent Benchmark</h1><p>${htmlEscape(report.suite.title)} · ${htmlEscape(report.runId)}</p><p class="note">Platform v${htmlEscape(report.environment.platformVersion ?? "unknown")} · Pi ${htmlEscape(report.environment.piVersion ?? "unknown")}${report.environment.codexVersion ? ` · Codex ${htmlEscape(report.environment.codexVersion)}` : ""} · Node ${htmlEscape(report.environment.nodeVersion ?? "unknown")} · Baseline ${htmlEscape(report.environment.treatmentBaseline ?? "unknown")} · Piagent treatment ${htmlEscape(report.environment.piagentTreatment?.id ?? "unrecorded")} · Suite ${htmlEscape(report.environment.suiteDigest ?? "unknown")}</p>
-<p class="metric"><strong>Comparison:</strong> ${htmlEscape(candidateLabel)} vs ${htmlEscape(baselineLabel)}</p><p class="metric"><strong>Purpose:</strong> ${htmlEscape(report.comparison.purpose ?? "unspecified")}</p><p class="metric"><strong>Claim tier:</strong> ${htmlEscape(report.comparison.claimEligibility?.achievedTier ?? "unavailable")}</p><p class="metric"><strong>Token claim scope:</strong> ${htmlEscape(report.comparison.claimEligibility?.tokenClaimScope ?? "unavailable")}</p>${report.comparison.tokenClaimUnavailableReason ? `<p class="metric"><strong>Token claim restriction:</strong> ${htmlEscape(report.comparison.tokenClaimUnavailableReason)}</p>` : ""}<p class="metric"><strong>Verdict:</strong> ${htmlEscape(report.verdict.status)}</p><p class="metric"><strong>Protocol:</strong> ${report.comparison.comparisonProtocolGate?.passed ? "PASS" : `FAIL (${htmlEscape((report.comparison.comparisonProtocolGate?.failedChecks ?? []).join(", ") || "missing evidence")})`}</p><p class="metric"><strong>Provider wire:</strong> ${report.comparison.providerWireSurfaceGate === null ? "n/a" : report.comparison.providerWireSurfaceGate ? "PASS" : "FAIL"} (${report.comparison.providerWireEvidence?.verifiedRuns ?? 0}/${report.comparison.providerWireEvidence?.runs ?? 0} runs, ${(report.comparison.providerWireEvidence?.driftGroups ?? []).length} drift groups)</p><p class="metric"><strong>Causal receipt:</strong> ${report.comparison.causalContextEvidenceGate === null ? "n/a" : report.comparison.causalContextEvidenceGate ? "PASS" : "FAIL"} (current v${causalContext?.requiredSchemaVersion ?? "?"} ${causalContext?.currentAvailableRuns ?? 0}/${causalContext?.runs ?? 0} runs)</p><p class="metric"><strong>Host readiness:</strong> ${report.comparison.hostReadinessGate === null ? "n/a" : report.comparison.hostReadinessGate ? "PASS" : "FAIL"} (${hostReadiness?.validReceiptCount ?? 0}/${hostReadiness?.receiptCount ?? 0} valid receipts · ${htmlEscape(hostReadiness?.windowCoverage ?? "n/a")} coverage)</p><p class="metric"><strong>Causal packs:</strong> ${causalAggregates ? `${causalAggregates.packCounts.offered}/${causalAggregates.packCounts.delivered}/${causalAggregates.packCounts.injected} offered/delivered/injected · ${causalAggregates.estimatedTokens.injected} injected estimated tokens · ${causalAggregates.selectedItemCounts.injected} selected items` : "unavailable"}</p><p class="metric"><strong>Causal direct rereads:</strong> ${causalAggregates ? `${causalAggregates.directFallbackRereads.successfulCalls} successful direct-path calls · ${causalAggregates.directFallbackRereads.shellToolCallsObserved} shell calls after injection` : "unavailable"}</p><p class="metric"><strong>Causal edit recovery:</strong> ${causalAggregates?.editRecoveryContext ? `${causalAggregates.editRecoveryContext.count ?? "n/a"} contexts · ${causalAggregates.editRecoveryContext.injectedChars ?? "n/a"} chars · ${causalAggregates.editRecoveryContext.injectedEstimatedTokens ?? "n/a"} estimated tokens · ${htmlEscape(causalAggregates.editRecoveryContext.evidenceCoverage?.status ?? "unavailable")} evidence` : "unavailable"}</p><p class="metric"><strong>Suite gate:</strong> ${report.comparison.suiteGate ? report.comparison.suiteGate.passed ? "PASS" : "FAIL" : "n/a"}</p><p class="metric"><strong>Infrastructure retries:</strong> ${report.infrastructure?.retries ?? 0}</p><p class="metric"><strong>Stability gate:</strong> ${report.comparison.stabilityGate === null ? "n/a" : report.comparison.stabilityGate ? "PASS" : "FAIL"}</p><p class="metric"><strong>Primary estimand:</strong> ${htmlEscape(report.comparison.primaryEfficiencyEstimand ?? "unavailable")}</p><p class="metric"><strong>Primary token delta:</strong> ${displayPercent(report.comparison.primaryEfficiencyDeltaPercent)}</p><p class="metric"><strong>Primary ratio 95% CI:</strong> ${primaryConfidence ? `${display(primaryConfidence.lower, 4)}–${display(primaryConfidence.upper, 4)}` : "n/a"}</p><p class="metric"><strong>Fixed-workload ratio:</strong> ${display(report.comparison.fixedWorkloadFamilyFreshTokenRatio, 4)}${fixedWorkloadConfidence ? ` (${display(fixedWorkloadConfidence.lower, 4)}–${display(fixedWorkloadConfidence.upper, 4)})` : ""}</p><p class="metric"><strong>Fixed-workload coverage:</strong> ${report.comparison.fixedWorkloadFamilyCoverage?.usableScenarioFamilies ?? 0}/${report.comparison.fixedWorkloadFamilyCoverage?.expectedScenarioFamilies ?? 0} families</p><p class="metric"><strong>Successful-pair diagnostic role:</strong> ${htmlEscape(report.comparison.successfulPairEfficiencyRole ?? "blocking")}</p><p class="metric"><strong>Successful-pair token delta:</strong> ${displayPercent(report.comparison.freshTokenDeltaPercent)}</p><p class="metric"><strong>Successful-pair ratio 95% CI:</strong> ${confidence ? `${display(confidence.lower, 4)}–${display(confidence.upper, 4)}` : "n/a"}</p><p class="metric"><strong>Normalized API-equivalent text-token cost ratio:</strong> ${display(normalizedCost?.ratio, 4)}${normalizedCostConfidence ? ` (${display(normalizedCostConfidence.lower, 4)}–${display(normalizedCostConfidence.upper, 4)})` : ""}</p><p class="metric"><strong>Normalized-cost gate:</strong> ${report.comparison.normalizedCostGate === null ? "n/a" : report.comparison.normalizedCostGate ? "PASS" : "FAIL"}</p><p class="metric"><strong>Failure-aware family ratio:</strong> ${display(report.comparison.failureAwareFamilyFreshTokenRatio, 4)}${failureAwareFamilyConfidence ? ` (${display(failureAwareFamilyConfidence.lower, 4)}–${display(failureAwareFamilyConfidence.upper, 4)})` : ""}</p><p class="metric"><strong>Failure-aware pooled ratio:</strong> ${display(report.comparison.failureAwareFreshTokenRatio, 4)}</p><p class="metric"><strong>Duration-ratio delta:</strong> ${displayPercent(report.comparison.durationDeltaPercent)}</p><p class="metric"><strong>Duration ratio 95% CI:</strong> ${durationConfidence ? `${display(durationConfidence.lower, 4)}–${display(durationConfidence.upper, 4)}` : "n/a"}</p><p class="metric"><strong>Performance gate:</strong> ${report.comparison.performanceGate === null ? "n/a" : report.comparison.performanceGate ? "PASS" : "FAIL"}</p><p class="metric"><strong>Provider-reported cost-ratio delta:</strong> ${displayPercent(report.comparison.costDeltaPercent)}</p><p class="metric"><strong>Comparable pairs:</strong> ${report.comparison.pairedUsageRuns}</p><p class="metric"><strong>Scenario families:</strong> ${report.comparison.pairedUsageScenarios ?? 0}</p><p class="metric"><strong>Complete families:</strong> ${report.comparison.pairedCompleteScenarios ?? 0}</p><p class="metric"><strong>Duration families:</strong> ${report.comparison.pairedCompleteDurationScenarios ?? 0}</p><p class="metric"><strong>Outcome families:</strong> ${report.comparison.pairedOutcomeScenarios ?? 0}</p><p class="metric"><strong>Pair wins:</strong> ${htmlEscape(candidateLabel)} ${report.comparison.pairedFreshTokenWins[candidateKey]} · ${htmlEscape(baselineLabel)} ${report.comparison.pairedFreshTokenWins[baselineKey]} · ties ${report.comparison.pairedFreshTokenWins.ties}</p>
+<p class="metric"><strong>Comparison:</strong> ${htmlEscape(candidateLabel)} vs ${htmlEscape(baselineLabel)}</p><p class="metric"><strong>Purpose:</strong> ${htmlEscape(report.comparison.purpose ?? "unspecified")}</p><p class="metric"><strong>Claim tier:</strong> ${htmlEscape(report.comparison.claimEligibility?.achievedTier ?? "unavailable")}</p><p class="metric"><strong>Token claim scope:</strong> ${htmlEscape(report.comparison.claimEligibility?.tokenClaimScope ?? "unavailable")}</p>${report.comparison.tokenClaimUnavailableReason ? `<p class="metric"><strong>Token claim restriction:</strong> ${htmlEscape(report.comparison.tokenClaimUnavailableReason)}</p>` : ""}<p class="metric"><strong>Verdict:</strong> ${htmlEscape(report.verdict.status)}</p><p class="metric"><strong>Protocol:</strong> ${report.comparison.comparisonProtocolGate?.passed ? "PASS" : `FAIL (${htmlEscape((report.comparison.comparisonProtocolGate?.failedChecks ?? []).join(", ") || "missing evidence")})`}</p><p class="metric"><strong>Codex-relative gate:</strong> ${report.comparison.codexRelativeEfficiencyGate === null ? "n/a" : report.comparison.codexRelativeEfficiencyGate ? "PASS" : "FAIL"}</p><p class="metric"><strong>Total token traffic:</strong> ${display(codexRelative?.totalTokenTraffic?.ratio, 4)}${codexRelativeTrafficConfidence ? ` (${display(codexRelativeTrafficConfidence.lower, 4)}–${display(codexRelativeTrafficConfidence.upper, 4)})` : ""} · upper ≤ ${display(codexRelative?.policy?.maximumTotalTokenTrafficRatioUpper95, 2)}</p><p class="metric"><strong>API-equivalent cost:</strong> ${display(codexRelative?.apiEquivalentCost?.ratio, 4)}${codexRelativeCostConfidence ? ` (${display(codexRelativeCostConfidence.lower, 4)}–${display(codexRelativeCostConfidence.upper, 4)})` : ""} · upper ≤ ${display(codexRelative?.policy?.maximumApiEquivalentCostRatioUpper95, 2)}</p><p class="metric"><strong>Subagent traffic:</strong> ${displayRatioPercent(codexRelative?.subagents?.trafficShare)} · ${codexRelative?.subagents?.sessions ?? "n/a"} sessions</p><p class="metric"><strong>Provider wire:</strong> ${report.comparison.providerWireSurfaceGate === null ? "n/a" : report.comparison.providerWireSurfaceGate ? "PASS" : "FAIL"} (${report.comparison.providerWireEvidence?.verifiedRuns ?? 0}/${report.comparison.providerWireEvidence?.runs ?? 0} runs, ${(report.comparison.providerWireEvidence?.driftGroups ?? []).length} drift groups)</p><p class="metric"><strong>Causal receipt:</strong> ${report.comparison.causalContextEvidenceGate === null ? "n/a" : report.comparison.causalContextEvidenceGate ? "PASS" : "FAIL"} (current v${causalContext?.requiredSchemaVersion ?? "?"} ${causalContext?.currentAvailableRuns ?? 0}/${causalContext?.runs ?? 0} runs)</p><p class="metric"><strong>Host readiness:</strong> ${report.comparison.hostReadinessGate === null ? "n/a" : report.comparison.hostReadinessGate ? "PASS" : "FAIL"} (${hostReadiness?.validReceiptCount ?? 0}/${hostReadiness?.receiptCount ?? 0} valid receipts · ${htmlEscape(hostReadiness?.windowCoverage ?? "n/a")} coverage)</p><p class="metric"><strong>Causal packs:</strong> ${causalAggregates ? `${causalAggregates.packCounts.offered}/${causalAggregates.packCounts.delivered}/${causalAggregates.packCounts.injected} offered/delivered/injected · ${causalAggregates.estimatedTokens.injected} injected estimated tokens · ${causalAggregates.selectedItemCounts.injected} selected items` : "unavailable"}</p><p class="metric"><strong>Causal direct rereads:</strong> ${causalAggregates ? `${causalAggregates.directFallbackRereads.successfulCalls} successful direct-path calls · ${causalAggregates.directFallbackRereads.shellToolCallsObserved} shell calls after injection` : "unavailable"}</p><p class="metric"><strong>Causal edit recovery:</strong> ${causalAggregates?.editRecoveryContext ? `${causalAggregates.editRecoveryContext.count ?? "n/a"} contexts · ${causalAggregates.editRecoveryContext.injectedChars ?? "n/a"} chars · ${causalAggregates.editRecoveryContext.injectedEstimatedTokens ?? "n/a"} estimated tokens · ${htmlEscape(causalAggregates.editRecoveryContext.evidenceCoverage?.status ?? "unavailable")} evidence` : "unavailable"}</p><p class="metric"><strong>Suite gate:</strong> ${report.comparison.suiteGate ? report.comparison.suiteGate.passed ? "PASS" : "FAIL" : "n/a"}</p><p class="metric"><strong>Infrastructure retries:</strong> ${report.infrastructure?.retries ?? 0}</p><p class="metric"><strong>Stability gate:</strong> ${report.comparison.stabilityGate === null ? "n/a" : report.comparison.stabilityGate ? "PASS" : "FAIL"}</p><p class="metric"><strong>Primary estimand:</strong> ${htmlEscape(report.comparison.primaryEfficiencyEstimand ?? "unavailable")}</p><p class="metric"><strong>Primary token delta:</strong> ${displayPercent(report.comparison.primaryEfficiencyDeltaPercent)}</p><p class="metric"><strong>Primary ratio 95% CI:</strong> ${primaryConfidence ? `${display(primaryConfidence.lower, 4)}–${display(primaryConfidence.upper, 4)}` : "n/a"}</p><p class="metric"><strong>Fixed-workload ratio:</strong> ${display(report.comparison.fixedWorkloadFamilyFreshTokenRatio, 4)}${fixedWorkloadConfidence ? ` (${display(fixedWorkloadConfidence.lower, 4)}–${display(fixedWorkloadConfidence.upper, 4)})` : ""}</p><p class="metric"><strong>Fixed-workload coverage:</strong> ${report.comparison.fixedWorkloadFamilyCoverage?.usableScenarioFamilies ?? 0}/${report.comparison.fixedWorkloadFamilyCoverage?.expectedScenarioFamilies ?? 0} families</p><p class="metric"><strong>Successful-pair diagnostic role:</strong> ${htmlEscape(report.comparison.successfulPairEfficiencyRole ?? "blocking")}</p><p class="metric"><strong>Successful-pair token delta:</strong> ${displayPercent(report.comparison.freshTokenDeltaPercent)}</p><p class="metric"><strong>Successful-pair ratio 95% CI:</strong> ${confidence ? `${display(confidence.lower, 4)}–${display(confidence.upper, 4)}` : "n/a"}</p><p class="metric"><strong>Normalized API-equivalent text-token cost ratio:</strong> ${display(normalizedCost?.ratio, 4)}${normalizedCostConfidence ? ` (${display(normalizedCostConfidence.lower, 4)}–${display(normalizedCostConfidence.upper, 4)})` : ""}</p><p class="metric"><strong>Normalized-cost gate:</strong> ${report.comparison.normalizedCostGate === null ? "n/a" : report.comparison.normalizedCostGate ? "PASS" : "FAIL"}</p><p class="metric"><strong>Failure-aware family ratio:</strong> ${display(report.comparison.failureAwareFamilyFreshTokenRatio, 4)}${failureAwareFamilyConfidence ? ` (${display(failureAwareFamilyConfidence.lower, 4)}–${display(failureAwareFamilyConfidence.upper, 4)})` : ""}</p><p class="metric"><strong>Failure-aware pooled ratio:</strong> ${display(report.comparison.failureAwareFreshTokenRatio, 4)}</p><p class="metric"><strong>Duration-ratio delta:</strong> ${displayPercent(report.comparison.durationDeltaPercent)}</p><p class="metric"><strong>Duration ratio 95% CI:</strong> ${durationConfidence ? `${display(durationConfidence.lower, 4)}–${display(durationConfidence.upper, 4)}` : "n/a"}</p><p class="metric"><strong>Performance gate:</strong> ${report.comparison.performanceGate === null ? "n/a" : report.comparison.performanceGate ? "PASS" : "FAIL"}</p><p class="metric"><strong>Provider-reported cost-ratio delta:</strong> ${displayPercent(report.comparison.costDeltaPercent)}</p><p class="metric"><strong>Comparable pairs:</strong> ${report.comparison.pairedUsageRuns}</p><p class="metric"><strong>Scenario families:</strong> ${report.comparison.pairedUsageScenarios ?? 0}</p><p class="metric"><strong>Complete families:</strong> ${report.comparison.pairedCompleteScenarios ?? 0}</p><p class="metric"><strong>Duration families:</strong> ${report.comparison.pairedCompleteDurationScenarios ?? 0}</p><p class="metric"><strong>Outcome families:</strong> ${report.comparison.pairedOutcomeScenarios ?? 0}</p><p class="metric"><strong>Pair wins:</strong> ${htmlEscape(candidateLabel)} ${report.comparison.pairedFreshTokenWins[candidateKey]} · ${htmlEscape(baselineLabel)} ${report.comparison.pairedFreshTokenWins[baselineKey]} · ties ${report.comparison.pairedFreshTokenWins.ties}</p>
 <p class="metric"><strong>Observational timing (${htmlEscape(baselineLabel)}):</strong> ${htmlEscape(timingSummary(report.timingDiagnostics, baselineSurface))}</p><p class="metric"><strong>Observational timing (${htmlEscape(candidateLabel)}):</strong> ${htmlEscape(timingSummary(report.timingDiagnostics, candidateSurface))}</p>
 <p class="note">Timing uses privacy-safe monotonic JSONL receipt boundaries only. Missing boundaries remain unavailable; these diagnostics do not alter authoritative duration or any release/spend gate.</p>
 <p class="metric"><strong>Edit-recovery failures:</strong> ${causalAggregates?.editRecoveryContext?.failuresObserved ?? "n/a"} observed · ${causalAggregates?.editRecoveryContext?.suppressedFailures ?? "n/a"} safely suppressed</p>