@piagent/platform 1.6.0 → 1.7.1

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

@@ -1,6 +1,24 @@
+function conditionalBothPassUsageText(report) {
+  const value = report.comparison?.conditionalBothPassFreshUsage;
+  if (!value) return "Conditional both-pass fresh usage: unavailable";
+  return `Conditional both-pass fresh usage: ratio ${display(value.ratio, 4)} | eligible ${value.eligiblePairs}/${value.observedPairs} observed pairs (${value.expectedPairs} expected) | both resolved ${value.bothResolvedPairs} | excluded unresolved ${value.excludedUnresolvedPairs}, usage ${value.excludedUsagePairs} | unpaired accepted runs ${value.unpairedAcceptedRuns}. Accepted attempts only; includes safety refusals; not overall token savings.`;
+}
+
+import { renderBenchmarkGoalHtml, renderBenchmarkGoalMarkdown, renderBenchmarkGoalText } from "./benchmark-goal-report.js";
+
 const SURFACE_LABELS = Object.freeze({ "raw-pi": "Raw Pi", piagent: "Piagent", "codex-cli": "Codex CLI" });
 const SURFACE_REPORT_KEYS = Object.freeze({ "raw-pi": "rawPi", piagent: "piagent", "codex-cli": "codexCli" });
 
+function publicationLines(report) {
+  const value = report.publicationAuthorization;
+  return value ? [
+    `Publication authorization: ${value.status} | claim allowed: ${value.allowed === true ? "yes" : "no"} | launcher stage: ${value.stageId}`,
+    `Publication boundary: ${value.explanation}`,
+    `Management-threshold overshoot: fresh tokens ${value.managementThresholdOvershoot?.freshTokens ?? "pending"}; active wall ms ${value.managementThresholdOvershoot?.activeWallTimeMs ?? "pending"}. No absolute provider-side hard cap is claimed.`,
+    `Launcher publication failures: ${(value.failures ?? []).join(", ") || "none recorded"}`
+  ] : [];
+}
+
 function surfaceLabel(surface) {
   return SURFACE_LABELS[surface] ?? surface;
 }
@@ -17,6 +35,10 @@ function displayPercent(value) {
   return Number.isFinite(value) ? `${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";
@@ -33,6 +55,25 @@ function workflowGapSummary(runs) {
   return topToolSummary(gaps);
 }
 
+function outcomeContractSummary(runs) {
+  const outcomeRuns = (runs ?? []).filter((run) => ["failureClass", "countsTowardQuality",
+    "countsTowardUsage", "runValidity"].every((field) => Object.hasOwn(run, field)));
+  if (outcomeRuns.length === 0) return null;
+  const counts = (field) => Object.entries(outcomeRuns.reduce((result, run) => {
+    const value = String(run[field]);
+    result[value] = (result[value] ?? 0) + 1;
+    return result;
+  }, {})).sort(([left], [right]) => left.localeCompare(right))
+    .map(([value, count]) => `${value}=${count}`).join(", ");
+  return {
+    runs: outcomeRuns.length,
+    quality: outcomeRuns.filter((run) => run.countsTowardQuality === true).length,
+    usage: outcomeRuns.filter((run) => run.countsTowardUsage === true).length,
+    failureClasses: counts("failureClass"),
+    validity: counts("runValidity")
+  };
+}
+
 function pairedOutcomeSummary(value, candidateLabel, baselineLabel) {
   if (!value) return "n/a";
   return `both ${value.bothPass} | ${candidateLabel}-only ${value.candidateOnlyPass} | ${baselineLabel}-only ${value.baselineOnlyPass} | neither ${value.bothFail}`;
@@ -107,14 +148,23 @@ export function renderBenchmarkText(report) {
   const durationConfidence = report.comparison.durationRatioConfidence95;
   const normalizedCost = report.comparison.normalizedCost;
   const normalizedCostConfidence = normalizedCost?.ratioConfidence95;
+  const codexRelative = report.comparison.codexRelativeEfficiency;
+  const codexRelativeTrafficConfidence = codexRelative?.totalTokenTraffic?.ratioConfidence95;
+  const codexRelativeCostConfidence = codexRelative?.apiEquivalentCost?.ratioConfidence95;
   const causalContext = report.comparison.causalContextEvidence;
   const causalAggregates = causalContext?.aggregates;
   const hostReadiness = report.environment.hostReadinessHistory;
+  const allAttemptEfficiency = report.comparison.allAttemptPooledEfficiency;
+  const serviceTier = report.comparison.serviceTierEvidence;
+  const campaignEvidence = report.environment.campaignEvidence;
+  const priorCampaignHistory = campaignEvidence?.priorCampaignHistory;
+  const outcomeSummary = outcomeContractSummary(report.runs);
   const suiteFailureReasons = (report.comparison.suiteGate?.failureReasons ?? [])
     .map((failure) => `${failure.id}: ${failure.message}`)
     .join("; ");
   const lines = [
     `Piagent Benchmark — ${report.suite.title}`,
+    ...publicationLines(report),
     `Run: ${report.runId} | ${report.runCount} sessions | ${report.repeats} repeat(s)`,
     runtimeParts.join(" | "),
     `Treatment baseline: ${report.environment.treatmentBaseline ?? "unknown"}`,
@@ -128,6 +178,11 @@ export function renderBenchmarkText(report) {
       : []),
     `Suite digest: ${report.environment.suiteDigest ?? "unknown"} | Source: ${report.environment.source?.kind ?? "unknown"}${report.environment.source?.dirty === true ? " (dirty)" : ""}`,
     `Infrastructure: ${report.infrastructure?.attempts ?? report.runCount} attempts | ${report.infrastructure?.retries ?? 0} retries across ${report.infrastructure?.retriedRuns ?? 0} measured runs`,
+    conditionalBothPassUsageText(report),
+    ...(outcomeSummary ? [
+      `Failure classes: ${outcomeSummary.failureClasses}`,
+      `Outcome accounting: quality ${outcomeSummary.quality}/${outcomeSummary.runs} | usage ${outcomeSummary.usage}/${outcomeSummary.runs} | validity ${outcomeSummary.validity}`
+    ] : []),
     "",
     "Surface     Resolved   Task grade    Scope     Quality  Safety  Reliability  Workflow  Efficiency  Overall",
     scoreLine(baselineLabel, baseline),
@@ -145,6 +200,14 @@ export function renderBenchmarkText(report) {
     `Piagent workflow gaps: ${workflowGapSummary(report.runs.filter((run) => run.surface === "piagent"))}`,
     `Token accounting: input=fresh input; reasoning is included in output; fresh=input+output; total=input+cache read+cache write+output`,
     `Attempt completeness: accepted ${report.tokenAccounting?.acceptedAttempts.exactAttempts ?? 0}/${report.tokenAccounting?.acceptedAttempts.attempts ?? 0} exact | failed ${report.tokenAccounting?.failedAttempts.exactAttempts ?? 0}/${report.tokenAccounting?.failedAttempts.attempts ?? 0} exact | failed fresh ${display(report.tokenAccounting?.failedAttempts.tokens.fresh, 0)}`,
+    `All-attempt net fresh tokens: ${candidateLabel} ${display(allAttemptEfficiency?.candidate?.freshTokens, 0)} | ${baselineLabel} ${display(allAttemptEfficiency?.baseline?.freshTokens, 0)} | ratio ${display(allAttemptEfficiency?.ratio, 4)} | reduction ${displayPercent(allAttemptEfficiency?.reductionPercent)} | gate ${allAttemptEfficiency?.passed === null ? "n/a" : allAttemptEfficiency?.passed ? "pass" : "fail"}`,
+    ...(campaignEvidence ? [
+      `Current paid campaign: ${campaignEvidence.status} | exact ${campaignEvidence.allAttempts?.exactAttempts ?? 0}/${campaignEvidence.providerStartedAttempts ?? 0} | unknown ${campaignEvidence.unknownAttempts ?? 0} | claim ${campaignEvidence.claimOutcome?.allowed === true ? "allowed" : campaignEvidence.claimOutcome?.allowed === false ? "not allowed" : "pending"}`,
+      `Prior paid campaigns (excluded from current net ratio): ${priorCampaignHistory?.paidCampaigns ?? 0} | provider attempts ${priorCampaignHistory?.providerStartedAttempts ?? 0} | exact ${priorCampaignHistory?.exactAttempts ?? 0} | unknown ${priorCampaignHistory?.unknownAttempts ?? 0} | ${priorCampaignHistory?.exactUsageComplete === true ? `exact known fresh ${display(priorCampaignHistory?.knownExactTokens?.fresh, 0)}` : `total usage unavailable; known-exact fresh subtotal ${display(priorCampaignHistory?.knownExactTokens?.fresh, 0)}`}`
+    ] : []),
+    `Fast execution configuration parity: ${serviceTier?.executionConfigurationParityGate === null ? "n/a" : serviceTier?.executionConfigurationParityGate ? "pass" : "fail"} | claim boundary ${serviceTier?.claimBoundary ?? "unavailable"}`,
+    `Provider response tier: ${serviceTier?.providerResponseEvidenceGate === null ? "not exposed by host" : serviceTier?.providerResponseEvidenceGate ? "verified fast/priority" : "default/invalid observed"} | actual provider-tier claim ${serviceTier?.actualProviderProcessingTierClaimAllowed ? "allowed" : "not allowed"}`,
+    `Fast billing references: ChatGPT credits ${display(serviceTier?.subscriptionCreditAccounting?.referenceRates?.chatgptSubscriptionCredits?.multiplier, 1)}x | API token price ${display(serviceTier?.subscriptionCreditAccounting?.referenceRates?.apiFastTokenPricing?.multiplier, 1)}x | actual billing mode/multiplier unverified | non-blocking | excluded from token ratio`,
     "",
     `Paired successful runs: ${report.comparison.pairedSuccessfulRuns}`,
     `Paired runs with comparable usage: ${report.comparison.pairedUsageRuns}`,
@@ -164,6 +227,10 @@ 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"}`,
@@ -232,7 +299,8 @@ export function renderBenchmarkText(report) {
     `Verdict: ${report.verdict.status}`,
     `Token-saving claim allowed: ${report.comparison.tokenClaimAllowed ? "yes" : "no"}`,
     `Generalization claim allowed: ${report.comparison.claimEligibility?.generalizationClaimAllowed ? "yes" : "no"}`,
-    `Claim limitations: ${(report.comparison.claimEligibility?.limitations ?? []).join(", ") || "none"}`
+    `Claim limitations: ${(report.comparison.claimEligibility?.limitations ?? []).join(", ") || "none"}`,
+    ...(report.goalAssessment ? ["", renderBenchmarkGoalText(report.goalAssessment)] : [])
   ];
   return `${lines.join("\n")}\n`;
 }
@@ -241,6 +309,14 @@ function htmlEscape(value) {
   return String(value ?? "").replaceAll("&", "&amp;").replaceAll("<", "&lt;").replaceAll(">", "&gt;").replaceAll('"', "&quot;");
 }
 
+export function renderBenchmarkMarkdown(report) {
+  const detail = renderBenchmarkText({ ...report, goalAssessment: undefined });
+  // A dynamic fence preserves arbitrary diagnostic text without markdown injection.
+  const fence = "`".repeat(Math.max(3, ...[...detail.matchAll(/`+/g)].map(match => match[0].length + 1)));
+  return `# Piagent benchmark report\n\n${renderBenchmarkGoalMarkdown(report.goalAssessment)}\n`
+    + `## Existing benchmark detail\n\n${fence}text\n${detail}${fence}\n`;
+}
+
 export function renderBenchmarkHtml(report) {
   const baselineSurface = report.comparison.baselineSurface ?? "raw-pi";
   const candidateSurface = report.comparison.candidateSurface ?? "piagent";
@@ -249,7 +325,7 @@ export function renderBenchmarkHtml(report) {
   const surfaceEntries = [[baselineSurface, report.surfaces[baselineKey]], [candidateSurface, report.surfaces[candidateKey]]];
   const baselineLabel = surfaceLabel(baselineSurface);
   const candidateLabel = surfaceLabel(candidateSurface);
-  const rows = report.runs.map((run) => `<tr><td>${htmlEscape(run.scenarioId)}</td><td>${htmlEscape(run.category ?? "unspecified")}</td><td>${htmlEscape(run.difficulty ?? "unspecified")}</td><td>${htmlEscape(run.profile ?? "unspecified")}</td><td>${htmlEscape(run.lifecycle ?? "unspecified")}</td><td>${htmlEscape(run.surface)}</td><td>${run.repeat}</td><td>${run.infrastructureRetries ?? 0}</td><td>${run.resolved ? "PASS" : "FAIL"}</td><td>${run.grade?.passed ? "PASS" : "FAIL"}</td><td>${run.scope?.passed ? "PASS" : "FAIL"}</td><td>${display(run.workflow?.score)}</td><td>${htmlEscape((run.workflow?.checks ?? []).filter((check) => !check.passed).map((check) => check.id).join(", ") || "none")}</td><td>${htmlEscape(run.usage?.model ?? "unknown")}</td><td>${htmlEscape(run.usage?.thinkingLevel ?? "unknown")}</td><td>${htmlEscape(run.usage?.usageSource ?? "unknown")}</td><td>${display(run.usage?.input, 0)}</td><td>${display(run.usage?.output, 0)}</td><td>${display(run.usage?.cacheRead, 0)}</td><td>${display(run.usage?.cacheWrite, 0)}</td><td>${display(run.usage?.reasoning, 0)}</td><td>${display(run.usage?.fresh, 0)}</td><td>${display(run.usage?.total, 0)}</td><td>${display(run.usage?.toolCalls, 0)}</td><td>${htmlEscape(topToolSummary(run.usage?.toolNames, 5))}</td><td>${display(run.usage?.cost, 6)}</td><td>${display(run.durationSeconds, 1)}</td><td>${htmlEscape(run.failure ?? "")}</td></tr>`).join("");
+  const rows = report.runs.map((run) => `<tr><td>${htmlEscape(run.scenarioId)}</td><td>${htmlEscape(run.category ?? "unspecified")}</td><td>${htmlEscape(run.difficulty ?? "unspecified")}</td><td>${htmlEscape(run.profile ?? "unspecified")}</td><td>${htmlEscape(run.lifecycle ?? "unspecified")}</td><td>${htmlEscape(run.surface)}</td><td>${run.repeat}</td><td>${run.infrastructureRetries ?? 0}</td><td>${run.resolved ? "PASS" : "FAIL"}</td><td>${run.grade?.passed ? "PASS" : "FAIL"}</td><td>${run.scope?.passed ? "PASS" : "FAIL"}</td><td>${htmlEscape(run.failureClass ?? "n/a")}</td><td>${run.countsTowardQuality === true ? "yes" : run.countsTowardQuality === false ? "no" : "n/a"}</td><td>${run.countsTowardUsage === true ? "yes" : run.countsTowardUsage === false ? "no" : "n/a"}</td><td>${htmlEscape(run.runValidity ?? "n/a")}</td><td>${display(run.workflow?.score)}</td><td>${htmlEscape((run.workflow?.checks ?? []).filter((check) => !check.passed).map((check) => check.id).join(", ") || "none")}</td><td>${htmlEscape(run.usage?.model ?? "unknown")}</td><td>${htmlEscape(run.usage?.thinkingLevel ?? "unknown")}</td><td>${htmlEscape(run.usage?.usageSource ?? "unknown")}</td><td>${display(run.usage?.input, 0)}</td><td>${display(run.usage?.output, 0)}</td><td>${display(run.usage?.cacheRead, 0)}</td><td>${display(run.usage?.cacheWrite, 0)}</td><td>${display(run.usage?.reasoning, 0)}</td><td>${display(run.usage?.fresh, 0)}</td><td>${display(run.usage?.total, 0)}</td><td>${display(run.usage?.toolCalls, 0)}</td><td>${htmlEscape(topToolSummary(run.usage?.toolNames, 5))}</td><td>${display(run.usage?.cost, 6)}</td><td>${display(run.durationSeconds, 1)}</td><td>${htmlEscape(run.failure ?? "")}</td></tr>`).join("");
   const scoreRows = surfaceEntries.map(([id, surface]) => `<tr><th>${htmlEscape(surfaceLabel(id))}</th><td>${surface.resolved}/${surface.runs}</td><td>${surface.qualityCorrect}/${surface.qualityRuns}</td><td>${surface.scopePassed}/${surface.runs}</td><td>${display(surface.scores.quality)}</td><td>${display(surface.scores.safety)}</td><td>${display(surface.scores.reliability)}</td><td>${display(surface.scores.workflow)}</td><td>${display(surface.scores.efficiency)}</td><td>${display(surface.scores.overall)}</td></tr>`).join("");
   const usageRows = surfaceEntries.map(([id, surface]) => { const usage = surface.usage.allMeasuredRuns; return `<tr><th>${htmlEscape(surfaceLabel(id))}</th><td>${display(usage.medianInputTokens, 0)}</td><td>${display(usage.medianOutputTokens, 0)}</td><td>${display(usage.medianCacheReadTokens, 0)}</td><td>${display(usage.medianCacheWriteTokens, 0)}</td><td>${display(usage.medianReasoningTokens, 0)}</td><td>${display(usage.medianFreshTokens, 0)}</td><td>${display(usage.medianTotalTokens, 0)}</td><td>${display(usage.medianToolCalls, 0)}</td><td>${display(usage.medianCost, 6)}</td><td>${display(usage.medianDurationSeconds, 1)}</td><td>${htmlEscape(topToolSummary(surface.usage.toolNames))}</td></tr>`; }).join("");
   const categoryRows = Object.entries(report.surfaces[candidateKey].bands?.categories ?? {}).map(([name, band]) => `<tr><th>${htmlEscape(name)}</th><td>${band.resolved}/${band.runs}</td><td>${display(band.score)}</td><td>${display(band.correctness)}</td></tr>`).join("");
@@ -261,9 +337,16 @@ 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;
+  const allAttemptEfficiency = report.comparison.allAttemptPooledEfficiency;
+  const serviceTier = report.comparison.serviceTierEvidence;
+  const campaignEvidence = report.environment.campaignEvidence;
+  const priorCampaignHistory = campaignEvidence?.priorCampaignHistory;
   const suiteFailureReasons = (report.comparison.suiteGate?.failureReasons ?? [])
     .map((failure) => `${failure.id}: ${failure.message}`)
     .join("; ");
@@ -271,17 +354,23 @@ 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 ${report.comparison.pairedFreshTokenWins.ties}</p>
+${publicationLines(report).map(line => `<p class="note">${htmlEscape(line)}</p>`).join("\n")}
+<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> ${htmlEscape(timingSummary(report.timingDiagnostics, candidateSurface))}</p>
+<p class="note">${htmlEscape(conditionalBothPassUsageText(report))}</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>
-<p class="metric"><strong>Exact attempts:</strong> accepted ${report.tokenAccounting?.acceptedAttempts.exactAttempts ?? 0}/${report.tokenAccounting?.acceptedAttempts.attempts ?? 0} · failed ${report.tokenAccounting?.failedAttempts.exactAttempts ?? 0}/${report.tokenAccounting?.failedAttempts.attempts ?? 0}</p><p class="metric"><strong>Failed-attempt fresh:</strong> ${display(report.tokenAccounting?.failedAttempts.tokens.fresh, 0)}</p>
+	<p class="metric"><strong>Exact attempts:</strong> accepted ${report.tokenAccounting?.acceptedAttempts.exactAttempts ?? 0}/${report.tokenAccounting?.acceptedAttempts.attempts ?? 0} · failed ${report.tokenAccounting?.failedAttempts.exactAttempts ?? 0}/${report.tokenAccounting?.failedAttempts.attempts ?? 0}</p><p class="metric"><strong>Failed-attempt fresh:</strong> ${display(report.tokenAccounting?.failedAttempts.tokens.fresh, 0)}</p>
+	<p class="metric"><strong>All-attempt net fresh:</strong> ${htmlEscape(candidateLabel)} ${display(allAttemptEfficiency?.candidate?.freshTokens, 0)} · ${htmlEscape(baselineLabel)} ${display(allAttemptEfficiency?.baseline?.freshTokens, 0)} · ratio ${display(allAttemptEfficiency?.ratio, 4)} · reduction ${displayPercent(allAttemptEfficiency?.reductionPercent)} · ${allAttemptEfficiency?.passed === null ? "n/a" : allAttemptEfficiency?.passed ? "PASS" : "FAIL"}</p>
+	${campaignEvidence ? `<p class="metric"><strong>Current paid campaign:</strong> ${htmlEscape(campaignEvidence.status)} · exact ${campaignEvidence.allAttempts?.exactAttempts ?? 0}/${campaignEvidence.providerStartedAttempts ?? 0} · unknown ${campaignEvidence.unknownAttempts ?? 0} · claim ${campaignEvidence.claimOutcome?.allowed === true ? "allowed" : campaignEvidence.claimOutcome?.allowed === false ? "not allowed" : "pending"}</p><p class="metric"><strong>Prior paid campaigns (excluded from current net ratio):</strong> ${priorCampaignHistory?.paidCampaigns ?? 0} · provider attempts ${priorCampaignHistory?.providerStartedAttempts ?? 0} · exact ${priorCampaignHistory?.exactAttempts ?? 0} · unknown ${priorCampaignHistory?.unknownAttempts ?? 0} · ${priorCampaignHistory?.exactUsageComplete === true ? `exact known fresh ${display(priorCampaignHistory?.knownExactTokens?.fresh, 0)}` : `total usage unavailable; known-exact fresh subtotal ${display(priorCampaignHistory?.knownExactTokens?.fresh, 0)}`}</p>` : ""}
+	<p class="metric"><strong>Fast execution configuration parity:</strong> ${serviceTier?.executionConfigurationParityGate === null ? "n/a" : serviceTier?.executionConfigurationParityGate ? "PASS" : "FAIL"} · boundary ${htmlEscape(serviceTier?.claimBoundary ?? "unavailable")}</p><p class="metric"><strong>Provider response tier:</strong> ${serviceTier?.providerResponseEvidenceGate === null ? "not exposed by host" : serviceTier?.providerResponseEvidenceGate ? "verified fast/priority" : "default/invalid observed"}</p><p class="metric"><strong>Fast billing references:</strong> ChatGPT credits ${display(serviceTier?.subscriptionCreditAccounting?.referenceRates?.chatgptSubscriptionCredits?.multiplier, 1)}x · API token price ${display(serviceTier?.subscriptionCreditAccounting?.referenceRates?.apiFastTokenPricing?.multiplier, 1)}x · actual billing mode/multiplier unverified · non-blocking · excluded from token ratio</p>
 <p class="note"><strong>Token definitions:</strong> Input is fresh input only. Reasoning is already included in output. Fresh = input + output. Total = input + cache read + cache write + output. Normalized API-equivalent cost uses the versioned suite pricing snapshot and is not OAuth/provider-billed cost.</p>
 ${suiteFailureReasons ? `<p><strong>Suite gate failures:</strong> ${htmlEscape(suiteFailureReasons)}</p>` : ""}
+${renderBenchmarkGoalHtml(report.goalAssessment)}
 <h2>Score bands</h2><div class="table-wrap"><table><thead><tr><th>Surface</th><th>Resolved</th><th>Task grader</th><th>Scope</th><th>Quality</th><th>Safety</th><th>Reliability</th><th>Workflow</th><th>Efficiency</th><th>Overall</th></tr></thead><tbody>${scoreRows}</tbody></table></div>
 ${categoryRows ? `<h2>${htmlEscape(candidateLabel)} category bands</h2><div class="table-wrap"><table><thead><tr><th>Category</th><th>Resolved</th><th>Score</th><th>Correctness</th></tr></thead><tbody>${categoryRows}</tbody></table></div>` : ""}
 ${tokenBandRows ? `<h2>Paired fresh-token ratio by category</h2><div class="table-wrap"><table><thead><tr><th>Category</th><th>Pairs</th><th>Families</th><th>Ratio</th><th>Median delta</th><th>${htmlEscape(candidateLabel)}/${htmlEscape(baselineLabel)}/ties</th></tr></thead><tbody>${tokenBandRows}</tbody></table></div>` : ""}
 <h2>Median usage across all measured runs</h2><div class="table-wrap"><table><thead><tr><th>Surface</th><th>Fresh input</th><th>Output</th><th>Cache read</th><th>Cache write</th><th>Reasoning ⊂ output</th><th>Fresh</th><th>Total</th><th>Tools</th><th>Provider-reported cost</th><th>Seconds</th><th>Top tools</th></tr></thead><tbody>${usageRows}</tbody></table></div>
-<h2>Runs</h2><div class="table-wrap"><table><thead><tr><th>Scenario</th><th>Category</th><th>Difficulty</th><th>Profile</th><th>Lifecycle</th><th>Surface</th><th>Repeat</th><th>Infra retries</th><th>Resolved</th><th>Grader</th><th>Scope</th><th>Workflow</th><th>Workflow gaps</th><th>Model</th><th>Thinking</th><th>Usage source</th><th>Fresh input</th><th>Output</th><th>Cache read</th><th>Cache write</th><th>Reasoning ⊂ output</th><th>Fresh</th><th>Total</th><th>Tools</th><th>Top tools</th><th>Provider-reported cost</th><th>Seconds</th><th>Failure</th></tr></thead><tbody>${rows}</tbody></table></div>
+<h2>Runs</h2><div class="table-wrap"><table><thead><tr><th>Scenario</th><th>Category</th><th>Difficulty</th><th>Profile</th><th>Lifecycle</th><th>Surface</th><th>Repeat</th><th>Infra retries</th><th>Resolved</th><th>Grader</th><th>Scope</th><th>Failure class</th><th>Counts toward quality</th><th>Counts toward usage</th><th>Run validity</th><th>Workflow</th><th>Workflow gaps</th><th>Model</th><th>Thinking</th><th>Usage source</th><th>Fresh input</th><th>Output</th><th>Cache read</th><th>Cache write</th><th>Reasoning ⊂ output</th><th>Fresh</th><th>Total</th><th>Tools</th><th>Top tools</th><th>Provider-reported cost</th><th>Seconds</th><th>Failure</th></tr></thead><tbody>${rows}</tbody></table></div>
 <p class="note">${htmlEscape(report.verdict.note)}</p></body></html>\n`;
 }