enigma-memory 0.1.15 → 0.1.17

package/scripts/build-production-workplan.mjs

@@ -234,6 +234,8 @@ export function buildProductionWorkplan(inputs = {}, options = {}) {
   const endpointRefs = missingHostedEndpointRefs(hosted);
   const stateBlockers = hostedStateBlockers(hosted);
 
+  const finalDependencyCommand = 'npm run production:dependencies -- --goal-audit .enigma/goal-audit-current.json --release-audit .enigma/release-audit-current.json --worker-inspect .enigma/worker-inspect-result-current.json --whitepaper .enigma/whitepaper-claims-current.json --cloudflare-credentials .enigma/cloudflare-credentials-current.json --edge-deploy .enigma/edge-backend-deployment-current.json --edge-live .enigma/edge-backend-bootstrap-live-current.json --storage-bootstrap .enigma/cloudflare-storage-bootstrap-current.json';
+
   const phases = [
     makePhase({
       id: 'cloudflare_credentials',
@@ -309,7 +311,7 @@ export function buildProductionWorkplan(inputs = {}, options = {}) {
       owner: 'operator-or-reviewer',
       prerequisites: ['cloudflare_credentials', 'cloudflare_worker_permission', 'hosted_backend_refs', 'operator_acceptance', 'release_gates'],
       blockers: dependencyReport.launch_ready === true ? [] : ['launch_ready is false'],
-      commands: [release.next_command, staticSite.next_command, whitepaper.next_command, 'npm run production:goal-audit -- --site <public-site-dir> --domain enigmamemory.com --release-audit .enigma/release-audit-current.json', 'npm run production:dependencies -- --goal-audit .enigma/goal-audit-current.json --release-audit .enigma/release-audit-current.json --worker-inspect .enigma/worker-inspect-validation-current.json --whitepaper .enigma/whitepaper-claims-current.json --cloudflare-credentials .enigma/cloudflare-credentials-current.json --edge-deploy .enigma/edge-backend-deployment-current.json --edge-live .enigma/edge-backend-bootstrap-live-current.json --storage-bootstrap .enigma/cloudflare-storage-bootstrap-current.json'],
+      commands: [release.next_command, staticSite.next_command, whitepaper.next_command, 'npm run production:goal-audit -- --site <public-site-dir> --domain enigmamemory.com --release-audit .enigma/release-audit-current.json', finalDependencyCommand],
       evidence: [...release.evidence, ...staticSite.evidence, ...whitepaper.evidence],
       details: { goal_complete: dependencyReport.goal_complete === true, launch_ready: dependencyReport.launch_ready === true },
     }),

package/scripts/build-proof-network-packet.mjs

@@ -15,7 +15,7 @@ import {
   validateProofNetworkPacket,
 } from '../packages/proof-network/src/index.js';
 
-export const PROOF_NETWORK_PACKET_RELEASE_TARGET = '0.1.15';
+export const PROOF_NETWORK_PACKET_RELEASE_TARGET = '0.1.17';
 
 const HASH_RE = /^(?:sha256:)?[a-f0-9]{64}$/iu;
 const SECRET_VALUE_RE = /(?:Bearer\s+[A-Za-z0-9._~+/=-]{12,}|Basic\s+[A-Za-z0-9+/=-]{12,}|-----BEGIN [A-Z ]*PRIVATE KEY-----|https?:\/\/[^\s/@]+:[^\s/@]+@|sk-[A-Za-z0-9_-]{16,}|AKIA[0-9A-Z]{16}|\b(?:raw[\s_-]*memory|plaintext[\s_-]*prompts?|plain[\s_-]*text[\s_-]*prompts?|private[\s_-]*prompts?|provider[\s_-]*responses?|full[\s_-]*transcript|decrypted[\s_-]*memory|credentials?|secrets?|passwords?|private[\s_-]*keys?|api[\s_-]*key[\s_-]*(?:secret|material|value)|api[\s_-]*secrets?|access[\s_-]*tokens?|refresh[\s_-]*tokens?|token[\s_-]*values?|credential[\s_-]*material|tenant[\s_-]*names?)\b)/iu;

package/scripts/check.mjs

@@ -3,7 +3,9 @@ import path from 'node:path';
 import { fileURLToPath, pathToFileURL } from 'node:url';
 
 const root = path.resolve(path.dirname(fileURLToPath(import.meta.url)), '..');
-const packageJsonPath = path.join(root, 'package.json');
+const packageJsonPath = process.env.ENIGMA_CHECK_PACKAGE_JSON_OVERRIDE
+  ? path.resolve(process.env.ENIGMA_CHECK_PACKAGE_JSON_OVERRIDE)
+  : path.join(root, 'package.json');
 const packageJson = JSON.parse(fs.readFileSync(packageJsonPath, 'utf8'));
 const productionRoots = [
   'packages/adapters/src/',

package/scripts/cloudflare-ops.mjs

@@ -6,6 +6,7 @@ import { dirname, join } from 'node:path';
 import { pathToFileURL } from 'node:url';
 import { promisify } from 'node:util';
 import { applyCloudflareSecretEnvFileFromArgv, CloudflareSecretEnvError } from './cloudflare-secret-env.mjs';
+import { PUBLIC_SITE_SECURITY_RESULT_SCHEMA, validatePublicSiteSecurity } from './validate-public-site-security.mjs';
 
 const execFile = promisify(execFileCallback);
 
@@ -46,6 +47,8 @@ Commands:
 
   pages deploy --site <dir> --project-name <name> [--execute]
       Without --execute, prints the exact Wrangler deploy plan only.
+      Dry-run output includes local public-site security validation.
+      --execute refuses artifacts with local security blockers before invoking Wrangler.
       With --execute, runs Wrangler through npm exec/npx without printing the token.
 
   pages verify --url <https-url> --project-name <name> [--domain <host>] \\
@@ -264,6 +267,35 @@ function redactPlanOutput(plan) {
   return redactOperationalPayload(plan);
 }
 
+function redactPublicSiteSecurityBlocker(entry) {
+  return {
+    message: redactOperationalText(entry?.message ?? ''),
+    ...(entry?.path === undefined ? {} : { path: redactOperationalText(String(entry.path)) }),
+  };
+}
+
+function publicSiteSecurityDeploySummary(result) {
+  return {
+    schema: PUBLIC_SITE_SECURITY_RESULT_SCHEMA,
+    ok: result?.ok === true,
+    status: result?.status ?? 'blocked',
+    blocker_count: Array.isArray(result?.blockers) ? result.blockers.length : 0,
+    blockers: Array.isArray(result?.blockers) ? result.blockers.map(redactPublicSiteSecurityBlocker) : [],
+    checked: redactOperationalPayload(result?.checked ?? {}),
+    claimBoundary: result?.claim_boundary ?? [],
+  };
+}
+
+function publicSiteSecurityErrorMessage(summary) {
+  const blockerText = summary.blockers
+    .slice(0, 5)
+    .map((entry) => (entry.path ? `${entry.path}: ${entry.message}` : entry.message))
+    .join('; ');
+  return blockerText.length > 0
+    ? `public site security validation blocked Pages deploy: ${blockerText}`
+    : 'public site security validation blocked Pages deploy';
+}
+
 function parsePositiveInteger(value, name) {
   const text = requireNonEmptyString(name, value);
   if (!/^\d+$/.test(text)) throw new UsageError(`${name} must be a positive integer`);
@@ -1518,6 +1550,7 @@ export async function runCloudflareOpsCommand(command, {
 
   if (command.kind === 'pages.deploy') {
     const plan = buildWranglerPagesDeployPlan(command);
+    const siteSecurity = publicSiteSecurityDeploySummary(await validatePublicSiteSecurity({ site: command.site }));
     if (!command.execute) {
       return {
         json: {
@@ -1526,10 +1559,12 @@ export async function runCloudflareOpsCommand(command, {
           dryRun: true,
           execute: false,
           plan: redactPlanOutput(plan),
+          siteSecurity,
           claimBoundary: 'Plan only; no Cloudflare Pages deployment was executed.',
         },
       };
     }
+    if (!siteSecurity.ok) throw new UsageError(publicSiteSecurityErrorMessage(siteSecurity));
     const result = await execFileImpl(plan.command, plan.args, { shell: plan.usesShell === true, windowsHide: true, maxBuffer: 10 * 1024 * 1024, env });
     return {
       json: {

package/scripts/collect-hosted-backend-live-evidence.mjs

@@ -1,4 +1,5 @@
 #!/usr/bin/env node
+import https from 'node:https';
 import { createHash } from 'node:crypto';
 import { mkdir, readFile, writeFile } from 'node:fs/promises';
 import { dirname, resolve } from 'node:path';
@@ -81,6 +82,42 @@ function redactProbeBody(value, path = 'probe.body') {
   return value;
 }
 
+export function localSimulationLoopbackFetch(url, init = {}) {
+  const parsed = new URL(url);
+  const host = parsed.hostname.toLowerCase();
+  if (parsed.protocol !== 'https:' || (host !== 'sim.enigmamemory.com' && !host.endsWith('.sim.enigmamemory.com'))) {
+    throw new Error('--local-simulation-loopback only supports https://*.sim.enigmamemory.com simulation probes');
+  }
+  const request = {
+    hostname: '127.0.0.1',
+    port: parsed.port || 443,
+    path: `${parsed.pathname}${parsed.search}`,
+    method: init.method || 'GET',
+    headers: init.headers,
+    rejectUnauthorized: false,
+    servername: parsed.hostname,
+  };
+  return new Promise((resolve, reject) => {
+    const req = https.request(request, (res) => {
+      const chunks = [];
+      res.on('data', (chunk) => chunks.push(chunk));
+      res.on('end', () => {
+        const text = Buffer.concat(chunks).toString('utf8');
+        resolve({
+          ok: res.statusCode >= 200 && res.statusCode < 300,
+          status: res.statusCode,
+          statusText: res.statusMessage || '',
+          url,
+          redirected: false,
+          text: async () => text,
+        });
+      });
+    });
+    req.on('error', reject);
+    req.end();
+  });
+}
+
 async function fetchProbe(url, { fetchImpl = globalThis.fetch, observedAt }) {
   if (typeof fetchImpl !== 'function') throw new Error('global fetch is not available in this Node runtime');
   const response = await fetchImpl(url, {
@@ -189,6 +226,10 @@ function parseArgs(argv) {
     }
     if (!arg.startsWith('--')) throw new Error(`Unexpected argument: ${arg}`);
     const name = arg.slice(2);
+    if (name === 'local-simulation-loopback') {
+      flags.set(name, true);
+      continue;
+    }
     const value = argv[index + 1];
     if (!value || value.startsWith('--')) throw new Error(`${arg} requires a value`);
     flags.set(name, value);
@@ -198,7 +239,7 @@ function parseArgs(argv) {
 }
 
 function usage() {
-  return `Usage: node scripts/collect-hosted-backend-live-evidence.mjs --relay-url <https-base> --gateway-url <https-base> --refs-json <refs.json> --domain <domain> --environment-id <id> --cloud-provider <provider> --region <region> --owner <owner> --operator-decision go --operator-packet-ref <ref> --operator-approved-at <iso> --operator-approved-by <name> [--out <collection.json>] [--evidence-out <evidence.json>]\n\nCollects public HTTPS /livez and /readyz evidence for relay and gateway, then validates it with validate-hosted-backend-live. It never sends credentials and does not deploy infrastructure.\n`;
+  return `Usage: node scripts/collect-hosted-backend-live-evidence.mjs --relay-url <https-base> --gateway-url <https-base> --refs-json <refs.json> --domain <domain> --environment-id <id> --cloud-provider <provider> --region <region> --owner <owner> --operator-decision go --operator-packet-ref <ref> --operator-approved-at <iso> --operator-approved-by <name> [--out <collection.json>] [--evidence-out <evidence.json>] [--local-simulation-loopback]\n\nCollects public HTTPS /livez and /readyz evidence for relay and gateway, then validates it with validate-hosted-backend-live. It never sends credentials and does not deploy infrastructure. The --local-simulation-loopback flag is restricted to https://*.sim.enigmamemory.com local simulation probes with self-signed TLS and must not be used as production evidence.\n`;
 }
 
 async function runCli(argv = process.argv.slice(2), { fetchImpl = globalThis.fetch } = {}) {
@@ -212,6 +253,7 @@ async function runCli(argv = process.argv.slice(2), { fetchImpl = globalThis.fet
   const refs = await readJsonFile(refsPath);
   const environment = await maybeReadJsonFile(readFlag(flags, 'environment-json'));
   const operatorAcceptance = await maybeReadJsonFile(readFlag(flags, 'operator-acceptance-json'));
+  const selectedFetchImpl = flags.get('local-simulation-loopback') === true ? localSimulationLoopbackFetch : fetchImpl;
   const collection = await collectHostedBackendLiveEvidence({
     relayBaseUrl: readFlag(flags, 'relay-url'),
     gatewayBaseUrl: readFlag(flags, 'gateway-url'),
@@ -233,7 +275,7 @@ async function runCli(argv = process.argv.slice(2), { fetchImpl = globalThis.fet
     operatorApprovedAt: readFlag(flags, 'operator-approved-at'),
     operatorApprovedBy: readFlag(flags, 'operator-approved-by'),
     observed_at: readFlag(flags, 'observed-at') ?? new Date().toISOString(),
-    fetchImpl,
+    fetchImpl: selectedFetchImpl,
   });
   const collectionJson = `${JSON.stringify(collection, null, 2)}\n`;
   const evidenceJson = `${JSON.stringify(collection.evidence, null, 2)}\n`;

package/scripts/run-memory-benchmarks.mjs

@@ -832,6 +832,11 @@ export function runMemoryBenchmarkSuite(options = {}) {
       credentials_required: false,
       external_downloads_required: false,
       external_provider_calls: false,
+      llm_answer_accuracy_scored: false,
+      provider_api_calls_made: false,
+      api_spend_possible: false,
+      mem0_adapter_run: false,
+      external_competitor_adapters_run: false,
       raw_private_memory_plaintext_included: false,
       provider_deletion_claim: false,
       model_forgetting_claim: false,

package/scripts/run-standard-memory-benchmarks.mjs

@@ -37,12 +37,41 @@ export const STANDARD_MEMORY_BENCHMARK_METHODS = Object.freeze([
   }),
 ]);
 
+export const STANDARD_EXTERNAL_COMPETITOR_ADAPTERS = Object.freeze([
+  Object.freeze({
+    id: 'mem0',
+    name: 'Mem0',
+    status: 'not_run_requires_credentials_or_runtime',
+    target_type: 'external_adapter',
+    can_run_in_this_harness: false,
+    scores_included: false,
+    required_artifacts: Object.freeze([
+      'Mem0 platform credentials or open-source runtime',
+      'Pinned Mem0 SDK/package versions',
+      'Fixed extraction, update, retrieval, reset, model, and tool policy',
+      'Same reviewed dataset manifest, split, top-k, and scorer as Enigma rows',
+    ]),
+    official_doc: 'https://docs.mem0.ai/',
+    boundary_reason: 'The standard runner has no Mem0 credentials, SDK/runtime, fixed memory loop, reset policy, model/tool environment, or reviewed adapter scorer, so no Mem0 score is produced.',
+  }),
+]);
+
 const LOCOMO_SOURCE_URL = 'https://raw.githubusercontent.com/snap-research/locomo/main/data/locomo10.json';
 const LONGMEMEVAL_SOURCE_URLS = Object.freeze([
   'https://huggingface.co/datasets/xiaowu0162/longmemeval-cleaned/resolve/main/longmemeval_oracle.json',
   'https://huggingface.co/datasets/xiaowu0162/longmemeval-cleaned/resolve/main/longmemeval_s_cleaned.json',
   'https://huggingface.co/datasets/xiaowu0162/longmemeval-cleaned/resolve/main/longmemeval_m_cleaned.json',
 ]);
+const LOCOMO_TASK_CATEGORIES = Object.freeze(['multi-session QA', 'event summarization', 'multimodal generation over long conversations']);
+const LONGMEMEVAL_TASK_CATEGORIES = Object.freeze(['information extraction', 'multi-session reasoning', 'temporal reasoning', 'knowledge updates', 'abstention']);
+const DATASET_TASK_CATEGORIES = Object.freeze({ locomo: LOCOMO_TASK_CATEGORIES, longmemeval: LONGMEMEVAL_TASK_CATEGORIES });
+export const STANDARD_MEMORY_BENCHMARK_PROTOCOL_PLAN_SCHEMA = 'enigma.standard_memory_benchmark_protocol_plan.v1';
+const PROTOCOL_REF_RE = /^[a-z0-9][a-z0-9._:/@+-]{2,191}$/u;
+const DEFAULT_ANSWERER_MODEL_REF = 'model:answerer-not-selected';
+const DEFAULT_JUDGE_MODEL_REF = 'model:judge-not-selected';
+const DEFAULT_ANSWER_PROMPT_REF = 'prompt:standard-answer@not-pinned';
+const DEFAULT_JUDGE_PROMPT_REF = 'prompt:standard-judge@not-pinned';
+const DEFAULT_PROTOCOL_REF = 'protocol:apples-to-apples-full-answer@not-pinned';
 
 const QUERY_RELEVANCE_STOPWORDS = new Set([
   'about',
@@ -121,6 +150,25 @@ function parseArgs(argv = process.argv.slice(2)) {
     } else if (arg === '--out') {
       options.out = requiredFlagValue(argv, index, arg);
       index += 1;
+    } else if (arg === '--protocol-plan') {
+      options.protocol_plan = true;
+    } else if (arg === '--answerer-ref') {
+      options.answerer_ref = requiredFlagValue(argv, index, arg);
+      index += 1;
+    } else if (arg === '--judge-ref') {
+      options.judge_ref = requiredFlagValue(argv, index, arg);
+      index += 1;
+    } else if (arg === '--answer-prompt-ref') {
+      options.answer_prompt_ref = requiredFlagValue(argv, index, arg);
+      index += 1;
+    } else if (arg === '--judge-prompt-ref') {
+      options.judge_prompt_ref = requiredFlagValue(argv, index, arg);
+      index += 1;
+    } else if (arg === '--protocol-ref') {
+      options.protocol_ref = requiredFlagValue(argv, index, arg);
+      index += 1;
+    } else if (arg === '--dry-run') {
+      options.dry_run = true;
     } else if (arg === '--help' || arg === '-h') {
       options.help = true;
     } else {
@@ -147,6 +195,229 @@ function optionalPositiveInteger(value, name) {
   return positiveInteger(value, name);
 }
 
+function datasetPlanRows(options) {
+  const rows = [];
+  if (options.locomo !== undefined || options.locomoPath !== undefined) {
+    rows.push({
+      id: 'locomo',
+      label: 'LoCoMo',
+      local_file_name: publicFileName(options.locomo ?? options.locomoPath),
+      source_url: LOCOMO_SOURCE_URL,
+      license: 'CC BY-NC 4.0',
+      sample_limit: optionalPositiveInteger(options.max_locomo_qa ?? options.maxLocomoQa, 'max_locomo_qa') ?? null,
+      parser: 'conversation session turns as memory records; qa evidence labels score support only',
+    });
+  }
+  if (options.longmemeval !== undefined || options.longmemevalPath !== undefined || options.longMemEvalPath !== undefined) {
+    rows.push({
+      id: 'longmemeval',
+      label: 'LongMemEval',
+      local_file_name: publicFileName(options.longmemeval ?? options.longmemevalPath ?? options.longMemEvalPath),
+      source_url: LONGMEMEVAL_SOURCE_URLS,
+      license: 'Review upstream Hugging Face dataset card and LongMemEval repository terms.',
+      sample_limit: optionalPositiveInteger(options.max_longmemeval_items ?? options.maxLongMemEvalItems, 'max_longmemeval_items') ?? null,
+      parser: 'haystack_sessions turns as memory records; answer-session labels score support only',
+    });
+  }
+  return rows;
+}
+
+function offlineCommandBoundaries({ scoresIncluded, datasetFilesRead }) {
+  return {
+    deterministic_offline_runner: true,
+    dataset_files_read_from_local_disk: datasetFilesRead,
+    network_calls_made: false,
+    provider_api_calls_made: false,
+    api_spend_possible: false,
+    hosted_memory_service_called: false,
+    external_competitor_adapters_run: false,
+    mem0_adapter_run: false,
+    llm_used: false,
+    llm_answer_accuracy_scored: false,
+    retrieval_evidence_proxy_scored: scoresIncluded,
+    benchmark_scores_included: scoresIncluded,
+    raw_question_text_included: false,
+    raw_answer_text_included: false,
+    raw_conversation_text_included: false,
+    gold_labels_used_for_retrieval: false,
+    gold_labels_used_for_scoring: scoresIncluded,
+  };
+}
+
+function applesToApplesControls(topK) {
+  return {
+    same_top_k_for_all_methods: true,
+    top_k: topK,
+    same_parser_per_dataset: true,
+    same_local_records_per_dataset: true,
+    same_gold_evidence_labels_per_dataset_for_scoring_only: true,
+    local_deterministic_methods_only: true,
+    provider_runtime_fixed: false,
+    competitor_runtime_fixed: false,
+    answer_generator_fixed: false,
+    evaluator_model_fixed: false,
+  };
+}
+
+export function buildStandardBenchmarkDryRunPlan(options = {}) {
+  const topK = optionalPositiveInteger(options.top_k ?? options.topK, 'top_k') ?? 5;
+  const datasets = datasetPlanRows(options);
+  if (datasets.length === 0) throw new Error('Provide --locomo <path> and/or --longmemeval <path>');
+  return {
+    schema: 'enigma.standard_memory_benchmark_plan.v1',
+    generated_at: options.generated_at ?? new Date().toISOString(),
+    package: {
+      name: 'enigma-memory',
+      version: '0.1.17',
+    },
+    public_safe: true,
+    dry_run: true,
+    top_k: topK,
+    datasets_planned: datasets,
+    local_methods: STANDARD_MEMORY_BENCHMARK_METHODS.map((method) => ({ ...method })),
+    external_competitor_adapters: STANDARD_EXTERNAL_COMPETITOR_ADAPTERS.map((adapter) => ({
+      ...adapter,
+      required_artifacts: [...adapter.required_artifacts],
+    })),
+    command_boundaries: offlineCommandBoundaries({ scoresIncluded: false, datasetFilesRead: false }),
+    apples_to_apples_controls: applesToApplesControls(topK),
+    non_claims: [
+      'This dry run does not read dataset files and produces no benchmark score.',
+      'No provider APIs, hosted memory services, Mem0 runtime, competitor SDKs, LLM generators, or evaluator models are called.',
+      'A scored report requires a separate non-dry-run command against the exact local dataset files and hashes.',
+    ],
+  };
+}
+function protocolRef(value, label, fallback) {
+  if (value === undefined || value === null) return fallback;
+  const normalized = typeof value === 'string' ? value.trim() : String(value);
+  if (normalized === '') throw new Error(`${label} must be a non-empty public ref`);
+  if (!PROTOCOL_REF_RE.test(normalized)) throw new Error(`${label} must be a lowercase public ref using letters, numbers, . _ : / @ + or -`);
+  return normalized;
+}
+
+function protocolPlanCategorySet(datasets) {
+  const categorySet = [];
+  const seen = new Set();
+  for (const row of datasets) {
+    for (const category of DATASET_TASK_CATEGORIES[row.id] ?? []) {
+      if (!seen.has(category)) {
+        seen.add(category);
+        categorySet.push(category);
+      }
+    }
+  }
+  return categorySet;
+}
+
+export function buildStandardBenchmarkProtocolPlan(options = {}) {
+  const topK = optionalPositiveInteger(options.top_k ?? options.topK, 'top_k') ?? 5;
+  const datasets = datasetPlanRows(options);
+  if (datasets.length === 0) throw new Error('Provide --locomo <path> and/or --longmemeval <path>');
+  const answererProvided = (options.answerer_ref ?? options.answererModelRef) !== undefined;
+  const judgeProvided = (options.judge_ref ?? options.judgeModelRef) !== undefined;
+  const answererRef = protocolRef(options.answerer_ref ?? options.answererModelRef, 'answerer_ref', DEFAULT_ANSWERER_MODEL_REF);
+  const judgeRef = protocolRef(options.judge_ref ?? options.judgeModelRef, 'judge_ref', DEFAULT_JUDGE_MODEL_REF);
+  const answerPromptRef = protocolRef(options.answer_prompt_ref ?? options.answerPromptRef, 'answer_prompt_ref', DEFAULT_ANSWER_PROMPT_REF);
+  const judgePromptRef = protocolRef(options.judge_prompt_ref ?? options.judgePromptRef, 'judge_prompt_ref', DEFAULT_JUDGE_PROMPT_REF);
+  const protocolRefValue = protocolRef(options.protocol_ref ?? options.protocolRef, 'protocol_ref', DEFAULT_PROTOCOL_REF);
+  const promptsFixed = answererProvided && judgeProvided;
+  return {
+    schema: STANDARD_MEMORY_BENCHMARK_PROTOCOL_PLAN_SCHEMA,
+    generated_at: options.generated_at ?? new Date().toISOString(),
+    package: {
+      name: 'enigma-memory',
+      version: '0.1.17',
+    },
+    public_safe: true,
+    protocol_plan: true,
+    dry_run: true,
+    top_k: topK,
+    category_set: protocolPlanCategorySet(datasets),
+    datasets_planned: datasets,
+    answerer: {
+      model_ref: answererRef,
+      temperature: 0,
+      max_tokens: 1024,
+      fixed: answererProvided,
+    },
+    judge: {
+      model_ref: judgeRef,
+      kind: 'llm-as-judge-or-exact-match-not-selected',
+      temperature: 0,
+      max_tokens: 512,
+      fixed: judgeProvided,
+    },
+    prompt_refs: [answerPromptRef, judgePromptRef],
+    protocol_refs: [protocolRefValue],
+    competitor_adapter_refs: STANDARD_EXTERNAL_COMPETITOR_ADAPTERS.map((adapter) => `adapter:${adapter.id}@not-pinned`),
+    external_competitor_adapters: STANDARD_EXTERNAL_COMPETITOR_ADAPTERS.map((adapter) => ({
+      ...adapter,
+      required_artifacts: [...adapter.required_artifacts],
+      adapter_ref: `adapter:${adapter.id}@not-pinned`,
+    })),
+    apples_to_apples_controls: applesToApplesControls(topK),
+    protocol_controls: {
+      same_answerer_model_for_all_rows: answererProvided,
+      same_judge_model_for_all_rows: judgeProvided,
+      same_prompts_for_all_rows: promptsFixed,
+      same_competitor_adapters_for_enigma_and_baselines: false,
+      answerer_model_fixed: answererProvided,
+      judge_model_fixed: judgeProvided,
+      prompts_fixed: promptsFixed,
+      temperature_fixed: false,
+      budget_caps_set: false,
+    },
+    cost_estimate_inputs: {
+      dataset_sample_limits: datasets.map((row) => ({ dataset: row.id, sample_limit: row.sample_limit })),
+      answerer_temperature: 0,
+      answerer_max_tokens: 1024,
+      judge_temperature: 0,
+      judge_max_tokens: 512,
+      max_retries: 0,
+      request_timeout_ms: null,
+      budget_cap_required_before_run: true,
+      budget_cap_set: false,
+    },
+    benchmark_boundaries: {
+      official_dataset_files_required: true,
+      credentials_required: false,
+      external_provider_calls: false,
+      llm_answer_accuracy_scored: false,
+      retrieval_evidence_proxy_scored: false,
+      raw_question_text_included: false,
+      raw_answer_text_included: false,
+      raw_conversation_text_included: false,
+      provider_deletion_claim: false,
+      model_forgetting_claim: false,
+      roi_or_provider_invoice_savings_claim: false,
+      compliance_certification_claim: false,
+      benchmark_leadership_claim: false,
+    },
+    protocol_boundaries: {
+      network_required: false,
+      provider_calls_made: false,
+      answers_generated: false,
+      judged: false,
+      competitor_adapters_run: false,
+      api_spend_possible: false,
+    },
+    command_boundaries: offlineCommandBoundaries({ scoresIncluded: false, datasetFilesRead: false }),
+    review_rules: [
+      'report hash format: sha256:<hex> over the public-safe protocol-plan JSON bytes',
+      'dataset ref, runner ref, package ref, environment ref, and verifier ref formats are pinned before a live run',
+      'metric scope: this artifact plans the full-answer protocol only; it is not a score',
+      'limitation: no provider answer-accuracy, competitor-performance, benchmark-leadership, ROI, provider-deletion, model-forgetting, or compliance claim is made',
+    ],
+    non_claims: [
+      'This protocol plan is a readiness artifact: it records the planned full-answer protocol dimensions and does not execute it.',
+      'No network is used, no provider APIs are called, no answers are generated, no answers are judged, and no competitor adapters are run.',
+      'Answerer/judge model refs, prompt refs, and competitor adapter refs are public-safe references only; they are not credentials, model calls, prompts, or scores.',
+      'A scored full-answer run requires a separate credentialed command with frozen models, prompts, evaluator, dataset manifest, and budget caps.',
+    ],
+  };
+}
+
 function publicFileName(path) {
   return path === undefined || path === null ? undefined : basename(String(path));
 }
@@ -715,7 +986,7 @@ export function parseLocomoDataset(data, options = {}) {
     source_url: LOCOMO_SOURCE_URL,
     license: 'CC BY-NC 4.0',
     parser: 'conversation session turns as memory records; qa evidence labels mapped to dialog ids such as D1:3 and semicolon-separated labels',
-    task_categories: ['multi-session QA', 'event summarization', 'multimodal generation over long conversations'],
+    task_categories: [...LOCOMO_TASK_CATEGORIES],
     records,
     queries,
   };
@@ -791,7 +1062,7 @@ export function parseLongMemEvalDataset(data, options = {}) {
     source_url: LONGMEMEVAL_SOURCE_URLS,
     license: 'See Hugging Face dataset card and upstream LongMemEval repository for the selected cleaned file.',
     parser: 'haystack_sessions turns as memory records; has_answer:true turns and answer_session_ids are used as evidence labels; _abs ids are evaluated as abstention cases',
-    task_categories: ['information extraction', 'multi-session reasoning', 'temporal reasoning', 'knowledge updates', 'abstention'],
+    task_categories: [...LONGMEMEVAL_TASK_CATEGORIES],
     records,
     queries,
   };
@@ -995,7 +1266,7 @@ function buildSuiteReport(datasetRows, topK, options) {
     generated_at: options.generated_at ?? new Date().toISOString(),
     package: {
       name: 'enigma-memory',
-      version: '0.1.15',
+      version: '0.1.17',
     },
     public_safe: true,
     top_k: topK,
@@ -1010,6 +1281,8 @@ function buildSuiteReport(datasetRows, topK, options) {
       'No provider APIs, hosted runtimes, competitor SDKs, or external accounts are called by this runner.',
       'Rows are local deterministic methods only; no third-party competitor scores or benchmark-leadership claims are emitted.',
     ],
+    command_boundaries: offlineCommandBoundaries({ scoresIncluded: true, datasetFilesRead: true }),
+    apples_to_apples_controls: applesToApplesControls(topK),
     benchmark_boundaries: {
       official_dataset_files_required: true,
       credentials_required: false,
@@ -1032,15 +1305,20 @@ function buildSuiteReport(datasetRows, topK, options) {
       enigma_relevance_fallback: 'falls back to all local candidates only when no enhanced relevance signal exists, then applies deterministic local ranking and --top-k',
       provider_api_used: false,
       llm_used: false,
+      gold_labels_used_for_retrieval: false,
     },
     local_methods: STANDARD_MEMORY_BENCHMARK_METHODS.map((method) => ({ ...method })),
+    external_competitor_adapters: STANDARD_EXTERNAL_COMPETITOR_ADAPTERS.map((adapter) => ({
+      ...adapter,
+      required_artifacts: [...adapter.required_artifacts],
+    })),
     datasets: datasetRows,
     dataset_rows: datasetRows,
   };
 }
 
 function usage() {
-  return `Usage: node scripts/run-standard-memory-benchmarks.mjs [--locomo <path>] [--longmemeval <path>] [--max-locomo-qa <n>] [--max-longmemeval-items <n>] [--top-k <n>] [--out <path>]\n\nProduces schema ${STANDARD_MEMORY_BENCHMARK_SUITE_SCHEMA}. Raw question, answer, and conversation text are never written to the report. With --longmemeval and --max-longmemeval-items, the local top-level JSON array is streamed for hashing and only the requested sample items are parsed.`;
+  return `Usage: node scripts/run-standard-memory-benchmarks.mjs [--locomo <path>] [--longmemeval <path>] [--max-locomo-qa <n>] [--max-longmemeval-items <n>] [--top-k <n>] [--out <path>] [--dry-run] [--protocol-plan [--answerer-ref <ref>] [--judge-ref <ref>] [--answer-prompt-ref <ref>] [--judge-prompt-ref <ref>] [--protocol-ref <ref>]]\n\nProduces schema ${STANDARD_MEMORY_BENCHMARK_SUITE_SCHEMA}. Raw question, answer, and conversation text are never written to the report. With --longmemeval and --max-longmemeval-items, the local top-level JSON array is streamed for hashing and only the requested sample items are parsed. Use --dry-run to print a public-safe offline execution plan without reading dataset files or producing scores. Use --protocol-plan to print a public-safe full-answer benchmark PROTOCOL plan (schema ${STANDARD_MEMORY_BENCHMARK_PROTOCOL_PLAN_SCHEMA}) recording the planned category set, top-k, answerer/judge model refs, prompt/protocol refs, competitor adapter refs, and cost-estimate inputs, with explicit network_required:false, provider_calls_made:false, answers_generated:false, and judged:false boundaries; it does not call providers, generate answers, judge answers, or run competitor adapters.`;
 }
 
 async function main() {
@@ -1049,7 +1327,11 @@ async function main() {
     console.log(usage());
     return;
   }
-  const report = await runStandardMemoryBenchmarkSuiteFromFiles(options);
+  const report = options.protocol_plan
+    ? buildStandardBenchmarkProtocolPlan(options)
+    : options.dry_run
+      ? buildStandardBenchmarkDryRunPlan(options)
+      : await runStandardMemoryBenchmarkSuiteFromFiles(options);
   const serialized = `${JSON.stringify(report, null, 2)}\n`;
   if (options.out) {
     const outPath = resolve(options.out);