ruvnet-brain 4.5.4 → 4.5.6

package/scripts/model-currency-evidence.mjs

@@ -0,0 +1,139 @@
+// DISTINCT-FROM: scripts/refresh-model-catalog.mjs — independent effort-specific evaluation evidence, never router authority.
+import { createHash } from 'node:crypto';
+
+export const WEEK_MS = 7 * 24 * 60 * 60 * 1000;
+export const digest = (bytes) => createHash('sha256').update(bytes).digest('hex');
+const finite = (value) => typeof value === 'number' && Number.isFinite(value) && value >= 0;
+
+/** Captured AA Next Flight contract. No eval/script execution; a changed schema fails closed. */
+export function parseArtificialAnalysis(html, { url, checkedAt, identityBindings = {} } = {}) {
+  const version = html.match(/Artificial Analysis Intelligence Index v(\d+\.\d+(?:\.\d+)?)/)?.[1];
+  if (!version) throw new Error('AA Intelligence Index version is absent');
+  const rows = new Map();
+  const walk = (value) => {
+    if (Array.isArray(value)) { for (const item of value) walk(item); return; }
+    if (!value || typeof value !== 'object') return;
+    if (Array.isArray(value.intelligenceIndexEvaluations) && typeof value.id === 'string') {
+      rows.set(value.id, value); return;
+    }
+    for (const child of Object.values(value)) walk(child);
+  };
+  for (const match of html.matchAll(/self\.__next_f\.push\((.*?)\)<\/script>/gs)) {
+    try {
+      const envelope = JSON.parse(match[1]);
+      if (typeof envelope[1] !== 'string' || !envelope[1].includes('intelligenceIndexEvaluations')) continue;
+      const line = envelope[1];
+      walk(JSON.parse(line.slice(line.indexOf(':') + 1)));
+    } catch { /* unrelated or chunked flight frame; required rows checked below */ }
+  }
+  const source = { url, checkedAt, sha256: digest(html), parser: 'aa-next-flight-v1' };
+  const records = [];
+  for (const row of rows.values()) {
+    const effortLabel = typeof row.suffix === 'string' ? row.suffix : row.name?.match(/\((Low|Medium|High|Xhigh|Max|Minimal|None)(?:,|\))/i)?.[1];
+    const effort = effortLabel?.split(',')[0]?.trim().toLowerCase();
+    if (!['low', 'medium', 'high', 'xhigh', 'max', 'minimal', 'none'].includes(effort)) continue;
+    if (!row.release?.slug || !finite(row.intelligenceIndex) || row.intelligenceIndexIsEstimated !== false) continue;
+    const benchmarks = row.intelligenceIndexEvaluations.filter((b) => typeof b.slug === 'string'
+      && Number.isFinite(b.score) && finite(b.costPerTask) && finite(b.timePerTask))
+      .map((b) => ({ suite: b.slug, score: b.score, costUsd: b.costPerTask, timeSeconds: b.timePerTask }));
+    const binding = identityBindings[row.release.slug];
+    if (binding && (typeof binding.model !== 'string' || !binding.evidence || !binding.checkedAt)) {
+      throw new Error(`Invalid native identity binding: ${row.release.slug}`);
+    }
+    records.push({
+      sourceModelId: row.id, sourceReleaseSlug: row.release.slug, sourceName: row.name,
+      model: binding?.model ?? null, identityEvidence: binding ?? null, effort,
+      benchmark: { suite: 'artificial-analysis-intelligence-index', version },
+      quality: { intelligenceIndex: row.intelligenceIndex, estimated: false }, benchmarks,
+      costPerTaskUsd: finite(row.intelligenceIndexCostPerTask?.cost?.total) ? row.intelligenceIndexCostPerTask.cost.total : null,
+      timePerTaskSeconds: finite(row.intelligenceIndexTimePerTask) ? row.intelligenceIndexTimePerTask : null,
+      speedTokensPerSecond: finite(row.outputSpeedVariance?.median ?? row.medianOutputSpeed) ? (row.outputSpeedVariance?.median ?? row.medianOutputSpeed) : null,
+      inputUsdPerMillion: finite(row.price1mInputTokens) ? row.price1mInputTokens : null,
+      outputUsdPerMillion: finite(row.price1mOutputTokens) ? row.price1mOutputTokens : null,
+      source,
+    });
+  }
+  if (records.length === 0 || !records.some((r) => r.benchmarks.length > 0 && r.costPerTaskUsd !== null && r.timePerTaskSeconds !== null)) {
+    throw new Error('AA effort benchmark matrix absent or incomplete');
+  }
+  return { source, records };
+}
+
+export function parseInventory(body, source) {
+  const data = JSON.parse(body).data;
+  if (!Array.isArray(data) || data.length < 50) throw new Error('OpenRouter inventory is suspiciously thin');
+  const models = data.map((row) => {
+    const input = Number(row.pricing?.prompt); const output = Number(row.pricing?.completion);
+    if (typeof row.id !== 'string' || !row.id || !Number.isFinite(input) || !Number.isFinite(output)) throw new Error('Invalid OpenRouter model/pricing');
+    return { id: row.id, pricing: { inputUsdPerMillion: finite(input) ? input * 1e6 : null, outputUsdPerMillion: finite(output) ? output * 1e6 : null },
+      supportedParameters: Array.isArray(row.supported_parameters) ? row.supported_parameters.filter((p) => typeof p === 'string') : [] };
+  });
+  return { checkedAt: source.checkedAt, source, models, discoveryOnly: true };
+}
+
+/** No benchmark row can certify native access, supported effort, or default-selection eligibility. */
+export function currencyStatus(record, now = Date.now()) {
+  const fresh = (section) => {
+    const checked = Date.parse(section?.checkedAt);
+    return Number.isFinite(checked) && checked <= now && now - checked < WEEK_MS;
+  };
+  const inventoryFresh = record?.schemaVersion === 1 && fresh(record.inventory)
+    && record.inventory.discoveryOnly === true && record.inventory.models?.length >= 50;
+  const evaluationsFresh = record?.schemaVersion === 1 && fresh(record.evaluations)
+    && record.evaluations.sources?.length > 0 && record.evaluations.records?.length > 0
+    && record.evaluations.sources.every((source) => fresh(source) && /^[a-f0-9]{64}$/.test(source.sha256 ?? ''));
+  const failed = !!record?.lastAttempt && record.lastAttempt.status !== 'complete';
+  return { status: inventoryFresh && evaluationsFresh && !failed ? 'current' : 'stale', inventoryFresh: !!inventoryFresh,
+    evaluationsFresh: !!evaluationsFresh, selectionQualified: false, maxAgeMs: WEEK_MS,
+    errors: record?.lastAttempt?.errors ?? [], assessment: record?.assessment ?? null, reason: failed ? 'last refresh incomplete' : inventoryFresh && evaluationsFresh ? 'fresh independent evidence' : 'weekly evidence required' };
+}
+
+/** Coding-agent workloads are a separate suite and harness/configuration, never model-index aliases. */
+export function parseCodingAgentEvidence(html, methodology, { url, checkedAt, identityBindings = {} } = {}) {
+  const version = methodology.match(/Coding Agent Index v(\d+\.\d+(?:\.\d+)?)/)?.[1];
+  if (!version) throw new Error('Coding Agent Index methodology version absent');
+  const rows = new Map();
+  const walk = (value) => {
+    if (Array.isArray(value)) { for (const child of value) walk(child); return; }
+    if (!value || typeof value !== 'object') return;
+    if (typeof value.hostModelSlug === 'string' && typeof value.id === 'string' && value.indexScore !== undefined) { rows.set(value.id, value); return; }
+    for (const child of Object.values(value)) walk(child);
+  };
+  for (const match of html.matchAll(/self\.__next_f\.push\((.*?)\)<\/script>/gs)) {
+    try {
+      const envelope = JSON.parse(match[1]);
+      if (typeof envelope[1] !== 'string') continue;
+      for (const line of envelope[1].split('\n')) {
+        if (!line.includes('hostModelSlug')) continue;
+        try { walk(JSON.parse(line.slice(line.indexOf(':') + 1))); } catch { /* unrelated frame */ }
+      }
+    } catch { /* required contract checked below */ }
+  }
+  const source = { url, checkedAt, sha256: digest(html), parser: 'aa-coding-agents-flight-v1' };
+  const records = [];
+  for (const row of rows.values()) {
+    if (!finite(row.indexScore) || row.indexScore > 1 || !finite(row.mean?.costUsd) || !finite(row.mean?.agentWallTimeSec)
+      || !row.display?.model || !Array.isArray(row.evals) || row.evals.length !== row.indexComponentCount || !row.versions
+      || row.evals.some((e) => !e.datasetIndexName || !finite(e.mean?.reward))) {
+      throw new Error(`Incomplete coding-agent record: ${row.id}`);
+    }
+    const nativeHost = row.agentName === 'Codex' && row.provider === 'openai' ? 'codex'
+      : row.agentName === 'Claude Code' && row.provider === 'anthropic' ? 'claude-code' : null;
+    const binding = nativeHost ? identityBindings[row.modelRelease?.slug] : null;
+    if (binding && (!binding.model || !binding.evidence || !binding.checkedAt)) throw new Error('Invalid coding-agent native identity binding');
+    const effort = nativeHost ? row.display.model.match(/\((low|medium|high|xhigh|max|minimal|none)\)/i)?.[1]?.toLowerCase() ?? null : null;
+    records.push({ sourceId: row.id, harness: row.agentName, provider: row.provider, nativeHost,
+      sourceModelSlug: row.hostModelSlug, sourceReleaseSlug: row.modelRelease?.slug ?? null,
+      displayModel: row.display.model, configurationLabel: row.displayLabel, effort,
+      model: binding?.model ?? null, identityEvidence: binding ?? null,
+      fallback: /fallback/i.test(row.displayLabel ?? row.display.model),
+      benchmark: { suite: 'artificial-analysis-coding-agent-index', version }, codingAgentIndexFraction: row.indexScore,
+      apiBenchmarkCostPerTaskUsd: row.mean.costUsd, timePerTaskSeconds: row.mean.agentWallTimeSec,
+      versions: row.versions, components: row.evals.map((e) => ({ suite: e.datasetIndexName,
+        dataset: e.refDatasetName, score: e.mean.reward, mean: e.mean })), source,
+      selectionQualified: false });
+  }
+  if (!records.length) throw new Error('Coding-agent structured records absent');
+  return { source, records, coverage: { scope: 'public serialized agent/configuration records; native entitlements unverified',
+    recordCount: records.length, renderedChartSelectionVerified: false }, selectionQualified: false };
+}

package/scripts/model-currency.mjs

@@ -0,0 +1,230 @@
+#!/usr/bin/env node
+// DISTINCT-FROM: scripts/model-router-catalog.mjs — per-user evidence currency, never candidate/default mutation.
+import fs from 'node:fs';
+import path from 'node:path';
+import os from 'node:os';
+import { spawn } from 'node:child_process';
+import { randomUUID } from 'node:crypto';
+import { buildWeeklyAssessment } from './model-weekly-assessment.mjs';
+import { fileURLToPath } from 'node:url';
+import { WEEK_MS, digest, parseInventory, parseArtificialAnalysis, parseCodingAgentEvidence, currencyStatus } from './model-currency-evidence.mjs';
+
+export const DEFAULT_ROUTER_DIR = path.join(os.homedir(), '.claude', 'model-router');
+export const AA_URLS = [
+  'https://artificialanalysis.ai/models/comparisons',
+  'https://artificialanalysis.ai/models/comparisons/claude-fable-5-1-medium-vs-claude-opus-5-medium',
+  'https://artificialanalysis.ai/models/releases/comparisons/gpt-6-1-sol-vs-claude-sonnet-5-5',
+  'https://artificialanalysis.ai/models/releases/comparisons/gpt-6-luna-vs-gpt-6-astra',
+  'https://artificialanalysis.ai/models/releases/comparisons/claude-opus-5-5-vs-gpt-6-astra',
+];
+export const OFFICIAL_SOURCES = [
+  { provider: 'openai', url: 'https://developers.openai.com/api/docs/models' },
+  { provider: 'anthropic', url: 'https://platform.claude.com/docs/en/models/overview' },
+];
+export const AGENT_SOURCE_URLS = [
+  'https://artificialanalysis.ai/agents/coding-agents',
+  'https://artificialanalysis.ai/methodology/coding-agents-benchmarking',
+];
+export const CODING_BENCHMARK_SOURCES = [
+  { suite: 'vulcanbench', url: 'https://vulcanbench.com/' },
+  { suite: 'vulcanbench-swe-v4', url: 'https://vulcanbench.com/benchmarks/swe-v4-gpt61-sol-v318.html' },
+  { suite: 'terminal-bench', url: 'https://www.tbench.ai/' },
+  { suite: 'swe-bench', url: 'https://www.swebench.com/' },
+];
+const INVENTORY_URL = 'https://openrouter.ai/api/v1/models';
+const LOCK_MS = 10 * 60 * 1000;
+const RETRY_MS = 60 * 60 * 1000;
+const SELF = fileURLToPath(import.meta.url);
+function readRecord(routerDir) {
+  try {
+    const target = path.join(routerDir, 'currency.json');
+    if (fs.statSync(target).size > 8 * 1024 * 1024) return null;
+    return JSON.parse(fs.readFileSync(target, 'utf8'));
+  } catch { return null; }
+}
+export function readCurrencyStatus({ routerDir = DEFAULT_ROUTER_DIR, now = Date.now() } = {}) {
+  return currencyStatus(readRecord(routerDir), now);
+}
+function atomicWrite(target, bytes, beforeCommit = () => {}) {
+  fs.mkdirSync(path.dirname(target), { recursive: true, mode: 0o700 });
+  const tmp = `${target}.tmp-${process.pid}-${Date.now()}`;
+  try {
+    const fd = fs.openSync(tmp, 'wx', 0o600);
+    try { fs.writeFileSync(fd, bytes); fs.fsyncSync(fd); } finally { fs.closeSync(fd); }
+    beforeCommit();
+    fs.renameSync(tmp, target);
+  } finally { try { fs.unlinkSync(tmp); } catch { /* renamed */ } }
+}
+// This guard is held only for synchronous filesystem transactions, never during fetches.
+// Never reap it: a crash inside this tiny transaction must fail closed rather than overlap writers.
+function transaction(routerDir, operation) {
+  fs.mkdirSync(routerDir, { recursive: true, mode: 0o700 });
+  const guard = path.join(routerDir, 'currency-mutation.lock');
+  try { fs.mkdirSync(guard, { mode: 0o700 }); } catch (error) { if (error.code === 'EEXIST') return null; throw error; }
+  try { return operation(); } finally { fs.rmdirSync(guard); }
+}
+function ownerPath(routerDir) { return path.join(routerDir, 'currency-refresh-owner.json'); }
+function readOwner(routerDir) {
+  try { return JSON.parse(fs.readFileSync(ownerPath(routerDir), 'utf8')); }
+  catch (error) { if (error.code === 'ENOENT') return null; throw error; }
+}
+function claim(routerDir, now) {
+  return transaction(routerDir, () => {
+    const owner = readOwner(routerDir);
+    if (owner && (!Number.isFinite(owner.claimedAt) || now - owner.claimedAt <= LOCK_MS)) return null;
+    const next = { token: randomUUID(), claimedAt: now };
+    atomicWrite(ownerPath(routerDir), JSON.stringify(next));
+    return next.token;
+  });
+}
+function release(routerDir, token) {
+  return transaction(routerDir, () => {
+    if (readOwner(routerDir)?.token !== token) return false;
+    fs.unlinkSync(ownerPath(routerDir)); return true;
+  });
+}
+function fencedWrite(routerDir, token, target, bytes) {
+  const written = transaction(routerDir, () => {
+    if (readOwner(routerDir)?.token !== token) throw new Error('refresh superseded: ownership token changed');
+    atomicWrite(target, bytes, () => {
+      if (readOwner(routerDir)?.token !== token) throw new Error('refresh superseded before commit');
+    }); return true;
+  });
+  if (!written) throw new Error('refresh transaction busy; no write committed');
+}
+
+/** Prompt path: local bounded read and one detached worker; never await a network request. */
+export function maybeLaunchCurrencyRefresh({ routerDir = DEFAULT_ROUTER_DIR, now = Date.now(), launch = spawn } = {}) {
+  const record = readRecord(routerDir); const status = currencyStatus(record, now);
+  if (status.status === 'current') return { ...status, launched: false };
+  const attempted = Date.parse(record?.lastAttempt?.checkedAt);
+  if (Number.isFinite(attempted) && attempted <= now && now - attempted < RETRY_MS) return { ...status, launched: false, deferred: 'retry cooldown' };
+  const lock = claim(routerDir, now);
+  if (!lock) return { ...status, launched: false, deferred: fs.existsSync(path.join(routerDir, 'currency-mutation.lock'))
+    ? 'refresh transaction blocked; inspect mutation guard before recovery' : 'refresh already running' };
+  try {
+    const child = launch(process.execPath, [SELF, '--refresh', '--claim-token', lock, '--router-dir', routerDir], { detached: true, stdio: 'ignore' });
+    child.once?.('error', () => release(routerDir, lock));
+    child.unref();
+    return { ...status, launched: true };
+  } catch (error) { release(routerDir, lock); return { ...status, launched: false, errors: [error.message] }; }
+}
+
+export async function refreshModelCurrency({ routerDir = DEFAULT_ROUTER_DIR, now = Date.now(), fetchImpl = fetch,
+  identityBindings, aaUrls = AA_URLS, officialSources = OFFICIAL_SOURCES, agentSourceUrls = AGENT_SOURCE_URLS, codingBenchmarkSources = CODING_BENCHMARK_SOURCES, claimToken = null } = {}) {
+  const lock = claimToken ?? claim(routerDir, now);
+  if (!lock) return { action: 'busy', status: 'stale', reason: 'refresh ownership or transaction guard unavailable' };
+  try {
+    const prior = readRecord(routerDir) ?? { schemaVersion: 1 };
+    const checkedAt = new Date(now).toISOString(); const errors = [];
+    let bindings = identityBindings;
+    if (!bindings) {
+      try { bindings = JSON.parse(fs.readFileSync(path.join(routerDir, 'identity-bindings.json'), 'utf8')); }
+      catch { bindings = {}; }
+    }
+    const collect = async (url) => {
+      const response = await fetchImpl(url, { signal: AbortSignal.timeout(20_000) });
+      if (!response.ok) throw new Error(`HTTP ${response.status}`);
+      const bytes = await response.text();
+      if (bytes.length > 6 * 1024 * 1024) throw new Error('source exceeds 6 MiB limit');
+      const source = { url, checkedAt, sha256: digest(bytes) };
+      fencedWrite(routerDir, lock, path.join(routerDir, 'evidence', `${source.sha256}.${url === INVENTORY_URL ? 'json' : 'html'}`), bytes);
+      return { source, bytes };
+    };
+    const results = await Promise.allSettled([INVENTORY_URL, ...aaUrls, ...officialSources.map((s) => s.url), ...agentSourceUrls, ...codingBenchmarkSources.map((s) => s.url)].map(collect));
+    let inventory = prior.inventory; let evaluations = prior.evaluations;
+    if (results[0].status === 'fulfilled') {
+      try { inventory = parseInventory(results[0].value.bytes, results[0].value.source); }
+      catch (error) { errors.push(`inventory: ${error.message}`); }
+    } else errors.push(`inventory: ${results[0].reason.message}`);
+    const parsed = [];
+    for (let i = 1; i <= aaUrls.length; i++) {
+      const result = results[i];
+      try {
+        if (result.status !== 'fulfilled') throw result.reason;
+        parsed.push(parseArtificialAnalysis(result.value.bytes, { ...result.value.source, identityBindings: bindings }));
+      } catch (error) { errors.push(`evaluations ${aaUrls[i - 1]}: ${error.message}`); }
+    }
+    // All requested pages must parse before certifying a new matrix; partial pages are evidence only.
+    if (parsed.length === aaUrls.length && parsed.length > 0) {
+      const records = new Map();
+      for (const page of parsed) for (const record of page.records) records.set(record.sourceModelId, record);
+      evaluations = { checkedAt, sources: parsed.map((p) => p.source), records: [...records.values()],
+        selectionQualified: false, limitation: 'Independent benchmark evidence; native access and supported effort require separate verification. Arena is not collected.' };
+    }
+    let official = prior.officialSources;
+    const publicDocs = [];
+    for (let i = 0; i < officialSources.length; i++) {
+      const result = results[1 + aaUrls.length + i];
+      if (result.status === 'fulfilled') publicDocs.push({ ...result.value.source, provider: officialSources[i].provider,
+        scope: 'official public/API documentation; not native subscription access', semanticallyQualified: false });
+      else errors.push(`official ${officialSources[i].url}: ${result.reason.message}`);
+    }
+    if (publicDocs.length === officialSources.length) official = { checkedAt, sources: publicDocs };
+    let agents = prior.agentSources;
+    if (agentSourceUrls.length) {
+      const agentResults = results.slice(1 + aaUrls.length + officialSources.length, 1 + aaUrls.length + officialSources.length + agentSourceUrls.length);
+      try {
+        if (agentResults.length !== 2) throw new Error('coding-agent source and methodology pair required');
+        for (const result of agentResults) if (result.status !== 'fulfilled') throw result.reason;
+        const parsedAgents = parseCodingAgentEvidence(agentResults[0].value.bytes, agentResults[1].value.bytes,
+          { ...agentResults[0].value.source, identityBindings: bindings });
+        agents = { checkedAt, sources: agentResults.map((r) => r.value.source), ...parsedAgents };
+      } catch (error) { errors.push(`coding agents: ${error.message}`); }
+    }
+    const additional = [];
+    for (let i = 0; i < codingBenchmarkSources.length; i++) {
+      const result = results[1 + aaUrls.length + officialSources.length + agentSourceUrls.length + i];
+      if (result.status === 'fulfilled' && result.value.bytes.trim().length >= 100) {
+        additional.push({ ...result.value.source, suite: codingBenchmarkSources[i].suite,
+          scope: 'raw source archive; independent suite, score parsing and semantic qualification not implemented' });
+      } else errors.push(`coding benchmark ${codingBenchmarkSources[i].url}: ${result.status === 'rejected' ? result.reason.message : 'empty or truncated source'}`);
+    }
+    if (additional.length === codingBenchmarkSources.length && agents && agents !== prior.agentSources) agents = { ...agents, additionalSources: additional };
+    else if (additional.length !== codingBenchmarkSources.length) agents = prior.agentSources;
+    const instructionPath = path.join(routerDir, 'weekly-analyst-instruction.md');
+    let instruction; let instructionSource = 'packaged-fallback';
+    try {
+      const fd = fs.openSync(instructionPath, 'r');
+      try {
+        const buffer = Buffer.alloc(128 * 1024 + 1);
+        const size = fs.readSync(fd, buffer, 0, buffer.length, 0);
+        if (size > 128 * 1024) throw new Error('effective weekly instruction exceeds 128 KiB');
+        instruction = buffer.subarray(0, size).toString('utf8');
+        if (!instruction.trim()) throw new Error('effective weekly instruction is empty');
+        instructionSource = 'effective-per-user-file';
+      } finally { fs.closeSync(fd); }
+    } catch (error) {
+      if (error.code !== 'ENOENT') { errors.push(`instruction: ${error.message}`); instruction = undefined; instructionSource = 'fallback-after-read-error'; }
+    }
+    const next = { schemaVersion: 1, maxAgeMs: WEEK_MS, inventory, evaluations, officialSources: official, agentSources: agents,
+      lastAttempt: { checkedAt, status: errors.length ? (inventory === prior.inventory && evaluations === prior.evaluations ? 'failed' : 'partial') : 'complete', errors } };
+    let priorPolicyBytes = null; let policy = null;
+    try { priorPolicyBytes = fs.readFileSync(path.join(routerDir, 'routing-policy.json'), 'utf8'); policy = JSON.parse(priorPolicyBytes); }
+    catch { /* no policy: report missing allocation, never synthesize one */ }
+    const assessment = buildWeeklyAssessment({ currency: next, policy, priorPolicyBytes, now, previousAssessment: prior.assessment, instruction, instructionSource });
+    const assessmentDir = path.join(routerDir, 'assessments', `${checkedAt.replaceAll(':', '-')}-${lock}`);
+    for (const [name, bytes] of [['report.json', JSON.stringify(assessment.report, null, 2)],
+      ['proposal.json', JSON.stringify(assessment.proposal, null, 2)], ['report.md', assessment.markdown],
+      ['instruction.md', assessment.instruction], ['prior-policy.json', priorPolicyBytes ?? 'null']]) {
+      fencedWrite(routerDir, lock, path.join(assessmentDir, name), bytes);
+    }
+    if (!fs.existsSync(instructionPath)) fencedWrite(routerDir, lock, instructionPath, assessment.instruction);
+    next.assessment = { instructionPath, instructionSha256: assessment.report.instructionSha256, instructionSource, checkedAt, signature: assessment.report.signature, reportPath: path.join(assessmentDir, 'report.json'),
+      proposalPath: path.join(assessmentDir, 'proposal.json'), status: 'complete-unqualified', analystExecuted: false,
+      notification: assessment.report.notification };
+    fencedWrite(routerDir, lock, path.join(routerDir, 'currency.json'), `${JSON.stringify(next, null, 2)}\n`);
+    return { action: 'refreshed', ...currencyStatus(next, now), lastAttempt: next.lastAttempt };
+  } finally { release(routerDir, lock); }
+}
+
+if (process.argv[1] && path.resolve(process.argv[1]) === SELF) {
+  const dirIndex = process.argv.indexOf('--router-dir');
+  const routerDir = dirIndex >= 0 ? process.argv[dirIndex + 1] : DEFAULT_ROUTER_DIR;
+  if (!routerDir || !path.isAbsolute(routerDir)) throw new Error('--router-dir must be absolute');
+  if (process.argv.includes('--refresh')) {
+    refreshModelCurrency({ routerDir, claimToken: process.argv.includes('--claim-token') ? process.argv[process.argv.indexOf('--claim-token') + 1] : null }).then((result) => {
+      console.log(JSON.stringify(result)); if (result.status === 'stale') process.exitCode = 1;
+    }).catch((error) => { console.error(error.message); process.exitCode = 1; });
+  } else console.log(JSON.stringify(process.argv.includes('--catch-up') ? maybeLaunchCurrencyRefresh({ routerDir }) : readCurrencyStatus({ routerDir })));
+}

package/scripts/model-native-catalog.mjs

@@ -0,0 +1,111 @@
+#!/usr/bin/env node
+// DISTINCT-FROM: model-router-catalog.mjs — fresh native subscription metadata, never model execution or policy promotion.
+import fs from 'node:fs';
+import path from 'node:path';
+import os from 'node:os';
+import { randomUUID } from 'node:crypto';
+import { fileURLToPath } from 'node:url';
+const HOSTS = { codex: 'openai', 'claude-code': 'anthropic' };
+const EFFORTS = new Set(['low', 'medium', 'high', 'xhigh', 'max']);
+const read = (file, optional = false) => {
+  try { const bytes = fs.readFileSync(file); if (bytes.length > 1048576) throw new Error('Native catalog input exceeds limit'); return bytes; }
+  catch (error) { if (optional && error.code === 'ENOENT') return null; throw error; }
+};
+function stage(file, bytes) {
+  const temporary = `${file}.${randomUUID()}.tmp`; const fd = fs.openSync(temporary, 'wx', 0o600);
+  try { fs.writeFileSync(fd, bytes); fs.fsyncSync(fd); } catch (error) { fs.closeSync(fd); fs.unlinkSync(temporary); throw error; }
+  fs.closeSync(fd);
+  return temporary;
+}
+const equal = (a, b) => a === null ? b === null : b !== null && a.equals(b);
+export async function refreshNativeCatalog({ routerDir = path.join(os.homedir(), '.claude', 'model-router'),
+  deadline = Date.now() + 60000, inspect, now = Date.now() } = {}) {
+  const checkedAt = new Date(now).toISOString();
+  const failed = reason => ({ status: 'failed', checkedAt, newNativeModelIds: [], policyApplied: false, reason });
+  const paths = { profile: path.join(routerDir, 'profile.json'), catalog: path.join(routerDir, 'catalog.json'),
+    availability: path.join(routerDir, 'native-availability.json') };
+  let guard; let acquired = false; const temporary = [];
+  try {
+    if (!path.isAbsolute(routerDir) || !Number.isFinite(deadline) || deadline <= Date.now()) return failed('Invalid native metadata directory or deadline');
+    const before = Object.fromEntries(Object.entries(paths).map(([key, file]) => [key, read(file, key === 'availability')]));
+    const profile = JSON.parse(before.profile); const catalog = JSON.parse(before.catalog);
+    if (profile.automaticModelRoutingUpdates !== true) return failed('Automatic native model registration is not authorized by this profile');
+    if (!Array.isArray(catalog.candidates) || catalog.candidates.some(c => typeof c.id !== 'string')
+      || new Set(catalog.candidates.map(c => c.id)).size !== catalog.candidates.length) return failed('Existing model catalog is malformed');
+    const hosts = Object.keys(HOSTS).filter(host => profile.harnesses?.[host]?.available === true
+      && profile.harnesses[host].subscription === true && profile.harnesses[host].enabled !== false);
+    if (!hosts.length) return failed('No enabled native subscription host in this profile');
+    inspect ??= (await import('./model-native-qualification.mjs')).inspectNativeQualificationHost;
+    const observations = []; const eligibleIds = []; const nativeIds = []; const unknownEffortIds = []; const withheldIds = [];
+    const candidates = [...catalog.candidates];
+    for (const host of hosts) {
+      if (Date.now() >= deadline) return failed('Native metadata deadline expired');
+      const result = await inspect({ host, deadline });
+      if (result?.failure || result?.available !== true || result.nativeSubscription !== true || result.sourceReceipt?.host !== host
+        || result.sourceReceipt.allowance?.verified !== true || result.catalogStatus !== 'verified'
+        || !Array.isArray(result.models) || !result.models.length) return failed(`Verified native model list unavailable for ${host}`);
+      const models = [];
+      for (const model of result.models) {
+        if (!/^[a-zA-Z0-9][a-zA-Z0-9._-]+$/.test(model.id ?? '')) return failed(`Exact native model identity unavailable for ${host}`);
+        const id = `${HOSTS[host]}/${model.id}`; nativeIds.push(id);
+        const nativeEfforts = Array.isArray(model.efforts) ? model.efforts.filter(e => typeof e === 'string' && /^[a-z][a-z0-9_-]{0,31}$/.test(e)) : [];
+        const efforts = [...new Set(nativeEfforts.filter(e => EFFORTS.has(e)))].sort();
+        const unrecognizedEfforts = [...new Set(nativeEfforts.filter(e => !EFFORTS.has(e)))].sort();
+        const knownEfforts = efforts.length > 0;
+        models.push({ id: model.id, efforts, effortSupport: knownEfforts ? 'verified' : 'unknown',
+          ...(unrecognizedEfforts.length ? { unrecognizedEfforts } : {}) });
+        if (!knownEfforts) { unknownEffortIds.push(id); continue; }
+        const existing = candidates.find(c => c.id === model.id);
+        if (existing) {
+          if (existing.provider !== HOSTS[host] || existing.disabled === true || existing.enabled === false
+            || !existing.harness?.includes(host) || !existing.subscription?.includes(host)) { withheldIds.push(id); continue; }
+        } else candidates.push({ id: model.id, provider: HOSTS[host], harness: [host], subscription: [host],
+          supportedEfforts: efforts, costPerMTok: null, verified: `${checkedAt} native subscription metadata; execution not tested`,
+          note: 'Native catalog availability is not role-quality qualification or active routing approval.' });
+        eligibleIds.push(id);
+      }
+      if (new Set(models.map(m => m.id)).size !== models.length) return failed(`Duplicate native model identity for ${host}`);
+      observations.push({ host, provider: HOSTS[host], checkedAt: result.checkedAt, harnessVersion: result.harnessVersion, allowanceVerified: true, models: models.sort((a, b) => a.id.localeCompare(b.id)) });
+    }
+    if (Date.now() >= deadline) return failed('Native metadata deadline expired');
+    const previous = before.availability ? JSON.parse(before.availability) : null;
+    if (previous && (previous.schemaVersion !== 1 || !Array.isArray(previous.nativeModelIds)
+      || previous.nativeModelIds.some(id => typeof id !== 'string'))) return failed('Previous native availability baseline is malformed');
+    const ids = [...new Set(eligibleIds)].sort(); const nativeModelIds = [...new Set(nativeIds)].sort();
+    const newNativeModelIds = previous ? nativeModelIds.filter(id => !previous.nativeModelIds.includes(id)) : [];
+    const snapshot = { schemaVersion: 1, status: 'current', checkedAt, hosts: observations,
+      nativeModelIds, eligibleModelIds: ids, unknownEffortModelIds: unknownEffortIds.sort(), withheldExistingModelIds: [...new Set(withheldIds)].sort(), inferenceRequests: 0, policyApplied: false };
+    const changedCatalog = candidates.length !== catalog.candidates.length;
+    const catalogBytes = changedCatalog ? Buffer.from(JSON.stringify({ ...catalog, candidates }, null, 2) + '\n') : before.catalog;
+    const snapshotBytes = Buffer.from(JSON.stringify(snapshot, null, 2) + '\n');
+    guard = path.join(routerDir, 'native-catalog-mutation.lock'); fs.mkdirSync(guard, { mode: 0o700 }); acquired = true;
+    for (const [key, file] of Object.entries(paths)) if (!equal(before[key], read(file, key === 'availability'))) throw new Error('Native catalog inputs changed');
+    const history = path.join(routerDir, 'native-catalog-history', `${now}-${randomUUID()}`); fs.mkdirSync(history, { recursive: true, mode: 0o700 });
+    for (const [name, bytes] of [['catalog.json', before.catalog], ['native-availability.json', before.availability]]) {
+      if (!bytes) continue; const target = path.join(history, name); const temp = stage(target, bytes); temporary.push(temp); fs.renameSync(temp, target);
+    }
+    const catalogTemp = changedCatalog ? stage(paths.catalog, catalogBytes) : null;
+    if (catalogTemp) temporary.push(catalogTemp);
+    const snapshotTemp = stage(paths.availability, snapshotBytes); temporary.push(snapshotTemp);
+    if (catalogTemp) fs.renameSync(catalogTemp, paths.catalog);
+    try { fs.renameSync(snapshotTemp, paths.availability); }
+    catch (error) {
+      if (catalogTemp && equal(catalogBytes, read(paths.catalog))) {
+        const restore = stage(paths.catalog, before.catalog); temporary.push(restore); fs.renameSync(restore, paths.catalog);
+      }
+      throw error;
+    }
+    return { status: 'current', checkedAt, newNativeModelIds, baseline: previous === null,
+      catalogAdded: candidates.length - catalog.candidates.length, unknownEffortModelIds: snapshot.unknownEffortModelIds, withheldExistingModelIds: snapshot.withheldExistingModelIds,
+      inferenceRequests: 0, policyApplied: false };
+  } catch { return failed('Native catalog refresh refused; previous availability and owner catalog retained where unchanged'); }
+  finally {
+    for (const file of temporary) { try { fs.unlinkSync(file); } catch { /* renamed */ } }
+    if (acquired) { try { fs.rmdirSync(guard); } catch { /* not acquired */ } }
+  }
+}
+if (process.argv[1] && path.resolve(process.argv[1]) === fileURLToPath(import.meta.url)) {
+  const option = name => { const at = process.argv.indexOf(name); return at < 0 ? undefined : process.argv[at + 1]; };
+  refreshNativeCatalog({ routerDir: option('--router-dir'), deadline: option('--deadline') === undefined ? undefined : Number(option('--deadline')) })
+    .then(result => { console.log(JSON.stringify(result)); if (result.status !== 'current') process.exitCode = 1; });
+}