nexus-agents 8.89.3 → 8.90.0

package/dist/{ast-rule-runner-AMZCLKRV.js → ast-rule-runner-SD665GB7.js}

@@ -9,8 +9,8 @@ import {
   getBuiltInAstRulesPath,
   loadRules,
   runAstQaRules
-} from "./chunk-LFFXUHX3.js";
-import "./chunk-NZN3SWSZ.js";
+} from "./chunk-O3MSZUQM.js";
+import "./chunk-F2H3MCEV.js";
 import "./chunk-UE7A2XHM.js";
 import "./chunk-LWOX65CP.js";
 import "./chunk-PX6F3LHL.js";
@@ -26,4 +26,4 @@ export {
   loadRules,
   runAstQaRules
 };
-//# sourceMappingURL=ast-rule-runner-AMZCLKRV.js.map
+//# sourceMappingURL=ast-rule-runner-SD665GB7.js.map

package/dist/{child-mcp-config-7DSYI325.js → child-mcp-config-BG3QNALD.js}

@@ -3,7 +3,7 @@ import {
 } from "./chunk-OITVVSWO.js";
 import {
   createLogger
-} from "./chunk-NZN3SWSZ.js";
+} from "./chunk-F2H3MCEV.js";
 import "./chunk-UE7A2XHM.js";
 import "./chunk-LWOX65CP.js";
 import "./chunk-PX6F3LHL.js";
@@ -56,4 +56,4 @@ async function generateMcpConfig(options) {
 export {
   generateMcpConfig
 };
-//# sourceMappingURL=child-mcp-config-7DSYI325.js.map
+//# sourceMappingURL=child-mcp-config-BG3QNALD.js.map

package/dist/{chunk-5UCGVAEZ.js → chunk-2N7XWZPW.js}

@@ -1,14 +1,14 @@
 import {
   GitHubProvider,
   ScmError
-} from "./chunk-SCXCJSTT.js";
+} from "./chunk-JWHIFEI3.js";
 import {
   resolveToken
-} from "./chunk-UDWZJY3V.js";
+} from "./chunk-HUCILYVO.js";
 import {
   err,
   ok
-} from "./chunk-NZN3SWSZ.js";
+} from "./chunk-F2H3MCEV.js";
 
 // src/scm/factory.ts
 async function createScmProvider(config) {
@@ -41,4 +41,4 @@ export {
   createScmProvider,
   createGitHubProvider
 };
-//# sourceMappingURL=chunk-5UCGVAEZ.js.map
+//# sourceMappingURL=chunk-2N7XWZPW.js.map

package/dist/{chunk-BJXSG5YL.js → chunk-4KMNLWAC.js}

@@ -4,12 +4,12 @@ import {
   planOptionalParams,
   requireApiKey,
   validateApiKeyPresence
-} from "./chunk-ZTTBGME2.js";
+} from "./chunk-VI5TF3KF.js";
 import {
   assertCustomApiHostResolvesPublic,
   hostnameOf,
   redactApiKey
-} from "./chunk-LZNYQYDU.js";
+} from "./chunk-KCWEMSU2.js";
 import {
   sanitizeErrorDetails
 } from "./chunk-2NDA4DHW.js";
@@ -31,7 +31,7 @@ import {
   isEndpointArmId,
   ok,
   recordUsageEvent
-} from "./chunk-NZN3SWSZ.js";
+} from "./chunk-F2H3MCEV.js";
 
 // src/adapters/openai-types.ts
 var OPENAI_MODELS = {
@@ -886,4 +886,4 @@ export {
   discoverModels,
   buildOpenAICompatAdapters
 };
-//# sourceMappingURL=chunk-BJXSG5YL.js.map
+//# sourceMappingURL=chunk-4KMNLWAC.js.map

package/dist/{chunk-QX42N7TE.js → chunk-4PMYNW3G.js}

@@ -1,6 +1,6 @@
 // src/version.ts
 import semver from "semver";
-var VERSION = true ? "8.89.3" : "dev";
+var VERSION = true ? "8.90.0" : "dev";
 var NODE_ENGINE_RANGE = ">=22.5.0";
 function isNodeVersionSupported(version) {
   return semver.satisfies(version, NODE_ENGINE_RANGE);
@@ -11,4 +11,4 @@ export {
   NODE_ENGINE_RANGE,
   isNodeVersionSupported
 };
-//# sourceMappingURL=chunk-QX42N7TE.js.map
+//# sourceMappingURL=chunk-4PMYNW3G.js.map

package/dist/{chunk-3PQIIMHN.js → chunk-5GI6Z5VE.js}

@@ -1,12 +1,19 @@
 import {
+  allOf,
+  verdictOver
+} from "./chunk-EZNIYU5T.js";
+import {
+  ApiArmIdSchema,
   TASK_CATEGORIES,
   getAdaptiveBonus,
-  getOutcomeStore
-} from "./chunk-NZN3SWSZ.js";
+  getOutcomeStore,
+  hasMeasuredCategory
+} from "./chunk-F2H3MCEV.js";
 
 // src/cli/doctor-deep.ts
 var CLI_NAMES = ["claude", "gemini", "codex", "opencode"];
 var COLD_START_THRESHOLD = 3;
+var ROUTED_ARMS = [...CLI_NAMES, ...ApiArmIdSchema.options];
 function checkLearningLoop() {
   const store = getOutcomeStore();
   const outcomes = store.query();
@@ -39,34 +46,50 @@ function checkDataSufficiency() {
   const categoriesWithData = /* @__PURE__ */ new Set();
   const allOutcomes = store.query();
   for (const o of allOutcomes) {
-    categoriesWithData.add(o.category);
+    if (hasMeasuredCategory(o)) categoriesWithData.add(o.category);
   }
   const missing = TASK_CATEGORIES.filter((c) => !categoriesWithData.has(c));
   return { cliStatus, missingCategories: missing, coldStartThreshold: COLD_START_THRESHOLD };
 }
+function roundRate(rate) {
+  return Math.round(rate * 1e3) / 1e3;
+}
+function tallyRowsByArm() {
+  const byArm = /* @__PURE__ */ new Map();
+  for (const o of getOutcomeStore().query()) {
+    if (!ROUTED_ARMS.includes(o.cli)) continue;
+    const tally = byArm.get(o.cli) ?? { total: 0, successes: 0 };
+    tally.total++;
+    if (o.success) tally.successes++;
+    byArm.set(o.cli, tally);
+  }
+  return byArm;
+}
 function checkConvergence() {
-  const store = getOutcomeStore();
+  const rowsByArm = tallyRowsByArm();
+  const measured = [];
   const rates = /* @__PURE__ */ new Map();
-  let totalRate = 0;
-  for (const cli of CLI_NAMES) {
-    const outcomes = store.query({ cli });
-    if (outcomes.length > 0) {
-      const rate = outcomes.filter((o) => o.success).length / outcomes.length;
-      rates.set(cli, Math.round(rate * 1e3) / 1e3);
-      totalRate += rate;
-    } else {
-      rates.set(cli, 0);
+  for (const arm of ROUTED_ARMS) {
+    const tally = rowsByArm.get(arm);
+    if (tally === void 0) {
+      rates.set(arm, { status: "unmeasured" });
+      continue;
     }
+    const rate = tally.successes / tally.total;
+    measured.push({ rate, sampleCount: tally.total });
+    rates.set(arm, { status: "measured", rate: roundRate(rate), sampleCount: tally.total });
   }
-  const avgRate = totalRate / CLI_NAMES.length;
-  const allAboveThreshold = CLI_NAMES.every((cli) => {
-    const outcomes = store.query({ cli });
-    return outcomes.length >= COLD_START_THRESHOLD;
-  });
+  const avgSuccessRate = verdictOver(
+    measured,
+    (arms) => roundRate(arms.reduce((sum, a) => sum + a.rate, 0) / arms.length),
+    "unmeasured"
+  );
+  const converged = allOf(measured, (a) => a.sampleCount >= COLD_START_THRESHOLD, false);
   return {
-    avgSuccessRate: Math.round(avgRate * 1e3) / 1e3,
-    cliSuccessRates: rates,
-    converged: allAboveThreshold
+    avgSuccessRate,
+    armSuccessRates: rates,
+    measuredArmCount: measured.length,
+    converged
   };
 }
 function runDeepDiagnostics() {
@@ -76,6 +99,23 @@ function runDeepDiagnostics() {
     routingConvergence: checkConvergence()
   };
 }
+function formatConvergenceRates(rc) {
+  const pct = (rate) => `${(rate * 100).toFixed(1)}%`;
+  if (rc.avgSuccessRate === "unmeasured") {
+    return ["  Avg success rate: unmeasured (no arm has outcome rows)"];
+  }
+  const noun = rc.measuredArmCount === 1 ? "arm" : "arms";
+  const lines = [
+    `  Avg success rate: ${pct(rc.avgSuccessRate)} over ${String(rc.measuredArmCount)} measured ${noun}`
+  ];
+  const unmeasured = [];
+  for (const [arm, r] of rc.armSuccessRates) {
+    if (r.status === "unmeasured") unmeasured.push(arm);
+    else lines.push(`    ${arm}: ${pct(r.rate)} (${String(r.sampleCount)} runs)`);
+  }
+  if (unmeasured.length > 0) lines.push(`    Unmeasured (no rows): ${unmeasured.join(", ")}`);
+  return lines;
+}
 function formatDeepDiagnostics(diag) {
   const lines = ["\n=== Deep Diagnostics ===\n"];
   const ll = diag.learningLoop;
@@ -97,7 +137,7 @@ function formatDeepDiagnostics(diag) {
   }
   const rc = diag.routingConvergence;
   lines.push("\nRouting Convergence:");
-  lines.push(`  Avg success rate: ${(rc.avgSuccessRate * 100).toFixed(1)}%`);
+  lines.push(...formatConvergenceRates(rc));
   lines.push(`  Converged: ${rc.converged ? "yes" : "no (still below cold-start threshold)"}`);
   return lines.join("\n");
 }
@@ -106,4 +146,4 @@ export {
   runDeepDiagnostics,
   formatDeepDiagnostics
 };
-//# sourceMappingURL=chunk-3PQIIMHN.js.map
+//# sourceMappingURL=chunk-5GI6Z5VE.js.map

package/dist/chunk-5GI6Z5VE.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/cli/doctor-deep.ts"],"sourcesContent":["/**\n * Deep diagnostics for the doctor command.\n *\n * Surfaces learning loop health, data sufficiency, routing convergence,\n * and memory system status. Opt-in via `--deep` flag.\n *\n * @module cli/doctor-deep\n * (Source: Issue #1031 — Enhanced doctor --deep diagnostics)\n */\n\nimport { getOutcomeStore } from '../orchestration/outcomes/outcome-store.js';\nimport { hasMeasuredCategory } from '../orchestration/outcomes/outcome-types.js';\nimport { TASK_CATEGORIES, type TaskCategory } from '../config/task-specialization-types.js';\nimport { getAdaptiveBonus } from '../mcp/tools/weather-report.js';\nimport { ApiArmIdSchema } from '../cli-adapters/types-core.js';\nimport { allOf, verdictOver } from '../utils/verdict-aggregation.js';\n\n// ============================================================================\n// Types\n// ============================================================================\n\n/** Per-CLI data sufficiency snapshot. */\nexport interface CliDataStatus {\n  readonly cli: string;\n  readonly taskCount: number;\n  readonly aboveThreshold: boolean;\n}\n\n/** Deep diagnostics result. */\nexport interface DeepDiagnostics {\n  readonly learningLoop: LearningLoopHealth;\n  readonly dataSufficiency: DataSufficiency;\n  readonly routingConvergence: RoutingConvergence;\n}\n\nexport interface LearningLoopHealth {\n  readonly totalOutcomes: number;\n  readonly latestTimestamp: string | null;\n  readonly activeBonuses: number;\n  readonly totalBonusPairs: number;\n}\n\nexport interface DataSufficiency {\n  readonly cliStatus: readonly CliDataStatus[];\n  readonly missingCategories: readonly string[];\n  readonly coldStartThreshold: number;\n}\n\n/**\n * One arm's success rate. An arm with no outcome rows is `unmeasured` — not a\n * 0% rate, which would claim every attempt failed (#6557).\n */\nexport type ArmSuccessRate =\n  | { readonly status: 'measured'; readonly rate: number; readonly sampleCount: number }\n  | { readonly status: 'unmeasured' };\n\nexport interface RoutingConvergence {\n  /**\n   * Mean success rate over MEASURED arms only. `'unmeasured'` when no arm has\n   * a row — the empty case, which is neither 0 nor NaN (#6557).\n   */\n  readonly avgSuccessRate: number | 'unmeasured';\n  /** Every routed arm (CLI slots and `api:*` arms), measured or not. */\n  readonly armSuccessRates: ReadonlyMap<string, ArmSuccessRate>;\n  /** Number of arms with at least one outcome row: the average's divisor. */\n  readonly measuredArmCount: number;\n  /** Every measured arm has cleared the cold-start threshold; false when none is measured. */\n  readonly converged: boolean;\n}\n\n// ============================================================================\n// Constants\n// ============================================================================\n\nconst CLI_NAMES = ['claude', 'gemini', 'codex', 'opencode'] as const;\nconst COLD_START_THRESHOLD = 3;\n\n/**\n * CLI slots plus API arms: every `cli` an attributed outcome row can carry.\n * Routed rows record the arm that ran (`api:anthropic`), not its slot, since\n * #6554. `unknown` is deliberately absent: an unattributed row is no arm's\n * measurement.\n */\nconst ROUTED_ARMS: readonly string[] = [...CLI_NAMES, ...ApiArmIdSchema.options];\n\n// ============================================================================\n// Diagnostics\n// ============================================================================\n\n/** Check learning loop health: outcome count, latest timestamp, bonuses. */\nfunction checkLearningLoop(): LearningLoopHealth {\n  const store = getOutcomeStore();\n  const outcomes = store.query();\n  const latestTimestamp =\n    outcomes.length > 0 ? (outcomes[outcomes.length - 1]?.timestamp ?? null) : null;\n\n  let activeBonuses = 0;\n  const totalPairs = CLI_NAMES.length * TASK_CATEGORIES.length;\n  for (const cli of CLI_NAMES) {\n    for (const cat of TASK_CATEGORIES) {\n      if (getAdaptiveBonus(cli, cat) !== 0) activeBonuses++;\n    }\n  }\n\n  return {\n    totalOutcomes: outcomes.length,\n    latestTimestamp,\n    activeBonuses,\n    totalBonusPairs: totalPairs,\n  };\n}\n\n/** Check per-CLI data sufficiency against cold-start threshold. */\nfunction checkDataSufficiency(): DataSufficiency {\n  const store = getOutcomeStore();\n  const cliStatus: CliDataStatus[] = [];\n\n  for (const cli of CLI_NAMES) {\n    const outcomes = store.query({ cli });\n    cliStatus.push({\n      cli,\n      taskCount: outcomes.length,\n      aboveThreshold: outcomes.length >= COLD_START_THRESHOLD,\n    });\n  }\n\n  const categoriesWithData = new Set<TaskCategory>();\n  const allOutcomes = store.query();\n  for (const o of allOutcomes) {\n    // A defaulted category is not coverage of that category (#6549).\n    if (hasMeasuredCategory(o)) categoriesWithData.add(o.category);\n  }\n  const missing = TASK_CATEGORIES.filter((c) => !categoriesWithData.has(c));\n\n  return { cliStatus, missingCategories: missing, coldStartThreshold: COLD_START_THRESHOLD };\n}\n\n/** Round a rate to three decimals for display-stable output. */\nfunction roundRate(rate: number): number {\n  return Math.round(rate * 1000) / 1000;\n}\n\ninterface MeasuredArm {\n  readonly rate: number;\n  readonly sampleCount: number;\n}\n\n/** Tally rows per routed arm; rows carrying no routed arm (`unknown`) are skipped. */\nfunction tallyRowsByArm(): Map<string, { total: number; successes: number }> {\n  const byArm = new Map<string, { total: number; successes: number }>();\n  for (const o of getOutcomeStore().query()) {\n    if (!ROUTED_ARMS.includes(o.cli)) continue;\n    const tally = byArm.get(o.cli) ?? { total: 0, successes: 0 };\n    tally.total++;\n    if (o.success) tally.successes++;\n    byArm.set(o.cli, tally);\n  }\n  return byArm;\n}\n\n/**\n * Check routing convergence from outcome success rates (#6557).\n *\n * Covers the arms rows actually carry, CLI and `api:*` alike. An arm with no\n * rows is `unmeasured` and stays out of the average, whose divisor is the\n * number of measured arms, never a fixed list.\n */\nfunction checkConvergence(): RoutingConvergence {\n  const rowsByArm = tallyRowsByArm();\n  const measured: MeasuredArm[] = [];\n  const rates = new Map<string, ArmSuccessRate>();\n  for (const arm of ROUTED_ARMS) {\n    const tally = rowsByArm.get(arm);\n    if (tally === undefined) {\n      rates.set(arm, { status: 'unmeasured' });\n      continue;\n    }\n    const rate = tally.successes / tally.total;\n    measured.push({ rate, sampleCount: tally.total });\n    rates.set(arm, { status: 'measured', rate: roundRate(rate), sampleCount: tally.total });\n  }\n\n  // Empty case named: with no measured arm there is no rate to average.\n  const avgSuccessRate = verdictOver<MeasuredArm, number | 'unmeasured'>(\n    measured,\n    (arms) => roundRate(arms.reduce((sum, a) => sum + a.rate, 0) / arms.length),\n    'unmeasured'\n  );\n  // Nothing measured has not converged on anything.\n  const converged = allOf(measured, (a) => a.sampleCount >= COLD_START_THRESHOLD, false);\n\n  return {\n    avgSuccessRate,\n    armSuccessRates: rates,\n    measuredArmCount: measured.length,\n    converged,\n  };\n}\n\n// ============================================================================\n// Public API\n// ============================================================================\n\n/** Run all deep diagnostics. */\nexport function runDeepDiagnostics(): DeepDiagnostics {\n  return {\n    learningLoop: checkLearningLoop(),\n    dataSufficiency: checkDataSufficiency(),\n    routingConvergence: checkConvergence(),\n  };\n}\n\n/** Render convergence rates, naming unmeasured arms rather than showing them as 0%. */\nfunction formatConvergenceRates(rc: RoutingConvergence): string[] {\n  const pct = (rate: number): string => `${(rate * 100).toFixed(1)}%`;\n  if (rc.avgSuccessRate === 'unmeasured') {\n    return ['  Avg success rate: unmeasured (no arm has outcome rows)'];\n  }\n  const noun = rc.measuredArmCount === 1 ? 'arm' : 'arms';\n  const lines = [\n    `  Avg success rate: ${pct(rc.avgSuccessRate)} over ${String(rc.measuredArmCount)} measured ${noun}`,\n  ];\n  const unmeasured: string[] = [];\n  for (const [arm, r] of rc.armSuccessRates) {\n    if (r.status === 'unmeasured') unmeasured.push(arm);\n    else lines.push(`    ${arm}: ${pct(r.rate)} (${String(r.sampleCount)} runs)`);\n  }\n  if (unmeasured.length > 0) lines.push(`    Unmeasured (no rows): ${unmeasured.join(', ')}`);\n  return lines;\n}\n\n/** Format deep diagnostics for CLI output. */\nexport function formatDeepDiagnostics(diag: DeepDiagnostics): string {\n  const lines: string[] = ['\\n=== Deep Diagnostics ===\\n'];\n\n  // Learning Loop\n  const ll = diag.learningLoop;\n  lines.push('Learning Loop:');\n  const outcomeIcon = ll.totalOutcomes > 0 ? '+' : '-';\n  lines.push(`  ${outcomeIcon} OutcomeStore: ${String(ll.totalOutcomes)} entries`);\n  const bonusIcon = ll.activeBonuses > 0 ? '+' : '-';\n  lines.push(\n    `  ${bonusIcon} Adaptive bonuses: ${String(ll.activeBonuses)}/${String(ll.totalBonusPairs)} active`\n  );\n\n  // Data Sufficiency\n  lines.push('\\nData Sufficiency:');\n  for (const cs of diag.dataSufficiency.cliStatus) {\n    const icon = cs.aboveThreshold ? '+' : '!';\n    const label = cs.aboveThreshold ? 'above threshold' : 'below threshold';\n    lines.push(`  ${icon} ${cs.cli}: ${String(cs.taskCount)} tasks (${label})`);\n  }\n  if (diag.dataSufficiency.missingCategories.length > 0) {\n    lines.push(`  Missing categories: ${diag.dataSufficiency.missingCategories.join(', ')}`);\n  }\n\n  // Routing Convergence\n  const rc = diag.routingConvergence;\n  lines.push('\\nRouting Convergence:');\n  lines.push(...formatConvergenceRates(rc));\n  lines.push(`  Converged: ${rc.converged ? 'yes' : 'no (still below cold-start threshold)'}`);\n\n  return lines.join('\\n');\n}\n"],"mappings":";;;;;;;;;;;;;AA0EA,IAAM,YAAY,CAAC,UAAU,UAAU,SAAS,UAAU;AAC1D,IAAM,uBAAuB;AAQ7B,IAAM,cAAiC,CAAC,GAAG,WAAW,GAAG,eAAe,OAAO;AAO/E,SAAS,oBAAwC;AAC/C,QAAM,QAAQ,gBAAgB;AAC9B,QAAM,WAAW,MAAM,MAAM;AAC7B,QAAM,kBACJ,SAAS,SAAS,IAAK,SAAS,SAAS,SAAS,CAAC,GAAG,aAAa,OAAQ;AAE7E,MAAI,gBAAgB;AACpB,QAAM,aAAa,UAAU,SAAS,gBAAgB;AACtD,aAAW,OAAO,WAAW;AAC3B,eAAW,OAAO,iBAAiB;AACjC,UAAI,iBAAiB,KAAK,GAAG,MAAM,EAAG;AAAA,IACxC;AAAA,EACF;AAEA,SAAO;AAAA,IACL,eAAe,SAAS;AAAA,IACxB;AAAA,IACA;AAAA,IACA,iBAAiB;AAAA,EACnB;AACF;AAGA,SAAS,uBAAwC;AAC/C,QAAM,QAAQ,gBAAgB;AAC9B,QAAM,YAA6B,CAAC;AAEpC,aAAW,OAAO,WAAW;AAC3B,UAAM,WAAW,MAAM,MAAM,EAAE,IAAI,CAAC;AACpC,cAAU,KAAK;AAAA,MACb;AAAA,MACA,WAAW,SAAS;AAAA,MACpB,gBAAgB,SAAS,UAAU;AAAA,IACrC,CAAC;AAAA,EACH;AAEA,QAAM,qBAAqB,oBAAI,IAAkB;AACjD,QAAM,cAAc,MAAM,MAAM;AAChC,aAAW,KAAK,aAAa;AAE3B,QAAI,oBAAoB,CAAC,EAAG,oBAAmB,IAAI,EAAE,QAAQ;AAAA,EAC/D;AACA,QAAM,UAAU,gBAAgB,OAAO,CAAC,MAAM,CAAC,mBAAmB,IAAI,CAAC,CAAC;AAExE,SAAO,EAAE,WAAW,mBAAmB,SAAS,oBAAoB,qBAAqB;AAC3F;AAGA,SAAS,UAAU,MAAsB;AACvC,SAAO,KAAK,MAAM,OAAO,GAAI,IAAI;AACnC;AAQA,SAAS,iBAAoE;AAC3E,QAAM,QAAQ,oBAAI,IAAkD;AACpE,aAAW,KAAK,gBAAgB,EAAE,MAAM,GAAG;AACzC,QAAI,CAAC,YAAY,SAAS,EAAE,GAAG,EAAG;AAClC,UAAM,QAAQ,MAAM,IAAI,EAAE,GAAG,KAAK,EAAE,OAAO,GAAG,WAAW,EAAE;AAC3D,UAAM;AACN,QAAI,EAAE,QAAS,OAAM;AACrB,UAAM,IAAI,EAAE,KAAK,KAAK;AAAA,EACxB;AACA,SAAO;AACT;AASA,SAAS,mBAAuC;AAC9C,QAAM,YAAY,eAAe;AACjC,QAAM,WAA0B,CAAC;AACjC,QAAM,QAAQ,oBAAI,IAA4B;AAC9C,aAAW,OAAO,aAAa;AAC7B,UAAM,QAAQ,UAAU,IAAI,GAAG;AAC/B,QAAI,UAAU,QAAW;AACvB,YAAM,IAAI,KAAK,EAAE,QAAQ,aAAa,CAAC;AACvC;AAAA,IACF;AACA,UAAM,OAAO,MAAM,YAAY,MAAM;AACrC,aAAS,KAAK,EAAE,MAAM,aAAa,MAAM,MAAM,CAAC;AAChD,UAAM,IAAI,KAAK,EAAE,QAAQ,YAAY,MAAM,UAAU,IAAI,GAAG,aAAa,MAAM,MAAM,CAAC;AAAA,EACxF;AAGA,QAAM,iBAAiB;AAAA,IACrB;AAAA,IACA,CAAC,SAAS,UAAU,KAAK,OAAO,CAAC,KAAK,MAAM,MAAM,EAAE,MAAM,CAAC,IAAI,KAAK,MAAM;AAAA,IAC1E;AAAA,EACF;AAEA,QAAM,YAAY,MAAM,UAAU,CAAC,MAAM,EAAE,eAAe,sBAAsB,KAAK;AAErF,SAAO;AAAA,IACL;AAAA,IACA,iBAAiB;AAAA,IACjB,kBAAkB,SAAS;AAAA,IAC3B;AAAA,EACF;AACF;AAOO,SAAS,qBAAsC;AACpD,SAAO;AAAA,IACL,cAAc,kBAAkB;AAAA,IAChC,iBAAiB,qBAAqB;AAAA,IACtC,oBAAoB,iBAAiB;AAAA,EACvC;AACF;AAGA,SAAS,uBAAuB,IAAkC;AAChE,QAAM,MAAM,CAAC,SAAyB,IAAI,OAAO,KAAK,QAAQ,CAAC,CAAC;AAChE,MAAI,GAAG,mBAAmB,cAAc;AACtC,WAAO,CAAC,0DAA0D;AAAA,EACpE;AACA,QAAM,OAAO,GAAG,qBAAqB,IAAI,QAAQ;AACjD,QAAM,QAAQ;AAAA,IACZ,uBAAuB,IAAI,GAAG,cAAc,CAAC,SAAS,OAAO,GAAG,gBAAgB,CAAC,aAAa,IAAI;AAAA,EACpG;AACA,QAAM,aAAuB,CAAC;AAC9B,aAAW,CAAC,KAAK,CAAC,KAAK,GAAG,iBAAiB;AACzC,QAAI,EAAE,WAAW,aAAc,YAAW,KAAK,GAAG;AAAA,QAC7C,OAAM,KAAK,OAAO,GAAG,KAAK,IAAI,EAAE,IAAI,CAAC,KAAK,OAAO,EAAE,WAAW,CAAC,QAAQ;AAAA,EAC9E;AACA,MAAI,WAAW,SAAS,EAAG,OAAM,KAAK,6BAA6B,WAAW,KAAK,IAAI,CAAC,EAAE;AAC1F,SAAO;AACT;AAGO,SAAS,sBAAsB,MAA+B;AACnE,QAAM,QAAkB,CAAC,8BAA8B;AAGvD,QAAM,KAAK,KAAK;AAChB,QAAM,KAAK,gBAAgB;AAC3B,QAAM,cAAc,GAAG,gBAAgB,IAAI,MAAM;AACjD,QAAM,KAAK,KAAK,WAAW,kBAAkB,OAAO,GAAG,aAAa,CAAC,UAAU;AAC/E,QAAM,YAAY,GAAG,gBAAgB,IAAI,MAAM;AAC/C,QAAM;AAAA,IACJ,KAAK,SAAS,sBAAsB,OAAO,GAAG,aAAa,CAAC,IAAI,OAAO,GAAG,eAAe,CAAC;AAAA,EAC5F;AAGA,QAAM,KAAK,qBAAqB;AAChC,aAAW,MAAM,KAAK,gBAAgB,WAAW;AAC/C,UAAM,OAAO,GAAG,iBAAiB,MAAM;AACvC,UAAM,QAAQ,GAAG,iBAAiB,oBAAoB;AACtD,UAAM,KAAK,KAAK,IAAI,IAAI,GAAG,GAAG,KAAK,OAAO,GAAG,SAAS,CAAC,WAAW,KAAK,GAAG;AAAA,EAC5E;AACA,MAAI,KAAK,gBAAgB,kBAAkB,SAAS,GAAG;AACrD,UAAM,KAAK,yBAAyB,KAAK,gBAAgB,kBAAkB,KAAK,IAAI,CAAC,EAAE;AAAA,EACzF;AAGA,QAAM,KAAK,KAAK;AAChB,QAAM,KAAK,wBAAwB;AACnC,QAAM,KAAK,GAAG,uBAAuB,EAAE,CAAC;AACxC,QAAM,KAAK,gBAAgB,GAAG,YAAY,QAAQ,uCAAuC,EAAE;AAE3F,SAAO,MAAM,KAAK,IAAI;AACxB;","names":[]}

package/dist/{chunk-TLFZPJOY.js → chunk-C6Q2NUIH.js}

@@ -1,10 +1,10 @@
 import {
   createAutoAdapter
-} from "./chunk-JYZNPK62.js";
+} from "./chunk-VIDSDHSQ.js";
 import {
   getDefaultCliCircuitBreakerRegistry,
   mapModelErrorToCategory
-} from "./chunk-XU2SJKFK.js";
+} from "./chunk-SMKBETIX.js";
 import {
   CircularBuffer,
   ConfigError,
@@ -28,7 +28,7 @@ import {
   resolveModelIdentitySync,
   toRateLimitError,
   warnIfGatewayCostUndeclared
-} from "./chunk-NZN3SWSZ.js";
+} from "./chunk-F2H3MCEV.js";
 
 // src/config/model-equivalence.ts
 function canonicalModelKey(modelId) {
@@ -975,4 +975,4 @@ export {
   getGlobalRegistry,
   resetGlobalRegistry
 };
-//# sourceMappingURL=chunk-TLFZPJOY.js.map
+//# sourceMappingURL=chunk-C6Q2NUIH.js.map

package/dist/{chunk-OQB3XC5O.js → chunk-EZNIYU5T.js}

@@ -7,9 +7,14 @@ function anyOf(items, predicate, whenEmpty) {
   if (items.length === 0) return whenEmpty;
   return items.some(predicate);
 }
+function verdictOver(items, aggregate, whenEmpty) {
+  if (items.length === 0) return whenEmpty;
+  return aggregate(items);
+}
 
 export {
   allOf,
-  anyOf
+  anyOf,
+  verdictOver
 };
-//# sourceMappingURL=chunk-OQB3XC5O.js.map
+//# sourceMappingURL=chunk-EZNIYU5T.js.map

package/dist/{chunk-OQB3XC5O.js.map → chunk-EZNIYU5T.js.map}

@@ -1 +1 @@
-{"version":3,"sources":["../src/utils/verdict-aggregation.ts"],"sourcesContent":["/**\n * Aggregating a verdict over a collection that might be empty (#4580).\n *\n * `[].every(p)` is `true` in JavaScript. So `checks.every((c) => c.passed)`\n * reports **pass** when `checks` is empty — absence rendered as health, on\n * exactly the code paths where that is most dangerous.\n *\n * Confirmed instances of the shape:\n *  - `verify_audit_chain` returns `ok: true` over zero events, so deleting\n *    every audit log file makes a tamper-evident chain verify clean (#4579)\n *  - a quality gate with zero checks reported `pass` (#4544)\n *  - the QA pipeline stage reported `success` after reviewing zero tasks,\n *    letting the graph advance as though implementations had been reviewed\n *\n * ## Why a helper rather than a lint or a type\n *\n * Measured before choosing: 64 non-test `.every()` calls, and most are fine —\n * `github-provider.ts:178` guards with `length > 0` and correctly leaves an\n * empty check set `pending`. A wide surface with sparse defects, so a blanket\n * lint would be mostly false positives, and false positives are what teach\n * people to bypass a gate.\n *\n * What was missing is not enforcement but a *decision point*: nothing made the\n * author say what empty means. `whenEmpty` is a required argument, so they\n * must. Chosen by a 7-voter `higher_order` panel at the supermajority bar.\n *\n * ## Choosing `whenEmpty`\n *\n * Ask what an empty collection is evidence OF. Usually nothing — in which case\n * the honest answer is the non-committal verdict (`pending`, `unmeasured`,\n * `skip`), not the optimistic one. Reserve `true` for cases where vacuous truth\n * is genuinely the contract, and say so at the call site.\n *\n * @module utils/verdict-aggregation\n * (Source: Issue #4580)\n */\n\n/**\n * `predicate` holds for every item — with the empty case named, not defaulted.\n *\n * @param items - The collection to judge.\n * @param predicate - Must hold for each item.\n * @param whenEmpty - The verdict when there is nothing to judge. Required.\n */\nexport function allOf<T>(\n  items: readonly T[],\n  predicate: (item: T) => boolean,\n  whenEmpty: boolean\n): boolean {\n  if (items.length === 0) return whenEmpty;\n  return items.every(predicate);\n}\n\n/**\n * `predicate` holds for at least one item — with the empty case named.\n *\n * `[].some(p)` is already `false`, which is usually right, but not always: a\n * \"did anything fail?\" check over zero results should often be `unmeasured`\n * rather than a clean `false`. Naming it keeps the reasoning visible.\n *\n * @param items - The collection to judge.\n * @param predicate - Must hold for at least one item.\n * @param whenEmpty - The verdict when there is nothing to judge. Required.\n */\nexport function anyOf<T>(\n  items: readonly T[],\n  predicate: (item: T) => boolean,\n  whenEmpty: boolean\n): boolean {\n  if (items.length === 0) return whenEmpty;\n  return items.some(predicate);\n}\n\n/**\n * Reduce a collection to an arbitrary verdict, with the empty case named.\n *\n * For verdicts richer than a boolean — a severity, a tri-state gate result —\n * where the empty case is usually `unmeasured` rather than the best value.\n *\n * @param items - The collection to judge.\n * @param aggregate - Folds a non-empty collection into a verdict.\n * @param whenEmpty - The verdict when there is nothing to judge. Required.\n */\nexport function verdictOver<T, V>(\n  items: readonly T[],\n  aggregate: (items: readonly T[]) => V,\n  whenEmpty: V\n): V {\n  if (items.length === 0) return whenEmpty;\n  return aggregate(items);\n}\n"],"mappings":";AA4CO,SAAS,MACd,OACA,WACA,WACS;AACT,MAAI,MAAM,WAAW,EAAG,QAAO;AAC/B,SAAO,MAAM,MAAM,SAAS;AAC9B;AAaO,SAAS,MACd,OACA,WACA,WACS;AACT,MAAI,MAAM,WAAW,EAAG,QAAO;AAC/B,SAAO,MAAM,KAAK,SAAS;AAC7B;","names":[]}
+{"version":3,"sources":["../src/utils/verdict-aggregation.ts"],"sourcesContent":["/**\n * Aggregating a verdict over a collection that might be empty (#4580).\n *\n * `[].every(p)` is `true` in JavaScript. So `checks.every((c) => c.passed)`\n * reports **pass** when `checks` is empty — absence rendered as health, on\n * exactly the code paths where that is most dangerous.\n *\n * Confirmed instances of the shape:\n *  - `verify_audit_chain` returns `ok: true` over zero events, so deleting\n *    every audit log file makes a tamper-evident chain verify clean (#4579)\n *  - a quality gate with zero checks reported `pass` (#4544)\n *  - the QA pipeline stage reported `success` after reviewing zero tasks,\n *    letting the graph advance as though implementations had been reviewed\n *\n * ## Why a helper rather than a lint or a type\n *\n * Measured before choosing: 64 non-test `.every()` calls, and most are fine —\n * `github-provider.ts:178` guards with `length > 0` and correctly leaves an\n * empty check set `pending`. A wide surface with sparse defects, so a blanket\n * lint would be mostly false positives, and false positives are what teach\n * people to bypass a gate.\n *\n * What was missing is not enforcement but a *decision point*: nothing made the\n * author say what empty means. `whenEmpty` is a required argument, so they\n * must. Chosen by a 7-voter `higher_order` panel at the supermajority bar.\n *\n * ## Choosing `whenEmpty`\n *\n * Ask what an empty collection is evidence OF. Usually nothing — in which case\n * the honest answer is the non-committal verdict (`pending`, `unmeasured`,\n * `skip`), not the optimistic one. Reserve `true` for cases where vacuous truth\n * is genuinely the contract, and say so at the call site.\n *\n * @module utils/verdict-aggregation\n * (Source: Issue #4580)\n */\n\n/**\n * `predicate` holds for every item — with the empty case named, not defaulted.\n *\n * @param items - The collection to judge.\n * @param predicate - Must hold for each item.\n * @param whenEmpty - The verdict when there is nothing to judge. Required.\n */\nexport function allOf<T>(\n  items: readonly T[],\n  predicate: (item: T) => boolean,\n  whenEmpty: boolean\n): boolean {\n  if (items.length === 0) return whenEmpty;\n  return items.every(predicate);\n}\n\n/**\n * `predicate` holds for at least one item — with the empty case named.\n *\n * `[].some(p)` is already `false`, which is usually right, but not always: a\n * \"did anything fail?\" check over zero results should often be `unmeasured`\n * rather than a clean `false`. Naming it keeps the reasoning visible.\n *\n * @param items - The collection to judge.\n * @param predicate - Must hold for at least one item.\n * @param whenEmpty - The verdict when there is nothing to judge. Required.\n */\nexport function anyOf<T>(\n  items: readonly T[],\n  predicate: (item: T) => boolean,\n  whenEmpty: boolean\n): boolean {\n  if (items.length === 0) return whenEmpty;\n  return items.some(predicate);\n}\n\n/**\n * Reduce a collection to an arbitrary verdict, with the empty case named.\n *\n * For verdicts richer than a boolean — a severity, a tri-state gate result —\n * where the empty case is usually `unmeasured` rather than the best value.\n *\n * @param items - The collection to judge.\n * @param aggregate - Folds a non-empty collection into a verdict.\n * @param whenEmpty - The verdict when there is nothing to judge. Required.\n */\nexport function verdictOver<T, V>(\n  items: readonly T[],\n  aggregate: (items: readonly T[]) => V,\n  whenEmpty: V\n): V {\n  if (items.length === 0) return whenEmpty;\n  return aggregate(items);\n}\n"],"mappings":";AA4CO,SAAS,MACd,OACA,WACA,WACS;AACT,MAAI,MAAM,WAAW,EAAG,QAAO;AAC/B,SAAO,MAAM,MAAM,SAAS;AAC9B;AAaO,SAAS,MACd,OACA,WACA,WACS;AACT,MAAI,MAAM,WAAW,EAAG,QAAO;AAC/B,SAAO,MAAM,KAAK,SAAS;AAC7B;AAYO,SAAS,YACd,OACA,WACA,WACG;AACH,MAAI,MAAM,WAAW,EAAG,QAAO;AAC/B,SAAO,UAAU,KAAK;AACxB;","names":[]}

package/dist/{chunk-NZN3SWSZ.js → chunk-F2H3MCEV.js}

@@ -2696,7 +2696,7 @@ function setDefaultRegistry(registry) {
 }
 async function reloadDefaultRegistry() {
   globalRegistry = buildDefaultRegistry();
-  const { resetGlobalRegistry } = await import("./unified-registry-LR4OCUFI.js");
+  const { resetGlobalRegistry } = await import("./unified-registry-HWBSGSOM.js");
   resetGlobalRegistry();
   return globalRegistry;
 }
@@ -10293,6 +10293,7 @@ var OutcomeFailureCategorySchema = z14.enum([
 var OutcomeCliSchema = z14.union([CliNameSchema, ApiArmIdSchema, z14.literal("unknown")]);
 var OutcomeCliSourceSchema = z14.enum(["executed", "category-default"]);
 var OutcomeRoutedBySchema = z14.enum(["composite-router"]);
+var OutcomeCategorySourceSchema = z14.enum(["detected", "defaulted"]);
 var TaskOutcomeSchema = z14.object({
   id: z14.string().min(1),
   cli: OutcomeCliSchema,
@@ -10301,6 +10302,8 @@ var TaskOutcomeSchema = z14.object({
   /** Set only when `CompositeRouter` selected the CLI for this task (#6521). */
   routedBy: OutcomeRoutedBySchema.optional(),
   category: TaskCategorySchema,
+  /** Whether `category` was detected or defaulted (#6549). Absent on legacy rows. */
+  categorySource: OutcomeCategorySourceSchema.optional(),
   model: z14.string().min(1),
   success: z14.boolean(),
   durationMs: z14.number().nonnegative(),
@@ -10357,6 +10360,16 @@ var OutcomeQuerySchema = z14.object({
   /** Restrict to outcomes recorded against a specific baseline (#2697). */
   baselineId: z14.string().min(1).max(64).optional()
 });
+var UNDETECTED_CATEGORY_FALLBACK = "exploration";
+function resolveOutcomeCategory(detected) {
+  if (detected === void 0) {
+    return { category: UNDETECTED_CATEGORY_FALLBACK, categorySource: "defaulted" };
+  }
+  return { category: detected, categorySource: "detected" };
+}
+function hasMeasuredCategory(outcome) {
+  return outcome.categorySource !== "defaulted";
+}
 var TIMEOUT_PATTERNS = ["timeout", "timed out", "deadline exceeded", "socket hang up", "aborted"];
 var AUTH_PATTERNS = [
   "auth",
@@ -10561,6 +10574,7 @@ function isDistillerEligible(outcome) {
   if (outcome.routedBy !== "composite-router") return false;
   if (outcome.source === "consensus") return false;
   if (!(outcome.durationMs > 0)) return false;
+  if (!hasMeasuredCategory(outcome)) return false;
   return ROUTABLE_CLIS.has(outcome.cli);
 }
 function countEligibleSince(outcomes, sinceMs) {
@@ -10579,7 +10593,8 @@ function getFileFieldsSchema() {
     cli: true,
     routedBy: true,
     durationMs: true,
-    timestamp: true
+    timestamp: true,
+    categorySource: true
   });
   return fileFieldsSchema;
 }
@@ -11074,7 +11089,9 @@ var OutcomeStore = class {
       successRate: successCount / outcomes.length,
       avgDurationMs: totalDuration / outcomes.length,
       byCli: groupBy(outcomes, (o) => o.cli),
-      byCategory: groupBy(outcomes, (o) => o.category)
+      // A defaulted category (#6549) is no category: totals count the row,
+      // the per-category breakdown does not.
+      byCategory: groupBy(outcomes.filter(hasMeasuredCategory), (o) => o.category)
     };
   }
   /** Number of stored outcomes. */
@@ -11194,7 +11211,9 @@ ${failLines.join("\n")}` : "";
 function buildPredicates(filter) {
   const preds = [];
   if (filter.cli !== void 0) preds.push((o) => o.cli === filter.cli);
-  if (filter.category !== void 0) preds.push((o) => o.category === filter.category);
+  if (filter.category !== void 0) {
+    preds.push((o) => hasMeasuredCategory(o) && o.category === filter.category);
+  }
   if (filter.source !== void 0) preds.push((o) => o.source === filter.source);
   if (filter.success !== void 0) preds.push((o) => o.success === filter.success);
   if (filter.failureCategory !== void 0) {
@@ -14111,7 +14130,7 @@ function buildCliWeather(summary, input) {
     const cliOutcomes = store.query({ cli });
     const byCategory = /* @__PURE__ */ new Map();
     for (const cat of TASK_CATEGORIES) {
-      const catOutcomes = cliOutcomes.filter((o) => o.category === cat);
+      const catOutcomes = cliOutcomes.filter((o) => hasMeasuredCategory(o) && o.category === cat);
       if (catOutcomes.length > 0) {
         const sc = catOutcomes.filter((o) => o.success).length;
         const td = catOutcomes.reduce((s, o) => s + o.durationMs, 0);
@@ -14276,7 +14295,9 @@ function buildSwarmHealth(expertPerf) {
   let observedCategories = 0;
   let analyzedCategories = 0;
   for (const category of TASK_CATEGORIES) {
-    const catOutcomes = allOutcomes.filter((o) => o.category === category);
+    const catOutcomes = allOutcomes.filter(
+      (o) => hasMeasuredCategory(o) && o.category === category
+    );
     if (catOutcomes.length < ROUTING_MIN_SAMPLES) continue;
     observedCategories++;
     const stats = analyzeCategoryRouting(catOutcomes);
@@ -16100,7 +16121,7 @@ var CompositeRouter = class _CompositeRouter {
    */
   async consultUnifiedContext(task) {
     try {
-      const { getContextForTask, inferTaskCategory } = await import("./context-retriever-UXQCDDRY.js");
+      const { getContextForTask, inferTaskCategory } = await import("./context-retriever-TWAN22PW.js");
       const ctx = await getContextForTask({
         task: task.content,
         category: inferTaskCategory(task.content),
@@ -17527,6 +17548,7 @@ export {
   loadGeneratedRegistryEntries,
   getFallbackChainForCategory,
   isCliName,
+  ApiArmIdSchema,
   apiArmId,
   routingArmDisplaySlot,
   isEndpointArmId,
@@ -17582,6 +17604,8 @@ export {
   validateTimeout,
   OutcomeFailureCategorySchema,
   TaskOutcomeSchema,
+  resolveOutcomeCategory,
+  hasMeasuredCategory,
   extractNonErrorMessage,
   categorizeOutcomeError,
   categorizeOutcomeErrorMessage,
@@ -17758,4 +17782,4 @@ export {
   AgentCapability,
   OrchestratorError
 };
-//# sourceMappingURL=chunk-NZN3SWSZ.js.map
+//# sourceMappingURL=chunk-F2H3MCEV.js.map