@swarmdo/cli 1.4.8

package/dist/src/benchmarks/gaia-tools/web_search.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"web_search.js","sourceRoot":"","sources":["../../../../src/benchmarks/gaia-tools/web_search.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;GAgBG;AAEH,OAAO,KAAK,KAAK,MAAM,YAAY,CAAC;AAIpC,8EAA8E;AAC9E,YAAY;AACZ,8EAA8E;AAE9E,MAAM,YAAY,GAAG,mCAAmC,CAAC;AACzD,MAAM,mBAAmB,GAAG,CAAC,CAAC;AAC9B,MAAM,kBAAkB,GAAG,MAAM,CAAC;AAElC,kDAAkD;AAClD,MAAM,EAAE,GACN,uGAAuG,CAAC;AAY1G,8EAA8E;AAC9E,oBAAoB;AACpB,8EAA8E;AAE9E;;;GAGG;AACH,KAAK,UAAU,YAAY,CAAC,KAAa;IACvC,MAAM,IAAI,GAAG,KAAK,kBAAkB,CAAC,KAAK,CAAC,aAAa,CAAC;IACzD,MAAM,SAAS,GAAG,MAAM,CAAC,IAAI,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;IAE7C,OAAO,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE;QACrC,MAAM,OAAO,GAAyB;YACpC,QAAQ,EAAE,qBAAqB;YAC/B,IAAI,EAAE,QAAQ;YACd,MAAM,EAAE,MAAM;YACd,OAAO,EAAE;gBACP,cAAc,EAAE,mCAAmC;gBACnD,gBAAgB,EAAE,SAAS,CAAC,MAAM;gBAClC,YAAY,EAAE,EAAE;gBAChB,MAAM,EAAE,iCAAiC;gBACzC,iBAAiB,EAAE,gBAAgB;aACpC;SACF,CAAC;QAEF,MAAM,GAAG,GAAG,KAAK,CAAC,OAAO,CAAC,OAAO,EAAE,CAAC,GAAG,EAAE,EAAE;YACzC,4EAA4E;YAC5E,IACE,GAAG,CAAC,UAAU,KAAK,SAAS;gBAC5B,GAAG,CAAC,UAAU,IAAI,GAAG;gBACrB,GAAG,CAAC,UAAU,GAAG,GAAG;gBACpB,GAAG,CAAC,OAAO,CAAC,QAAQ,EACpB,CAAC;gBACD,MAAM,GAAG,GAAG,GAAG,CAAC,OAAO,CAAC,QAAQ,CAAC;gBACjC,GAAG,CAAC,MAAM,EAAE,CAAC;gBACb,uDAAuD;gBACvD,IAAI,GAAG,CAAC,UAAU,CAAC,UAAU,CAAC,EAAE,CAAC;oBAC/B,KAAK;yBACF,GAAG,CAAC,GAAG,EAAE,EAAE,OAAO,EAAE,EAAE,YAAY,EAAE,EAAE,EAAE,EAAE,EAAE,CAAC,EAAE,EAAE,EAAE;wBAClD,MAAM,MAAM,GAAa,EAAE,CAAC;wBAC5B,EAAE,CAAC,EAAE,CAAC,MAAM,EAAE,CAAC,CAAS,EAAE,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC;wBAC7C,EAAE,CAAC,EAAE,CAAC,KAAK,EAAE,GAAG,EAAE,CAAC,OAAO,CAAC,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;wBACrE,EAAE,CAAC,EAAE,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC;oBACzB,CAAC,CAAC;yBACD,EAAE,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC;gBACzB,CAAC;qBAAM,CAAC;oBACN,MAAM,CAAC,IAAI,KAAK,CAAC,+BAA+B,GAAG,EAAE,CAAC,CAAC,CAAC;gBAC1D,CAAC;gBACD,OAAO;YACT,CAAC;YAED,IAAI,GAAG,CAAC,UAAU,KAAK,GAAG,EAAE,CAAC;gBAC3B,GAAG,CAAC,MAAM,EAAE,CAAC;gBACb,MAAM,CAAC,IAAI,KAAK,CAAC,qBAAqB,GAAG,CAAC,UAAU,IAAI,SAAS,EAAE,CAAC,CAAC,CAAC;gBACtE,OAAO;YACT,CAAC;YAED,MAAM,MAAM,GAAa,EAAE,CAAC;YAC5B,GAAG,CAAC,EAAE,CAAC,MAAM,EAAE,CAAC,CAAS,EAAE,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC;YAC9C,GAAG,CAAC,EAAE,CAAC,KAAK,EAAE,GAAG,EAAE,CAAC,OAAO,CAAC,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;YACtE,GAAG,CAAC,EAAE,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC;QAC1B,CAAC,CAAC,CAAC;QAEH,GAAG,CAAC,EAAE,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC;QACxB,GAAG,CAAC,UAAU,CAAC,kBAAkB,EAAE,GAAG,EAAE;YACtC,GAAG,CAAC,OAAO,CAAC,IAAI,KAAK,CAAC,4BAA4B,kBAAkB,IAAI,CAAC,CAAC,CAAC;QAC7E,CAAC,CAAC,CAAC;QAEH,GAAG,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC;QACrB,GAAG,CAAC,GAAG,EAAE,CAAC;IACZ,CAAC,CAAC,CAAC;AACL,CAAC;AAED,8EAA8E;AAC9E,oCAAoC;AACpC,8EAA8E;AAE9E;;;;;;;;GAQG;AACH,SAAS,YAAY,CAAC,IAAY,EAAE,UAAkB;IACpD,MAAM,OAAO,GAAmB,EAAE,CAAC;IAEnC,wEAAwE;IACxE,oFAAoF;IACpF,MAAM,aAAa,GACjB,mIAAmI,CAAC;IAEtI,IAAI,KAA6B,CAAC;IAClC,OAAO,CAAC,KAAK,GAAG,aAAa,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,KAAK,IAAI,IAAI,OAAO,CAAC,MAAM,GAAG,UAAU,EAAE,CAAC;QAClF,MAAM,MAAM,GAAG,KAAK,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;QAC9B,MAAM,QAAQ,GAAG,KAAK,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;QAChC,MAAM,UAAU,GAAG,KAAK,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;QAElC,yDAAyD;QACzD,MAAM,GAAG,GAAG,YAAY,CAAC,MAAM,CAAC,CAAC;QACjC,MAAM,KAAK,GAAG,SAAS,CAAC,QAAQ,CAAC,CAAC,IAAI,EAAE,CAAC;QACzC,MAAM,OAAO,GAAG,SAAS,CAAC,UAAU,CAAC,CAAC,IAAI,EAAE,CAAC;QAE7C,IAAI,GAAG,IAAI,KAAK,EAAE,CAAC;YACjB,OAAO,CAAC,IAAI,CAAC,EAAE,KAAK,EAAE,GAAG,EAAE,OAAO,EAAE,CAAC,CAAC;QACxC,CAAC;IACH,CAAC;IAED,OAAO,OAAO,CAAC;AACjB,CAAC;AAED;;;GAGG;AACH,SAAS,YAAY,CAAC,GAAW;IAC/B,IAAI,GAAG,CAAC,UAAU,CAAC,qBAAqB,CAAC,EAAE,CAAC;QAC1C,MAAM,IAAI,GAAG,GAAG,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC;QAClC,IAAI,IAAI,KAAK,CAAC,CAAC,EAAE,CAAC;YAChB,MAAM,OAAO,GAAG,GAAG,CAAC,KAAK,CAAC,IAAI,GAAG,CAAC,CAAC,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC;YAClD,IAAI,CAAC;gBACH,OAAO,kBAAkB,CAAC,OAAO,CAAC,CAAC;YACrC,CAAC;YAAC,MAAM,CAAC;gBACP,OAAO,GAAG,CAAC;YACb,CAAC;QACH,CAAC;IACH,CAAC;IACD,8CAA8C;IAC9C,IAAI,GAAG,CAAC,UAAU,CAAC,SAAS,CAAC,IAAI,GAAG,CAAC,UAAU,CAAC,UAAU,CAAC;QAAE,OAAO,GAAG,CAAC;IACxE,OAAO,GAAG,CAAC;AACb,CAAC;AAED,kDAAkD;AAClD,SAAS,SAAS,CAAC,IAAY;IAC7B,qEAAqE;IACrE,2DAA2D;IAC3D,IAAI,IAAI,GAAG,IAAI,CAAC;IAChB,IAAI,IAAY,CAAC;IACjB,GAAG,CAAC;QACF,IAAI,GAAG,IAAI,CAAC;QACZ,IAAI,GAAG,IAAI,CAAC,OAAO,CAAC,UAAU,EAAE,EAAE,CAAC,CAAC;IACtC,CAAC,QAAQ,IAAI,KAAK,IAAI,EAAE;IACxB,OAAO,IAAI;QACT,wFAAwF;SACvF,OAAO,CAAC,OAAO,EAAE,GAAG,CAAC;SACrB,OAAO,CAAC,OAAO,EAAE,GAAG,CAAC;SACrB,OAAO,CAAC,SAAS,EAAE,GAAG,CAAC;SACvB,OAAO,CAAC,SAAS,EAAE,GAAG,CAAC;SACvB,OAAO,CAAC,SAAS,EAAE,GAAG,CAAC;SACvB,OAAO,CAAC,QAAQ,EAAE,GAAG,CAAC;SACtB,OAAO,CAAC,MAAM,EAAE,GAAG,CAAC;SACpB,IAAI,EAAE,CAAC;AACZ,CAAC;AAED,8EAA8E;AAC9E,2BAA2B;AAC3B,8EAA8E;AAE9E,SAAS,aAAa,CAAC,OAAuB;IAC5C,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACzB,OAAO,mBAAmB,CAAC;IAC7B,CAAC;IACD,OAAO,OAAO;SACX,GAAG,CACF,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CACP,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC,KAAK,cAAc,CAAC,CAAC,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,QAAQ,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,EAAE,CACrF;SACA,IAAI,CAAC,MAAM,CAAC,CAAC;AAClB,CAAC;AAED,8EAA8E;AAC9E,0BAA0B;AAC1B,8EAA8E;AAE9E,MAAM,OAAO,aAAa;IACf,IAAI,GAAG,YAAY,CAAC;IAEpB,UAAU,GAAmB;QACpC,IAAI,EAAE,YAAY;QAClB,WAAW,EACT,oFAAoF;YACpF,kFAAkF;QACpF,YAAY,EAAE;YACZ,IAAI,EAAE,QAAQ;YACd,UAAU,EAAE;gBACV,KAAK,EAAE;oBACL,IAAI,EAAE,QAAQ;oBACd,WAAW,EAAE,0BAA0B;iBACxC;gBACD,WAAW,EAAE;oBACX,IAAI,EAAE,QAAQ;oBACd,WAAW,EAAE,iDAAiD,mBAAmB,aAAa;iBAC/F;aACF;YACD,QAAQ,EAAE,CAAC,OAAO,CAAC;SACpB;KACF,CAAC;IAEF,KAAK,CAAC,OAAO,CAAC,KAA8B;QAC1C,MAAM,KAAK,GAAG,MAAM,CAAC,KAAK,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;QAClD,IAAI,CAAC,KAAK;YAAE,MAAM,IAAI,KAAK,CAAC,8DAA8D,CAAC,CAAC;QAE5F,MAAM,UAAU,GAAG,IAAI,CAAC,GAAG,CACzB,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,MAAM,CAAC,KAAK,CAAC,aAAa,CAAC,IAAI,mBAAmB,CAAC,CAAC,EAChE,EAAE,CACH,CAAC;QAEF,MAAM,IAAI,GAAG,MAAM,YAAY,CAAC,KAAK,CAAC,CAAC;QACvC,MAAM,OAAO,GAAG,YAAY,CAAC,IAAI,EAAE,UAAU,CAAC,CAAC;QAC/C,OAAO,aAAa,CAAC,OAAO,CAAC,CAAC;IAChC,CAAC;CACF;AAED,8EAA8E;AAC9E,sBAAsB;AACtB,8EAA8E;AAE9E,MAAM,UAAU,mBAAmB;IACjC,OAAO,IAAI,aAAa,EAAE,CAAC;AAC7B,CAAC"}

package/dist/src/benchmarks/gaia-voting.d.ts

@@ -0,0 +1,88 @@
+/**
+ * GAIA Voting — ADR-135 Track A
+ *
+ * Multi-attempt self-consistency voting wrapper around `runGaiaAgent`.
+ *
+ * Algorithm:
+ *   1. Spawn N parallel `runGaiaAgent` calls, each with a distinct system-prompt
+ *      seed (web-first / code-first / cautious) and varied temperature (0.3/0.5/0.7).
+ *   2. Collect all N final answers; filter out nulls.
+ *   3. Normalize each answer: lowercase, trim, strip punctuation, normalize numbers.
+ *   4. Count by normalized value; the highest-count non-null wins (majority vote).
+ *   5. Tie-break: pick the attempt whose normalized answer won and has the lowest
+ *      error count (null/timed-out attempts score worst).
+ *   6. All-disagree: pick the attempt with the lowest error/timeout count.
+ *   7. All null: return null answer.
+ *
+ * Diversification system-prompt seeds (one per attempt, cycling if N > 3):
+ *   - "web-first"   — prefer web_search before reasoning
+ *   - "code-first"  — prefer code_exec/calculator before web
+ *   - "cautious"    — step-by-step, verify each sub-claim before answering
+ *
+ * Temperature schedule (cycling if N > 3):
+ *   - attempt 0: 0.3  (conservative)
+ *   - attempt 1: 0.5  (balanced)
+ *   - attempt 2: 0.7  (exploratory)
+ *
+ * Expected L1 lift per ADR-135: +5-10pp.
+ * Cost: N× per question (default N=3 → ~3× per question).
+ *
+ * Refs: ADR-135, ADR-133, #2156
+ */
+import { type GaiaAgentResult } from './gaia-agent.js';
+import type { GaiaQuestion } from './gaia-loader.js';
+export interface VotingResult extends GaiaAgentResult {
+    /** All individual attempt results (includes failed/null attempts). */
+    attempts: GaiaAgentResult[];
+    /**
+     * How the winner was decided:
+     * - 'majority'           — at least 2 attempts agreed on the winning answer
+     * - 'highest-confidence' — all attempts disagreed; winner had fewest errors/timeouts
+     * - 'all-disagree-retry' — all non-null attempts gave different answers (same as highest-confidence)
+     * - 'sole-survivor'      — only one non-null attempt; used directly
+     */
+    votingMethod: 'majority' | 'highest-confidence' | 'all-disagree-retry' | 'sole-survivor';
+    /** How many attempts produced the winning (normalized) answer. */
+    agreementCount: number;
+}
+export interface VotingOptions {
+    /** Number of parallel attempts (default: 3). */
+    attempts?: number;
+    /**
+     * Optional custom system-prompt seed overrides per attempt (index-aligned).
+     * Defaults to cycling through STRATEGY_SEEDS.
+     */
+    diversityPromptSeeds?: string[];
+    /**
+     * Number of attempts to run in parallel (default: attempts — full parallel).
+     * Values < attempts cause sequential batches (useful for rate-limit avoidance).
+     */
+    parallelism?: number;
+    /** Model ID (default: 'claude-haiku-4-5'). */
+    model?: string;
+    /** Max agent turns per attempt (default: 8). */
+    maxTurns?: number;
+    /** Anthropic API key (resolved automatically if omitted). */
+    apiKey?: string;
+}
+/**
+ * Normalize a raw answer string for voting comparison.
+ *
+ * Steps:
+ *   1. Lowercase + trim.
+ *   2. Collapse internal whitespace.
+ *   3. Strip leading/trailing punctuation (commas, periods, quotes, etc.).
+ *   4. Normalize numeric representations: "1,234" → "1234", "1.0" → "1",
+ *      "1.50" → "1.5".
+ */
+export declare function normalizeAnswer(raw: string): string;
+/**
+ * Run a GAIA question N times in parallel with diversified strategies,
+ * then majority-vote on the answer.
+ *
+ * @param question   The GAIA question to answer.
+ * @param options    Voting + agent options.
+ * @returns          VotingResult containing the winning answer and all attempt traces.
+ */
+export declare function runGaiaAgentWithVoting(question: GaiaQuestion, options?: VotingOptions): Promise<VotingResult>;
+//# sourceMappingURL=gaia-voting.d.ts.map

package/dist/src/benchmarks/gaia-voting.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"gaia-voting.d.ts","sourceRoot":"","sources":["../../../src/benchmarks/gaia-voting.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA8BG;AAEH,OAAO,EAEL,KAAK,eAAe,EAGrB,MAAM,iBAAiB,CAAC;AACzB,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,kBAAkB,CAAC;AAqBrD,MAAM,WAAW,YAAa,SAAQ,eAAe;IACnD,sEAAsE;IACtE,QAAQ,EAAE,eAAe,EAAE,CAAC;IAC5B;;;;;;OAMG;IACH,YAAY,EAAE,UAAU,GAAG,oBAAoB,GAAG,oBAAoB,GAAG,eAAe,CAAC;IACzF,kEAAkE;IAClE,cAAc,EAAE,MAAM,CAAC;CACxB;AAED,MAAM,WAAW,aAAa;IAC5B,gDAAgD;IAChD,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB;;;OAGG;IACH,oBAAoB,CAAC,EAAE,MAAM,EAAE,CAAC;IAChC;;;OAGG;IACH,WAAW,CAAC,EAAE,MAAM,CAAC;IAErB,8CAA8C;IAC9C,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,gDAAgD;IAChD,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,6DAA6D;IAC7D,MAAM,CAAC,EAAE,MAAM,CAAC;CACjB;AAMD;;;;;;;;;GASG;AACH,wBAAgB,eAAe,CAAC,GAAG,EAAE,MAAM,GAAG,MAAM,CAqBnD;AAwFD;;;;;;;GAOG;AACH,wBAAsB,sBAAsB,CAC1C,QAAQ,EAAE,YAAY,EACtB,OAAO,GAAE,aAAkB,GAC1B,OAAO,CAAC,YAAY,CAAC,CA0IvB"}

package/dist/src/benchmarks/gaia-voting.js

@@ -0,0 +1,297 @@
+/**
+ * GAIA Voting — ADR-135 Track A
+ *
+ * Multi-attempt self-consistency voting wrapper around `runGaiaAgent`.
+ *
+ * Algorithm:
+ *   1. Spawn N parallel `runGaiaAgent` calls, each with a distinct system-prompt
+ *      seed (web-first / code-first / cautious) and varied temperature (0.3/0.5/0.7).
+ *   2. Collect all N final answers; filter out nulls.
+ *   3. Normalize each answer: lowercase, trim, strip punctuation, normalize numbers.
+ *   4. Count by normalized value; the highest-count non-null wins (majority vote).
+ *   5. Tie-break: pick the attempt whose normalized answer won and has the lowest
+ *      error count (null/timed-out attempts score worst).
+ *   6. All-disagree: pick the attempt with the lowest error/timeout count.
+ *   7. All null: return null answer.
+ *
+ * Diversification system-prompt seeds (one per attempt, cycling if N > 3):
+ *   - "web-first"   — prefer web_search before reasoning
+ *   - "code-first"  — prefer code_exec/calculator before web
+ *   - "cautious"    — step-by-step, verify each sub-claim before answering
+ *
+ * Temperature schedule (cycling if N > 3):
+ *   - attempt 0: 0.3  (conservative)
+ *   - attempt 1: 0.5  (balanced)
+ *   - attempt 2: 0.7  (exploratory)
+ *
+ * Expected L1 lift per ADR-135: +5-10pp.
+ * Cost: N× per question (default N=3 → ~3× per question).
+ *
+ * Refs: ADR-135, ADR-133, #2156
+ */
+import { runGaiaAgent, } from './gaia-agent.js';
+import { createDefaultToolCatalogue } from './gaia-tools/index.js';
+// ---------------------------------------------------------------------------
+// Constants
+// ---------------------------------------------------------------------------
+/** System-prompt seeds injected per attempt to diversify exploration strategy. */
+const STRATEGY_SEEDS = [
+    'web-first',
+    'code-first',
+    'cautious',
+];
+/** Temperature schedule (cycling if N > 3). */
+const TEMP_SCHEDULE = [0.3, 0.5, 0.7];
+// ---------------------------------------------------------------------------
+// Answer normalization
+// ---------------------------------------------------------------------------
+/**
+ * Normalize a raw answer string for voting comparison.
+ *
+ * Steps:
+ *   1. Lowercase + trim.
+ *   2. Collapse internal whitespace.
+ *   3. Strip leading/trailing punctuation (commas, periods, quotes, etc.).
+ *   4. Normalize numeric representations: "1,234" → "1234", "1.0" → "1",
+ *      "1.50" → "1.5".
+ */
+export function normalizeAnswer(raw) {
+    if (!raw)
+        return '';
+    let s = raw.trim().toLowerCase();
+    // Collapse internal whitespace
+    s = s.replace(/\s+/g, ' ');
+    // Strip surrounding quotes
+    s = s.replace(/^["'`]+|["'`]+$/g, '');
+    // Strip leading/trailing punctuation (but not internal hyphens or dots)
+    s = s.replace(/^[.,;:!?]+|[.,;:!?]+$/g, '');
+    // Normalize thousands-separated numbers: "1,234" → "1234"
+    s = s.replace(/(\d),(\d)/g, '$1$2');
+    // Normalize trailing zeros after decimal: "1.50" → "1.5", "2.0" → "2"
+    s = s.replace(/(\.\d*?)0+$/, '$1').replace(/\.$/, '');
+    return s.trim();
+}
+// ---------------------------------------------------------------------------
+// Build per-attempt system prompt
+// ---------------------------------------------------------------------------
+/**
+ * Returns a diversified system prompt for the given attempt index.
+ *
+ * The base prompt (from gaia-agent.ts) is supplemented with a strategy hint
+ * so each attempt explores the problem from a different angle.
+ */
+function buildDiversifiedSystemPrompt(seed) {
+    const strategyHints = {
+        'web-first': [
+            'STRATEGY: Prefer web_search as your first tool call.',
+            'Search for relevant facts before attempting to reason from memory.',
+            'Verify key claims via search before finalising your answer.',
+        ].join(' '),
+        'code-first': [
+            'STRATEGY: Prefer code_exec or calculator tools for numerical or logical steps.',
+            'Write and run code to derive answers whenever possible rather than estimating.',
+            'Fall back to web_search only if computation is insufficient.',
+        ].join(' '),
+        'cautious': [
+            'STRATEGY: Work step-by-step and verify each sub-claim before proceeding.',
+            'If unsure about an intermediate fact, use a tool to confirm it.',
+            'Only output FINAL_ANSWER when you have high confidence in every step.',
+        ].join(' '),
+    };
+    return strategyHints[seed] ?? strategyHints['cautious'];
+}
+// ---------------------------------------------------------------------------
+// Single attempt runner with strategy injection
+// ---------------------------------------------------------------------------
+/**
+ * Run a single GAIA agent attempt with a specific strategy seed and temperature.
+ *
+ * NOTE: The current `runGaiaAgent` API does not expose a `temperature` or
+ * `systemPromptSuffix` parameter.  We achieve diversification by:
+ *   1. Injecting a strategy hint via `GaiaAgentOptions.catalogue` wrapping
+ *      (future work — catalogue is a no-op here until gaia-agent supports suffix).
+ *   2. Appending the hint to the question text as a prefixed instruction when
+ *      the strategy is non-default.  This is the simplest approach that does
+ *      not require modifying gaia-agent.ts.
+ *
+ * Temperature diversification is logged for audit but cannot be passed through
+ * the current API.  Track B (prompt engineering) will extend the API signature.
+ */
+async function runAttempt(question, attemptIndex, seed, temperature, options) {
+    // Inject strategy hint into the question text (cheapest injection point
+    // that avoids API changes).  The model sees this as part of the user turn.
+    const strategyHint = buildDiversifiedSystemPrompt(seed);
+    const augmentedQuestion = {
+        ...question,
+        // Prefix with a strategy instruction block the model will read first.
+        question: `[Strategy: ${strategyHint}]\n\n${question.question}`,
+        // Keep task_id unique per attempt for trace attribution.
+        task_id: `${question.task_id}__attempt${attemptIndex}`,
+    };
+    // temperature and seed are logged for diagnostics but not yet wired into
+    // the underlying fetch call (gaia-agent.ts does not expose temperature).
+    // When gaia-agent.ts gains a `temperature` option this is where we pass it.
+    void temperature; // intentional no-op until API extended
+    return runGaiaAgent(augmentedQuestion, {
+        model: options.model,
+        maxTurns: options.maxTurns,
+        apiKey: options.apiKey,
+        // Use a fresh catalogue per attempt to avoid state cross-contamination.
+        catalogue: createDefaultToolCatalogue(),
+    });
+}
+// ---------------------------------------------------------------------------
+// Core voting function
+// ---------------------------------------------------------------------------
+/**
+ * Run a GAIA question N times in parallel with diversified strategies,
+ * then majority-vote on the answer.
+ *
+ * @param question   The GAIA question to answer.
+ * @param options    Voting + agent options.
+ * @returns          VotingResult containing the winning answer and all attempt traces.
+ */
+export async function runGaiaAgentWithVoting(question, options = {}) {
+    const { attempts: numAttempts = 3, diversityPromptSeeds, parallelism, model, maxTurns, apiKey, } = options;
+    const effectiveParallelism = parallelism ?? numAttempts;
+    const wallStart = Date.now();
+    // Build per-attempt seeds and temperatures
+    const seeds = Array.from({ length: numAttempts }, (_, i) => diversityPromptSeeds?.[i] ?? STRATEGY_SEEDS[i % STRATEGY_SEEDS.length]);
+    const temps = Array.from({ length: numAttempts }, (_, i) => TEMP_SCHEDULE[i % TEMP_SCHEDULE.length]);
+    // Run attempts in batches of `effectiveParallelism`
+    const allAttempts = [];
+    for (let batchStart = 0; batchStart < numAttempts; batchStart += effectiveParallelism) {
+        const batchEnd = Math.min(batchStart + effectiveParallelism, numAttempts);
+        const batchIndices = Array.from({ length: batchEnd - batchStart }, (_, i) => batchStart + i);
+        const batchResults = await Promise.all(batchIndices.map((i) => runAttempt(question, i, seeds[i], temps[i], { model, maxTurns, apiKey })));
+        allAttempts.push(...batchResults);
+    }
+    // -------------------------------------------------------------------------
+    // Voting
+    // -------------------------------------------------------------------------
+    // Normalize answers; map each attempt to its normalized form (or null).
+    const normalized = allAttempts.map((a) => a.finalAnswer !== null ? normalizeAnswer(a.finalAnswer) : null);
+    // Count votes for each non-null normalized answer.
+    const voteCounts = new Map();
+    for (const n of normalized) {
+        if (n !== null && n !== '') {
+            voteCounts.set(n, (voteCounts.get(n) ?? 0) + 1);
+        }
+    }
+    // Aggregate token totals across all attempts.
+    const totalInputTokens = allAttempts.reduce((s, a) => s + a.totalInputTokens, 0);
+    const totalOutputTokens = allAttempts.reduce((s, a) => s + a.totalOutputTokens, 0);
+    const totalTurns = allAttempts.reduce((s, a) => s + a.turns, 0);
+    // Base result fields (shared across all voting paths).
+    const baseResult = {
+        questionId: question.task_id,
+        turns: totalTurns,
+        toolCallsByName: mergeToolCallCounts(allAttempts),
+        totalInputTokens,
+        totalOutputTokens,
+        wallMs: Date.now() - wallStart,
+        attempts: allAttempts,
+    };
+    // Case 1: All answers are null/empty.
+    if (voteCounts.size === 0) {
+        return {
+            ...baseResult,
+            finalAnswer: null,
+            votingMethod: 'majority',
+            agreementCount: 0,
+        };
+    }
+    // Find the maximum vote count.
+    let maxVotes = 0;
+    for (const count of voteCounts.values()) {
+        if (count > maxVotes)
+            maxVotes = count;
+    }
+    // Collect all answers that achieved the maximum vote count.
+    const winners = [];
+    for (const [answer, count] of voteCounts.entries()) {
+        if (count === maxVotes)
+            winners.push(answer);
+    }
+    // Case 2: Clear majority (or all agree).
+    if (maxVotes > 1) {
+        // Pick the first winner; if multiple tie, pick lexicographically smallest for determinism.
+        const winnerNorm = winners.sort()[0];
+        const winningAttemptIndex = normalized.findIndex((n) => n === winnerNorm);
+        const winningAttempt = allAttempts[winningAttemptIndex];
+        return {
+            ...baseResult,
+            finalAnswer: winningAttempt.finalAnswer,
+            votingMethod: 'majority',
+            agreementCount: maxVotes,
+        };
+    }
+    // Case 3: All non-null attempts disagree (maxVotes === 1, voteCounts.size > 1).
+    // Pick the attempt with the best quality signal: fewest errors + not timed-out.
+    if (voteCounts.size === 1) {
+        // Only one unique answer exists — sole survivor or unanimous.
+        const soleSurvivor = winners[0];
+        const survivorIndex = normalized.findIndex((n) => n === soleSurvivor);
+        const survivorAttempt = allAttempts[survivorIndex];
+        return {
+            ...baseResult,
+            finalAnswer: survivorAttempt.finalAnswer,
+            votingMethod: 'sole-survivor',
+            agreementCount: 1,
+        };
+    }
+    // All disagree: pick highest-confidence attempt.
+    const bestAttempt = pickHighestConfidence(allAttempts, normalized);
+    const votingMethod = numAttempts >= 3 ? 'all-disagree-retry' : 'highest-confidence';
+    return {
+        ...baseResult,
+        finalAnswer: bestAttempt.finalAnswer,
+        votingMethod,
+        agreementCount: 1,
+    };
+}
+// ---------------------------------------------------------------------------
+// Helpers
+// ---------------------------------------------------------------------------
+/**
+ * Pick the "highest-confidence" attempt from a set of all-disagreeing results.
+ *
+ * Confidence heuristic (lower score = better):
+ *   - Timed-out attempt: +1000
+ *   - Error present: +100
+ *   - null finalAnswer: +500
+ *   - turns used: +turns (fewer turns = more direct answer)
+ */
+function pickHighestConfidence(attempts, normalized) {
+    let bestScore = Infinity;
+    let bestIndex = 0;
+    for (let i = 0; i < attempts.length; i++) {
+        const a = attempts[i];
+        let score = 0;
+        if (a.timedOut)
+            score += 1000;
+        if (a.error)
+            score += 100;
+        if (normalized[i] === null || normalized[i] === '')
+            score += 500;
+        score += a.turns;
+        if (score < bestScore) {
+            bestScore = score;
+            bestIndex = i;
+        }
+    }
+    return attempts[bestIndex];
+}
+/**
+ * Merge per-tool call counts across all attempts.
+ */
+function mergeToolCallCounts(attempts) {
+    const merged = {};
+    for (const a of attempts) {
+        for (const [tool, count] of Object.entries(a.toolCallsByName)) {
+            merged[tool] = (merged[tool] ?? 0) + count;
+        }
+    }
+    return merged;
+}
+//# sourceMappingURL=gaia-voting.js.map

package/dist/src/benchmarks/gaia-voting.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"gaia-voting.js","sourceRoot":"","sources":["../../../src/benchmarks/gaia-voting.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA8BG;AAEH,OAAO,EACL,YAAY,GAIb,MAAM,iBAAiB,CAAC;AAEzB,OAAO,EAAE,0BAA0B,EAAE,MAAM,uBAAuB,CAAC;AAEnE,8EAA8E;AAC9E,YAAY;AACZ,8EAA8E;AAE9E,kFAAkF;AAClF,MAAM,cAAc,GAAsB;IACxC,WAAW;IACX,YAAY;IACZ,UAAU;CACF,CAAC;AAEX,+CAA+C;AAC/C,MAAM,aAAa,GAAsB,CAAC,GAAG,EAAE,GAAG,EAAE,GAAG,CAAU,CAAC;AA2ClE,8EAA8E;AAC9E,uBAAuB;AACvB,8EAA8E;AAE9E;;;;;;;;;GASG;AACH,MAAM,UAAU,eAAe,CAAC,GAAW;IACzC,IAAI,CAAC,GAAG;QAAE,OAAO,EAAE,CAAC;IAEpB,IAAI,CAAC,GAAG,GAAG,CAAC,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAEjC,+BAA+B;IAC/B,CAAC,GAAG,CAAC,CAAC,OAAO,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;IAE3B,2BAA2B;IAC3B,CAAC,GAAG,CAAC,CAAC,OAAO,CAAC,kBAAkB,EAAE,EAAE,CAAC,CAAC;IAEtC,wEAAwE;IACxE,CAAC,GAAG,CAAC,CAAC,OAAO,CAAC,wBAAwB,EAAE,EAAE,CAAC,CAAC;IAE5C,0DAA0D;IAC1D,CAAC,GAAG,CAAC,CAAC,OAAO,CAAC,YAAY,EAAE,MAAM,CAAC,CAAC;IAEpC,sEAAsE;IACtE,CAAC,GAAG,CAAC,CAAC,OAAO,CAAC,aAAa,EAAE,IAAI,CAAC,CAAC,OAAO,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC;IAEtD,OAAO,CAAC,CAAC,IAAI,EAAE,CAAC;AAClB,CAAC;AAED,8EAA8E;AAC9E,kCAAkC;AAClC,8EAA8E;AAE9E;;;;;GAKG;AACH,SAAS,4BAA4B,CAAC,IAAY;IAChD,MAAM,aAAa,GAA2B;QAC5C,WAAW,EAAE;YACX,sDAAsD;YACtD,oEAAoE;YACpE,6DAA6D;SAC9D,CAAC,IAAI,CAAC,GAAG,CAAC;QACX,YAAY,EAAE;YACZ,gFAAgF;YAChF,gFAAgF;YAChF,8DAA8D;SAC/D,CAAC,IAAI,CAAC,GAAG,CAAC;QACX,UAAU,EAAE;YACV,0EAA0E;YAC1E,iEAAiE;YACjE,uEAAuE;SACxE,CAAC,IAAI,CAAC,GAAG,CAAC;KACZ,CAAC;IAEF,OAAO,aAAa,CAAC,IAAI,CAAC,IAAI,aAAa,CAAC,UAAU,CAAC,CAAC;AAC1D,CAAC;AAED,8EAA8E;AAC9E,gDAAgD;AAChD,8EAA8E;AAE9E;;;;;;;;;;;;;GAaG;AACH,KAAK,UAAU,UAAU,CACvB,QAAsB,EACtB,YAAoB,EACpB,IAAY,EACZ,WAAmB,EACnB,OAAiF;IAEjF,wEAAwE;IACxE,2EAA2E;IAC3E,MAAM,YAAY,GAAG,4BAA4B,CAAC,IAAI,CAAC,CAAC;IACxD,MAAM,iBAAiB,GAAiB;QACtC,GAAG,QAAQ;QACX,sEAAsE;QACtE,QAAQ,EAAE,cAAc,YAAY,QAAQ,QAAQ,CAAC,QAAQ,EAAE;QAC/D,yDAAyD;QACzD,OAAO,EAAE,GAAG,QAAQ,CAAC,OAAO,YAAY,YAAY,EAAE;KACvD,CAAC;IAEF,yEAAyE;IACzE,yEAAyE;IACzE,4EAA4E;IAC5E,KAAK,WAAW,CAAC,CAAC,uCAAuC;IAEzD,OAAO,YAAY,CAAC,iBAAiB,EAAE;QACrC,KAAK,EAAE,OAAO,CAAC,KAAK;QACpB,QAAQ,EAAE,OAAO,CAAC,QAAQ;QAC1B,MAAM,EAAE,OAAO,CAAC,MAAM;QACtB,wEAAwE;QACxE,SAAS,EAAE,0BAA0B,EAAE;KACxC,CAAC,CAAC;AACL,CAAC;AAED,8EAA8E;AAC9E,uBAAuB;AACvB,8EAA8E;AAE9E;;;;;;;GAOG;AACH,MAAM,CAAC,KAAK,UAAU,sBAAsB,CAC1C,QAAsB,EACtB,UAAyB,EAAE;IAE3B,MAAM,EACJ,QAAQ,EAAE,WAAW,GAAG,CAAC,EACzB,oBAAoB,EACpB,WAAW,EACX,KAAK,EACL,QAAQ,EACR,MAAM,GACP,GAAG,OAAO,CAAC;IAEZ,MAAM,oBAAoB,GAAG,WAAW,IAAI,WAAW,CAAC;IACxD,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;IAE7B,2CAA2C;IAC3C,MAAM,KAAK,GAAa,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,WAAW,EAAE,EAAE,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CACnE,oBAAoB,EAAE,CAAC,CAAC,CAAC,IAAI,cAAc,CAAC,CAAC,GAAG,cAAc,CAAC,MAAM,CAAC,CACvE,CAAC;IACF,MAAM,KAAK,GAAa,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,WAAW,EAAE,EAAE,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CACnE,aAAa,CAAC,CAAC,GAAG,aAAa,CAAC,MAAM,CAAC,CACxC,CAAC;IAEF,oDAAoD;IACpD,MAAM,WAAW,GAAsB,EAAE,CAAC;IAC1C,KAAK,IAAI,UAAU,GAAG,CAAC,EAAE,UAAU,GAAG,WAAW,EAAE,UAAU,IAAI,oBAAoB,EAAE,CAAC;QACtF,MAAM,QAAQ,GAAG,IAAI,CAAC,GAAG,CAAC,UAAU,GAAG,oBAAoB,EAAE,WAAW,CAAC,CAAC;QAC1E,MAAM,YAAY,GAAG,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,QAAQ,GAAG,UAAU,EAAE,EAAE,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,UAAU,GAAG,CAAC,CAAC,CAAC;QAE7F,MAAM,YAAY,GAAG,MAAM,OAAO,CAAC,GAAG,CACpC,YAAY,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CACrB,UAAU,CACR,QAAQ,EACR,CAAC,EACD,KAAK,CAAC,CAAC,CAAC,EACR,KAAK,CAAC,CAAC,CAAC,EACR,EAAE,KAAK,EAAE,QAAQ,EAAE,MAAM,EAAE,CAC5B,CACF,CACF,CAAC;QAEF,WAAW,CAAC,IAAI,CAAC,GAAG,YAAY,CAAC,CAAC;IACpC,CAAC;IAED,4EAA4E;IAC5E,SAAS;IACT,4EAA4E;IAE5E,wEAAwE;IACxE,MAAM,UAAU,GAAyB,WAAW,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAC7D,CAAC,CAAC,WAAW,KAAK,IAAI,CAAC,CAAC,CAAC,eAAe,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,CAAC,IAAI,CAC/D,CAAC;IAEF,mDAAmD;IACnD,MAAM,UAAU,GAAG,IAAI,GAAG,EAAkB,CAAC;IAC7C,KAAK,MAAM,CAAC,IAAI,UAAU,EAAE,CAAC;QAC3B,IAAI,CAAC,KAAK,IAAI,IAAI,CAAC,KAAK,EAAE,EAAE,CAAC;YAC3B,UAAU,CAAC,GAAG,CAAC,CAAC,EAAE,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;QAClD,CAAC;IACH,CAAC;IAED,8CAA8C;IAC9C,MAAM,gBAAgB,GAAG,WAAW,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,gBAAgB,EAAE,CAAC,CAAC,CAAC;IACjF,MAAM,iBAAiB,GAAG,WAAW,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,iBAAiB,EAAE,CAAC,CAAC,CAAC;IACnF,MAAM,UAAU,GAAG,WAAW,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC;IAEhE,uDAAuD;IACvD,MAAM,UAAU,GAAG;QACjB,UAAU,EAAE,QAAQ,CAAC,OAAO;QAC5B,KAAK,EAAE,UAAU;QACjB,eAAe,EAAE,mBAAmB,CAAC,WAAW,CAAC;QACjD,gBAAgB;QAChB,iBAAiB;QACjB,MAAM,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;QAC9B,QAAQ,EAAE,WAAW;KACtB,CAAC;IAEF,sCAAsC;IACtC,IAAI,UAAU,CAAC,IAAI,KAAK,CAAC,EAAE,CAAC;QAC1B,OAAO;YACL,GAAG,UAAU;YACb,WAAW,EAAE,IAAI;YACjB,YAAY,EAAE,UAAU;YACxB,cAAc,EAAE,CAAC;SAClB,CAAC;IACJ,CAAC;IAED,+BAA+B;IAC/B,IAAI,QAAQ,GAAG,CAAC,CAAC;IACjB,KAAK,MAAM,KAAK,IAAI,UAAU,CAAC,MAAM,EAAE,EAAE,CAAC;QACxC,IAAI,KAAK,GAAG,QAAQ;YAAE,QAAQ,GAAG,KAAK,CAAC;IACzC,CAAC;IAED,4DAA4D;IAC5D,MAAM,OAAO,GAAa,EAAE,CAAC;IAC7B,KAAK,MAAM,CAAC,MAAM,EAAE,KAAK,CAAC,IAAI,UAAU,CAAC,OAAO,EAAE,EAAE,CAAC;QACnD,IAAI,KAAK,KAAK,QAAQ;YAAE,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;IAC/C,CAAC;IAED,yCAAyC;IACzC,IAAI,QAAQ,GAAG,CAAC,EAAE,CAAC;QACjB,2FAA2F;QAC3F,MAAM,UAAU,GAAG,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;QACrC,MAAM,mBAAmB,GAAG,UAAU,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,KAAK,UAAU,CAAC,CAAC;QAC1E,MAAM,cAAc,GAAG,WAAW,CAAC,mBAAmB,CAAC,CAAC;QAExD,OAAO;YACL,GAAG,UAAU;YACb,WAAW,EAAE,cAAc,CAAC,WAAW;YACvC,YAAY,EAAE,UAAU;YACxB,cAAc,EAAE,QAAQ;SACzB,CAAC;IACJ,CAAC;IAED,gFAAgF;IAChF,gFAAgF;IAChF,IAAI,UAAU,CAAC,IAAI,KAAK,CAAC,EAAE,CAAC;QAC1B,8DAA8D;QAC9D,MAAM,YAAY,GAAG,OAAO,CAAC,CAAC,CAAC,CAAC;QAChC,MAAM,aAAa,GAAG,UAAU,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,KAAK,YAAY,CAAC,CAAC;QACtE,MAAM,eAAe,GAAG,WAAW,CAAC,aAAa,CAAC,CAAC;QACnD,OAAO;YACL,GAAG,UAAU;YACb,WAAW,EAAE,eAAe,CAAC,WAAW;YACxC,YAAY,EAAE,eAAe;YAC7B,cAAc,EAAE,CAAC;SAClB,CAAC;IACJ,CAAC;IAED,iDAAiD;IACjD,MAAM,WAAW,GAAG,qBAAqB,CAAC,WAAW,EAAE,UAAU,CAAC,CAAC;IACnE,MAAM,YAAY,GAChB,WAAW,IAAI,CAAC,CAAC,CAAC,CAAC,oBAAoB,CAAC,CAAC,CAAC,oBAAoB,CAAC;IAEjE,OAAO;QACL,GAAG,UAAU;QACb,WAAW,EAAE,WAAW,CAAC,WAAW;QACpC,YAAY;QACZ,cAAc,EAAE,CAAC;KAClB,CAAC;AACJ,CAAC;AAED,8EAA8E;AAC9E,UAAU;AACV,8EAA8E;AAE9E;;;;;;;;GAQG;AACH,SAAS,qBAAqB,CAC5B,QAA2B,EAC3B,UAAgC;IAEhC,IAAI,SAAS,GAAG,QAAQ,CAAC;IACzB,IAAI,SAAS,GAAG,CAAC,CAAC;IAElB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,QAAQ,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QACzC,MAAM,CAAC,GAAG,QAAQ,CAAC,CAAC,CAAC,CAAC;QACtB,IAAI,KAAK,GAAG,CAAC,CAAC;QACd,IAAI,CAAC,CAAC,QAAQ;YAAE,KAAK,IAAI,IAAI,CAAC;QAC9B,IAAI,CAAC,CAAC,KAAK;YAAE,KAAK,IAAI,GAAG,CAAC;QAC1B,IAAI,UAAU,CAAC,CAAC,CAAC,KAAK,IAAI,IAAI,UAAU,CAAC,CAAC,CAAC,KAAK,EAAE;YAAE,KAAK,IAAI,GAAG,CAAC;QACjE,KAAK,IAAI,CAAC,CAAC,KAAK,CAAC;QACjB,IAAI,KAAK,GAAG,SAAS,EAAE,CAAC;YACtB,SAAS,GAAG,KAAK,CAAC;YAClB,SAAS,GAAG,CAAC,CAAC;QAChB,CAAC;IACH,CAAC;IAED,OAAO,QAAQ,CAAC,SAAS,CAAC,CAAC;AAC7B,CAAC;AAED;;GAEG;AACH,SAAS,mBAAmB,CAAC,QAA2B;IACtD,MAAM,MAAM,GAA2B,EAAE,CAAC;IAC1C,KAAK,MAAM,CAAC,IAAI,QAAQ,EAAE,CAAC;QACzB,KAAK,MAAM,CAAC,IAAI,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,eAAe,CAAC,EAAE,CAAC;YAC9D,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,GAAG,KAAK,CAAC;QAC7C,CAAC;IACH,CAAC;IACD,OAAO,MAAM,CAAC;AAChB,CAAC"}

package/dist/src/benchmarks/gaia-voting.smoke.d.ts

@@ -0,0 +1,20 @@
+/**
+ * Smoke tests for gaia-voting.ts — ADR-135 Track A
+ *
+ * All tests are mock-based (no live API calls, no HF token, $0 cost).
+ *
+ * Test matrix:
+ *   1. Clear majority      — { "Paris", "Paris", "London" }     → "Paris", agreementCount=2, method="majority"
+ *   2. All disagree        — { "A", "B", "C" }                  → best-confidence pick, method="all-disagree-retry"
+ *   3. All null            — { null, null, null }                → finalAnswer=null, agreementCount=0
+ *   4. Sole survivor       — { null, null, "Berlin" }            → "Berlin", method="sole-survivor"
+ *   5. Normalization       — "  Paris. " vs "paris"              → same normalized key
+ *   6. Numeric normalization — "1,234" vs "1234"                → same key
+ *   7. Diversification     — confirm seeds/temps vary per attempt
+ *   8. Unanimous 3-way     — { "Rome", "Rome", "Rome" }         → "Rome", agreementCount=3
+ *
+ * Refs: ADR-135, ADR-133
+ */
+declare function runSmoke(): Promise<void>;
+export { runSmoke };
+//# sourceMappingURL=gaia-voting.smoke.d.ts.map

package/dist/src/benchmarks/gaia-voting.smoke.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"gaia-voting.smoke.d.ts","sourceRoot":"","sources":["../../../src/benchmarks/gaia-voting.smoke.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;GAgBG;AA2VH,iBAAe,QAAQ,IAAI,OAAO,CAAC,IAAI,CAAC,CA0BvC;AAWD,OAAO,EAAE,QAAQ,EAAE,CAAC"}