@swarmdo/cli 1.4.8

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

@@ -0,0 +1,437 @@
+/**
+ * GAIA Judge — ADR-133-PR6
+ *
+ * Two-stage answer scorer for the GAIA benchmark:
+ *
+ *   Stage 1 — Fast path: normalized exact-match.
+ *     Normalise = lowercase + strip surrounding whitespace + strip surrounding
+ *     single/double quotes + collapse internal whitespace runs to one space.
+ *     Roughly 30 % of GAIA Level-1 answers satisfy this; no API call required.
+ *
+ *   Stage 2 — LLM-as-judge: when exact-match fails, ask Claude Sonnet whether
+ *     the candidate answer is semantically equivalent to the ground truth.
+ *     The prompt embeds GAIA's official evaluation guideline (see
+ *     https://huggingface.co/datasets/gaia-benchmark/GAIA for full spec).
+ *
+ * Caching: judgment results are persisted under
+ *   ~/.cache/swarmdo/gaia/judgments/<hash>.json
+ * keyed on (question_id, candidate_answer, model_id, JUDGE_PROMPT_VERSION).
+ * Re-running the same pair hits the cache and returns instantly.
+ *
+ * API pattern: raw fetch() against https://api.anthropic.com/v1/messages —
+ * mirrors the pattern established in gaia-agent.ts (ADR-133-PR3).
+ *
+ * Refs: ADR-133, #2156
+ */
+import { createHash } from 'node:crypto';
+import * as fs from 'node:fs';
+import * as path from 'node:path';
+import * as os from 'node:os';
+import { execSync } from 'node:child_process';
+// ---------------------------------------------------------------------------
+// Constants
+// ---------------------------------------------------------------------------
+const ANTHROPIC_API_URL = 'https://api.anthropic.com/v1/messages';
+const ANTHROPIC_API_VERSION = '2023-06-01';
+const DEFAULT_JUDGE_MODEL = 'claude-sonnet-4-6';
+const DEFAULT_CACHE_DIR = path.join(os.homedir(), '.cache', 'swarmdo', 'gaia', 'judgments');
+/**
+ * Bump this string whenever the judge prompt changes so stale cached verdicts
+ * are automatically invalidated (different key → cache miss).
+ */
+const JUDGE_PROMPT_VERSION = 'v1';
+// Sonnet pricing (input/output per million tokens, 2026-05-27).
+const SONNET_INPUT_COST_PER_M = 3.0;
+const SONNET_OUTPUT_COST_PER_M = 15.0;
+// ---------------------------------------------------------------------------
+// Normalisation
+// ---------------------------------------------------------------------------
+/**
+ * GAIA normalisation as specified in the dataset paper:
+ *   - strip surrounding whitespace
+ *   - lowercase
+ *   - strip a single pair of surrounding quotes (single or double)
+ *   - collapse internal whitespace runs to one space
+ */
+export function normaliseAnswer(raw) {
+    if (raw == null)
+        return '';
+    let s = raw.trim().toLowerCase();
+    // Strip one pair of surrounding quotes
+    if ((s.startsWith('"') && s.endsWith('"')) ||
+        (s.startsWith("'") && s.endsWith("'"))) {
+        s = s.slice(1, -1);
+    }
+    // Collapse internal whitespace
+    s = s.replace(/\s+/g, ' ').trim();
+    return s;
+}
+// ---------------------------------------------------------------------------
+// Unit-aware numeric matching
+// ---------------------------------------------------------------------------
+/**
+ * Attempt to match a candidate numeric answer to an expected answer where the
+ * question implies a unit scale.
+ *
+ * Examples that this catches:
+ *   candidate="17000", expected="17", question contains "thousand"
+ *     → candidate / 1000 ≈ expected → MATCH
+ *   candidate="17", expected="17000", question contains "thousand"
+ *     → candidate × 1000 ≈ expected → MATCH (reverse direction)
+ *
+ * Returns true only when a numeric match is found under one of the scale
+ * multipliers mentioned in the question text.  Returns false for non-numeric
+ * inputs or when no multiplier matches.
+ *
+ * @param candidate    - The raw string from the model (may include commas/spaces).
+ * @param expected     - The raw ground-truth string.
+ * @param questionText - The original question (used to detect multiplier words).
+ */
+export function unitAwareNumberMatch(candidate, expected, questionText) {
+    // Strip commas, spaces, and trailing unit suffixes for numeric parsing
+    const toNum = (s) => parseFloat(s.replace(/[,\s]/g, ''));
+    const candNum = toNum(candidate);
+    const expNum = toNum(expected);
+    if (isNaN(candNum) || isNaN(expNum))
+        return false;
+    // Exact numeric equality (handles "17" vs "17.0", etc.)
+    if (Math.abs(candNum - expNum) < 0.001 * (Math.abs(expNum) + 1))
+        return true;
+    if (!questionText)
+        return false;
+    const qLower = questionText.toLowerCase();
+    const MULTIPLIERS = [
+        ['trillion', 1e12],
+        ['billion', 1e9],
+        ['million', 1e6],
+        ['thousand', 1e3],
+        ['hundred', 1e2],
+    ];
+    for (const [word, mult] of MULTIPLIERS) {
+        if (!qLower.includes(word))
+            continue;
+        // Model returned raw, expected is already in scaled units
+        // e.g. model says "17000", question asks "how many thousand hours", expected is "17"
+        if (Math.abs(candNum / mult - expNum) < 0.01 * (Math.abs(expNum) + 1))
+            return true;
+        // Reverse: model returned scaled, expected is raw
+        if (Math.abs(candNum - expNum / mult) < 0.01 * (Math.abs(candNum) + 1))
+            return true;
+    }
+    return false;
+}
+// ---------------------------------------------------------------------------
+// Cache helpers
+// ---------------------------------------------------------------------------
+function cacheKey(questionId, candidateAnswer, judgeModel) {
+    const raw = `${questionId}||${candidateAnswer}||${judgeModel}||${JUDGE_PROMPT_VERSION}`;
+    return createHash('sha256').update(raw).digest('hex');
+}
+function cacheRead(cacheDir, key) {
+    const file = path.join(cacheDir, `${key}.json`);
+    try {
+        const txt = fs.readFileSync(file, 'utf-8');
+        return JSON.parse(txt);
+    }
+    catch {
+        return null;
+    }
+}
+function cacheWrite(cacheDir, key, result) {
+    try {
+        fs.mkdirSync(cacheDir, { recursive: true });
+        fs.writeFileSync(path.join(cacheDir, `${key}.json`), JSON.stringify(result, null, 2), 'utf-8');
+    }
+    catch {
+        // Non-fatal — cache write failure should not abort a benchmark run.
+    }
+}
+// ---------------------------------------------------------------------------
+// API key resolution (mirrors gaia-agent.ts resolveAnthropicApiKey)
+// ---------------------------------------------------------------------------
+function resolveApiKey(supplied) {
+    if (supplied && supplied.trim())
+        return supplied.trim();
+    const envKey = process.env.ANTHROPIC_API_KEY;
+    if (envKey && envKey.trim())
+        return envKey.trim();
+    try {
+        const out = execSync('gcloud secrets versions access latest --secret=ANTHROPIC_API_KEY 2>/dev/null', { encoding: 'utf-8', timeout: 10_000 }).trim();
+        if (out)
+            return out;
+    }
+    catch {
+        /* fall through */
+    }
+    throw new Error('ANTHROPIC_API_KEY not found.  Set the env var or store it in GCP Secret Manager under ' +
+        '"ANTHROPIC_API_KEY".');
+}
+// ---------------------------------------------------------------------------
+// LLM-as-judge prompt
+// ---------------------------------------------------------------------------
+/**
+ * Build the judge system prompt.
+ *
+ * References the GAIA official scoring guideline:
+ *   "The evaluation of the answer is done by exact string match after
+ *    normalisation.  For numerical answers, units are ignored unless
+ *    the question explicitly asks for them.  For named-entity answers,
+ *    common aliases are accepted.  For open-ended questions where an exact
+ *    match is not possible, the answer is judged correct if it is semantically
+ *    equivalent to the ground truth and contains all required information."
+ *   Source: https://huggingface.co/datasets/gaia-benchmark/GAIA (README, §Evaluation)
+ */
+function buildJudgeSystemPrompt() {
+    return [
+        'You are a precise judge evaluating whether a candidate answer to a',
+        'question-answering benchmark is correct.',
+        '',
+        'SCORING RULES (from the GAIA benchmark specification):',
+        '1. Exact-string equivalence (after normalisation) is always correct.',
+        '2. For NUMERICAL answers: ignore units unless the question explicitly requests them.',
+        '   "3.14" and "approximately 3.14" for "what is pi to 2 decimal places" are both correct.',
+        '3. For NAMED-ENTITY answers: accept common aliases and alternative spellings.',
+        '   "UK" and "United Kingdom" are equivalent.',
+        '4. For LIST answers: the candidate must contain all required items; extra items are ok.',
+        '5. Do NOT accept answers that are vague or incomplete when the ground truth is specific.',
+        '   "a European city" is NOT correct if the ground truth is "Paris".',
+        '',
+        'You MUST respond with a single JSON object on one line, exactly this shape:',
+        '{"passed": true, "reason": "..."}  or  {"passed": false, "reason": "..."}',
+        'The "reason" must be 200 characters or fewer.',
+        'Do not output anything outside the JSON object.',
+    ].join('\n');
+}
+function buildJudgeUserMessage(question, groundTruth, candidate) {
+    return [
+        `QUESTION: ${question}`,
+        `GROUND TRUTH: ${groundTruth}`,
+        `CANDIDATE ANSWER: ${candidate}`,
+        '',
+        'Is the candidate answer correct per the scoring rules above?',
+    ].join('\n');
+}
+async function callJudge(systemPrompt, userMessage, model, apiKey) {
+    const messages = [
+        { role: 'user', content: userMessage },
+    ];
+    const body = JSON.stringify({
+        model,
+        max_tokens: 256,
+        system: systemPrompt,
+        messages,
+    });
+    const response = await fetch(ANTHROPIC_API_URL, {
+        method: 'POST',
+        headers: {
+            'Content-Type': 'application/json',
+            'x-api-key': apiKey,
+            'anthropic-version': ANTHROPIC_API_VERSION,
+        },
+        body,
+    });
+    if (!response.ok) {
+        const errText = await response.text().catch(() => '(no body)');
+        throw new Error(`Anthropic API error ${response.status}: ${errText}`);
+    }
+    const data = (await response.json());
+    const textBlock = data.content.find((c) => c.type === 'text');
+    const rawText = textBlock?.text ?? '';
+    // Parse the JSON the model produced
+    let passed = false;
+    let reason = '';
+    try {
+        // The model might wrap the JSON in a code fence — strip it
+        const jsonStr = rawText.replace(/^```(?:json)?\s*/i, '').replace(/\s*```$/, '').trim();
+        const parsed = JSON.parse(jsonStr);
+        passed = Boolean(parsed.passed);
+        reason = String(parsed.reason ?? '').slice(0, 200);
+    }
+    catch {
+        // Fallback: scan the text for obvious pass/fail signals
+        const lower = rawText.toLowerCase();
+        passed = lower.includes('"passed": true') || lower.includes('"passed":true');
+        reason = `parse error — raw: ${rawText.slice(0, 100)}`;
+    }
+    return {
+        passed,
+        reason,
+        tokensIn: data.usage.input_tokens,
+        tokensOut: data.usage.output_tokens,
+    };
+}
+// ---------------------------------------------------------------------------
+// Public API
+// ---------------------------------------------------------------------------
+/**
+ * Judge a single GAIA answer.
+ *
+ * @param question   - Object with `id` (task_id), `expected` (ground truth),
+ *                     and optional `questionText` (the full question string,
+ *                     used for unit-aware numeric matching in Stage 1).
+ * @param candidateAnswer - The answer produced by the agent; `null` counts as a miss.
+ * @param options    - Optional overrides (model, cache dir, API key, etc.).
+ * @returns          - JudgeResult with pass/fail, scoring path, and cost metrics.
+ */
+export async function judgeAnswer(question, candidateAnswer, options) {
+    const judgeModel = options?.judgeModel ?? DEFAULT_JUDGE_MODEL;
+    const cacheDir = options?.cacheDir ?? DEFAULT_CACHE_DIR;
+    const candidate = candidateAnswer ?? '';
+    // ── Stage 0: null / empty candidate is always a miss ──
+    if (!candidate.trim()) {
+        return {
+            questionId: question.id,
+            passed: false,
+            scoringPath: 'exact-match',
+            candidateAnswer: candidate,
+            groundTruth: question.expected,
+        };
+    }
+    // ── Stage 1a: normalised exact-match (no API call) ──
+    const normCandidate = normaliseAnswer(candidate);
+    const normExpected = normaliseAnswer(question.expected);
+    if (normCandidate === normExpected) {
+        return {
+            questionId: question.id,
+            passed: true,
+            scoringPath: 'exact-match',
+            candidateAnswer: candidate,
+            groundTruth: question.expected,
+        };
+    }
+    // ── Stage 1b: unit-aware numeric match (no API call) ──
+    // Handles cases like model returns "17000" but expected is "17" and
+    // the question asks "how many thousand hours".
+    if (unitAwareNumberMatch(normCandidate, normExpected, question.questionText)) {
+        return {
+            questionId: question.id,
+            passed: true,
+            scoringPath: 'exact-match',
+            candidateAnswer: candidate,
+            groundTruth: question.expected,
+        };
+    }
+    // ── Cache lookup (before calling the LLM) ──
+    const key = cacheKey(question.id, candidate, judgeModel);
+    if (!options?.skipCache) {
+        const cached = cacheRead(cacheDir, key);
+        if (cached !== null) {
+            return { ...cached, scoringPath: 'cache' };
+        }
+    }
+    // ── Stage 2: LLM-as-judge ──
+    const apiKey = resolveApiKey(options?.apiKey);
+    const systemPrompt = buildJudgeSystemPrompt();
+    const userMessage = buildJudgeUserMessage(question.questionText ?? question.expected, question.expected, candidate);
+    const { passed, reason, tokensIn, tokensOut } = await callJudge(systemPrompt, userMessage, judgeModel, apiKey);
+    const costUsd = (tokensIn / 1_000_000) * SONNET_INPUT_COST_PER_M +
+        (tokensOut / 1_000_000) * SONNET_OUTPUT_COST_PER_M;
+    const result = {
+        questionId: question.id,
+        passed,
+        scoringPath: 'llm-judge',
+        candidateAnswer: candidate,
+        groundTruth: question.expected,
+        judgeReason: reason,
+        judgeModel,
+        judgeTokensIn: tokensIn,
+        judgeTokensOut: tokensOut,
+        judgeCostUsd: costUsd,
+    };
+    // Persist to cache (even failures — avoids repeated LLM calls on re-run)
+    cacheWrite(cacheDir, key, result);
+    return result;
+}
+// ---------------------------------------------------------------------------
+// Smoke runner
+// ---------------------------------------------------------------------------
+/**
+ * Self-contained smoke test.  Run with:
+ *   npx tsx src/benchmarks/gaia-judge.ts
+ *
+ * Does NOT require an ANTHROPIC_API_KEY for the exact-match cases.
+ * The LLM-judge cases require a live key and cost ~$0.001 total.
+ *
+ * Expected cost: ≤ 2 Sonnet judge calls × ~300 tokens ≈ $0.001
+ */
+async function runSmoke() {
+    const PASS = '\x1b[32mPASS\x1b[0m';
+    const FAIL = '\x1b[31mFAIL\x1b[0m';
+    let failures = 0;
+    function check(label, condition) {
+        if (condition) {
+            console.log(`  ${PASS}  ${label}`);
+        }
+        else {
+            console.log(`  ${FAIL}  ${label}`);
+            failures++;
+        }
+    }
+    // Use a temp cache dir so smoke runs are isolated
+    const tmpCacheDir = path.join(os.tmpdir(), `gaia-judge-smoke-${Date.now()}`);
+    const baseOpts = { cacheDir: tmpCacheDir };
+    console.log('\n=== gaia-judge smoke ===\n');
+    // ── Stage 1a: normaliseAnswer unit tests (no API call, no judgeAnswer) ──
+    console.log('-- Stage 1a: normaliseAnswer --');
+    check('normalise("346") === "346"', normaliseAnswer('346') === '346');
+    check('normalise(" YES ") === "yes"', normaliseAnswer(' YES ') === 'yes');
+    check('normalise(\'"Paris"\') === "paris"', normaliseAnswer('"Paris"') === 'paris');
+    check('normalise("hello  world") === "hello world"', normaliseAnswer('hello  world') === 'hello world');
+    check('normalise(null) === ""', normaliseAnswer(null) === '');
+    check('"346" !== "347" after normalise', normaliseAnswer('346') !== normaliseAnswer('347'));
+    // ── Stage 1b: exact-match hit and null-candidate cases (no API call) ──
+    console.log('\n-- Stage 1b: exact-match path --');
+    const r1 = await judgeAnswer({ id: 'em-1', expected: '346' }, '346', baseOpts);
+    check('exact match "346" vs "346" → pass, exact-match path', r1.passed && r1.scoringPath === 'exact-match');
+    const r3 = await judgeAnswer({ id: 'em-3', expected: 'yes' }, ' YES ', baseOpts);
+    check('normalised "yes" vs " YES " → pass, exact-match path', r3.passed && r3.scoringPath === 'exact-match');
+    const r4 = await judgeAnswer({ id: 'em-4', expected: 'Paris' }, '"Paris"', baseOpts);
+    check('quote-stripped "Paris" vs \'"Paris"\' → pass, exact-match path', r4.passed && r4.scoringPath === 'exact-match');
+    const r5 = await judgeAnswer({ id: 'em-5', expected: 'hello world' }, 'hello  world', baseOpts);
+    check('whitespace-collapsed "hello world" → pass, exact-match path', r5.passed && r5.scoringPath === 'exact-match');
+    const rNull = await judgeAnswer({ id: 'em-null', expected: '346' }, null, baseOpts);
+    check('null candidate → fail, exact-match path', !rNull.passed && rNull.scoringPath === 'exact-match');
+    // ── LLM-judge cases (requires ANTHROPIC_API_KEY) ──
+    const hasKey = !!(process.env.ANTHROPIC_API_KEY?.trim());
+    if (!hasKey) {
+        console.log('\n-- Stage 2: llm-judge (SKIPPED — no ANTHROPIC_API_KEY) --');
+    }
+    else {
+        console.log('\n-- Stage 2: llm-judge --');
+        // Case 1: semantically equivalent (should pass)
+        const r6 = await judgeAnswer({ id: 'llm-1', expected: 'Paris' }, 'The capital of France is Paris', baseOpts);
+        check(`llm-judge "Paris" vs "The capital of France is Paris" → pass (path=${r6.scoringPath})`, r6.passed);
+        // Case 2: numerically wrong (should fail)
+        const r7 = await judgeAnswer({ id: 'llm-2', expected: '3.14159' }, 'approximately three', baseOpts);
+        check(`llm-judge "3.14159" vs "approximately three" → fail (path=${r7.scoringPath})`, !r7.passed);
+        const llmCost = (r6.judgeCostUsd ?? 0) + (r7.judgeCostUsd ?? 0);
+        console.log(`  cost: $${llmCost.toFixed(5)} (${(r6.judgeTokensIn ?? 0) + (r7.judgeTokensIn ?? 0)} in, ${(r6.judgeTokensOut ?? 0) + (r7.judgeTokensOut ?? 0)} out)`);
+        // ── Cache hit verification ──
+        console.log('\n-- Stage 3: cache hit --');
+        // Re-run case 1 — must return from cache
+        const r8 = await judgeAnswer({ id: 'llm-1', expected: 'Paris' }, 'The capital of France is Paris', baseOpts);
+        check('second run of llm-1 → cache hit', r8.scoringPath === 'cache');
+        check('cache hit preserves original verdict', r8.passed === r6.passed);
+        // skipCache forces an LLM call even if cached
+        const r9 = await judgeAnswer({ id: 'llm-1', expected: 'Paris' }, 'The capital of France is Paris', { ...baseOpts, skipCache: true });
+        check('skipCache=true bypasses cache → llm-judge', r9.scoringPath === 'llm-judge');
+    }
+    // Cleanup temp cache dir
+    try {
+        fs.rmSync(tmpCacheDir, { recursive: true, force: true });
+    }
+    catch { /* ignore */ }
+    console.log(`\n=== smoke ${failures === 0 ? 'PASSED' : `FAILED (${failures} assertion(s))`} ===\n`);
+    if (failures > 0)
+        process.exit(1);
+}
+// Run smoke when executed directly
+const isMain = process.argv[1] && (process.argv[1].endsWith('gaia-judge.ts') ||
+    process.argv[1].endsWith('gaia-judge.js'));
+if (isMain) {
+    runSmoke().catch((err) => {
+        console.error('Smoke failed:', err);
+        process.exit(1);
+    });
+}
+//# sourceMappingURL=gaia-judge.js.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"gaia-judge.js","sourceRoot":"","sources":["../../../src/benchmarks/gaia-judge.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;GAwBG;AAEH,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AACzC,OAAO,KAAK,EAAE,MAAM,SAAS,CAAC;AAC9B,OAAO,KAAK,IAAI,MAAM,WAAW,CAAC;AAClC,OAAO,KAAK,EAAE,MAAM,SAAS,CAAC;AAC9B,OAAO,EAAE,QAAQ,EAAE,MAAM,oBAAoB,CAAC;AAE9C,8EAA8E;AAC9E,YAAY;AACZ,8EAA8E;AAE9E,MAAM,iBAAiB,GAAG,uCAAuC,CAAC;AAClE,MAAM,qBAAqB,GAAG,YAAY,CAAC;AAC3C,MAAM,mBAAmB,GAAG,mBAAmB,CAAC;AAChD,MAAM,iBAAiB,GAAG,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,OAAO,EAAE,EAAE,QAAQ,EAAE,SAAS,EAAE,MAAM,EAAE,WAAW,CAAC,CAAC;AAE5F;;;GAGG;AACH,MAAM,oBAAoB,GAAG,IAAI,CAAC;AAElC,gEAAgE;AAChE,MAAM,uBAAuB,GAAG,GAAG,CAAC;AACpC,MAAM,wBAAwB,GAAG,IAAI,CAAC;AA4BtC,8EAA8E;AAC9E,gBAAgB;AAChB,8EAA8E;AAE9E;;;;;;GAMG;AACH,MAAM,UAAU,eAAe,CAAC,GAA8B;IAC5D,IAAI,GAAG,IAAI,IAAI;QAAE,OAAO,EAAE,CAAC;IAC3B,IAAI,CAAC,GAAG,GAAG,CAAC,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IACjC,uCAAuC;IACvC,IAAI,CAAC,CAAC,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC;QACtC,CAAC,CAAC,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,EAAE,CAAC;QAC3C,CAAC,GAAG,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC;IACrB,CAAC;IACD,+BAA+B;IAC/B,CAAC,GAAG,CAAC,CAAC,OAAO,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC,IAAI,EAAE,CAAC;IAClC,OAAO,CAAC,CAAC;AACX,CAAC;AAED,8EAA8E;AAC9E,8BAA8B;AAC9B,8EAA8E;AAE9E;;;;;;;;;;;;;;;;;GAiBG;AACH,MAAM,UAAU,oBAAoB,CAClC,SAAiB,EACjB,QAAgB,EAChB,YAAqB;IAErB,uEAAuE;IACvE,MAAM,KAAK,GAAG,CAAC,CAAS,EAAU,EAAE,CAAC,UAAU,CAAC,CAAC,CAAC,OAAO,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC,CAAC;IAEzE,MAAM,OAAO,GAAG,KAAK,CAAC,SAAS,CAAC,CAAC;IACjC,MAAM,MAAM,GAAG,KAAK,CAAC,QAAQ,CAAC,CAAC;IAE/B,IAAI,KAAK,CAAC,OAAO,CAAC,IAAI,KAAK,CAAC,MAAM,CAAC;QAAE,OAAO,KAAK,CAAC;IAElD,wDAAwD;IACxD,IAAI,IAAI,CAAC,GAAG,CAAC,OAAO,GAAG,MAAM,CAAC,GAAG,KAAK,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;QAAE,OAAO,IAAI,CAAC;IAE7E,IAAI,CAAC,YAAY;QAAE,OAAO,KAAK,CAAC;IAEhC,MAAM,MAAM,GAAG,YAAY,CAAC,WAAW,EAAE,CAAC;IAE1C,MAAM,WAAW,GAA4B;QAC3C,CAAC,UAAU,EAAE,IAAI,CAAC;QAClB,CAAC,SAAS,EAAE,GAAG,CAAC;QAChB,CAAC,SAAS,EAAE,GAAG,CAAC;QAChB,CAAC,UAAU,EAAE,GAAG,CAAC;QACjB,CAAC,SAAS,EAAE,GAAG,CAAC;KACjB,CAAC;IAEF,KAAK,MAAM,CAAC,IAAI,EAAE,IAAI,CAAC,IAAI,WAAW,EAAE,CAAC;QACvC,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,IAAI,CAAC;YAAE,SAAS;QAErC,0DAA0D;QAC1D,qFAAqF;QACrF,IAAI,IAAI,CAAC,GAAG,CAAC,OAAO,GAAG,IAAI,GAAG,MAAM,CAAC,GAAG,IAAI,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;YAAE,OAAO,IAAI,CAAC;QAEnF,kDAAkD;QAClD,IAAI,IAAI,CAAC,GAAG,CAAC,OAAO,GAAG,MAAM,GAAG,IAAI,CAAC,GAAG,IAAI,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;YAAE,OAAO,IAAI,CAAC;IACtF,CAAC;IAED,OAAO,KAAK,CAAC;AACf,CAAC;AAED,8EAA8E;AAC9E,gBAAgB;AAChB,8EAA8E;AAE9E,SAAS,QAAQ,CACf,UAAkB,EAClB,eAAuB,EACvB,UAAkB;IAElB,MAAM,GAAG,GAAG,GAAG,UAAU,KAAK,eAAe,KAAK,UAAU,KAAK,oBAAoB,EAAE,CAAC;IACxF,OAAO,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;AACxD,CAAC;AAED,SAAS,SAAS,CAAC,QAAgB,EAAE,GAAW;IAC9C,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,GAAG,GAAG,OAAO,CAAC,CAAC;IAChD,IAAI,CAAC;QACH,MAAM,GAAG,GAAG,EAAE,CAAC,YAAY,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;QAC3C,OAAO,IAAI,CAAC,KAAK,CAAC,GAAG,CAAgB,CAAC;IACxC,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,IAAI,CAAC;IACd,CAAC;AACH,CAAC;AAED,SAAS,UAAU,CAAC,QAAgB,EAAE,GAAW,EAAE,MAAmB;IACpE,IAAI,CAAC;QACH,EAAE,CAAC,SAAS,CAAC,QAAQ,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QAC5C,EAAE,CAAC,aAAa,CACd,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,GAAG,GAAG,OAAO,CAAC,EAClC,IAAI,CAAC,SAAS,CAAC,MAAM,EAAE,IAAI,EAAE,CAAC,CAAC,EAC/B,OAAO,CACR,CAAC;IACJ,CAAC;IAAC,MAAM,CAAC;QACP,oEAAoE;IACtE,CAAC;AACH,CAAC;AAED,8EAA8E;AAC9E,oEAAoE;AACpE,8EAA8E;AAE9E,SAAS,aAAa,CAAC,QAAiB;IACtC,IAAI,QAAQ,IAAI,QAAQ,CAAC,IAAI,EAAE;QAAE,OAAO,QAAQ,CAAC,IAAI,EAAE,CAAC;IAExD,MAAM,MAAM,GAAG,OAAO,CAAC,GAAG,CAAC,iBAAiB,CAAC;IAC7C,IAAI,MAAM,IAAI,MAAM,CAAC,IAAI,EAAE;QAAE,OAAO,MAAM,CAAC,IAAI,EAAE,CAAC;IAElD,IAAI,CAAC;QACH,MAAM,GAAG,GAAG,QAAQ,CAClB,8EAA8E,EAC9E,EAAE,QAAQ,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE,CACvC,CAAC,IAAI,EAAE,CAAC;QACT,IAAI,GAAG;YAAE,OAAO,GAAG,CAAC;IACtB,CAAC;IAAC,MAAM,CAAC;QACP,kBAAkB;IACpB,CAAC;IAED,MAAM,IAAI,KAAK,CACb,wFAAwF;QACxF,sBAAsB,CACvB,CAAC;AACJ,CAAC;AAED,8EAA8E;AAC9E,sBAAsB;AACtB,8EAA8E;AAE9E;;;;;;;;;;;GAWG;AACH,SAAS,sBAAsB;IAC7B,OAAO;QACL,oEAAoE;QACpE,0CAA0C;QAC1C,EAAE;QACF,wDAAwD;QACxD,sEAAsE;QACtE,sFAAsF;QACtF,2FAA2F;QAC3F,+EAA+E;QAC/E,8CAA8C;QAC9C,yFAAyF;QACzF,0FAA0F;QAC1F,qEAAqE;QACrE,EAAE;QACF,6EAA6E;QAC7E,2EAA2E;QAC3E,+CAA+C;QAC/C,iDAAiD;KAClD,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AACf,CAAC;AAED,SAAS,qBAAqB,CAC5B,QAAgB,EAChB,WAAmB,EACnB,SAAiB;IAEjB,OAAO;QACL,aAAa,QAAQ,EAAE;QACvB,iBAAiB,WAAW,EAAE;QAC9B,qBAAqB,SAAS,EAAE;QAChC,EAAE;QACF,8DAA8D;KAC/D,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AACf,CAAC;AAqBD,KAAK,UAAU,SAAS,CACtB,YAAoB,EACpB,WAAmB,EACnB,KAAa,EACb,MAAc;IAEd,MAAM,QAAQ,GAAuB;QACnC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,WAAW,EAAE;KACvC,CAAC;IAEF,MAAM,IAAI,GAAG,IAAI,CAAC,SAAS,CAAC;QAC1B,KAAK;QACL,UAAU,EAAE,GAAG;QACf,MAAM,EAAE,YAAY;QACpB,QAAQ;KACT,CAAC,CAAC;IAEH,MAAM,QAAQ,GAAG,MAAM,KAAK,CAAC,iBAAiB,EAAE;QAC9C,MAAM,EAAE,MAAM;QACd,OAAO,EAAE;YACP,cAAc,EAAE,kBAAkB;YAClC,WAAW,EAAE,MAAM;YACnB,mBAAmB,EAAE,qBAAqB;SAC3C;QACD,IAAI;KACL,CAAC,CAAC;IAEH,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,CAAC;QACjB,MAAM,OAAO,GAAG,MAAM,QAAQ,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,WAAW,CAAC,CAAC;QAC/D,MAAM,IAAI,KAAK,CAAC,uBAAuB,QAAQ,CAAC,MAAM,KAAK,OAAO,EAAE,CAAC,CAAC;IACxE,CAAC;IAED,MAAM,IAAI,GAAG,CAAC,MAAM,QAAQ,CAAC,IAAI,EAAE,CAAsB,CAAC;IAE1D,MAAM,SAAS,GAAG,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,MAAM,CAAC,CAAC;IAC9D,MAAM,OAAO,GAAG,SAAS,EAAE,IAAI,IAAI,EAAE,CAAC;IAEtC,oCAAoC;IACpC,IAAI,MAAM,GAAG,KAAK,CAAC;IACnB,IAAI,MAAM,GAAG,EAAE,CAAC;IAChB,IAAI,CAAC;QACH,2DAA2D;QAC3D,MAAM,OAAO,GAAG,OAAO,CAAC,OAAO,CAAC,mBAAmB,EAAE,EAAE,CAAC,CAAC,OAAO,CAAC,SAAS,EAAE,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;QACvF,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,OAAO,CAA0C,CAAC;QAC5E,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;QAChC,MAAM,GAAG,MAAM,CAAC,MAAM,CAAC,MAAM,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC;IACrD,CAAC;IAAC,MAAM,CAAC;QACP,wDAAwD;QACxD,MAAM,KAAK,GAAG,OAAO,CAAC,WAAW,EAAE,CAAC;QACpC,MAAM,GAAG,KAAK,CAAC,QAAQ,CAAC,gBAAgB,CAAC,IAAI,KAAK,CAAC,QAAQ,CAAC,eAAe,CAAC,CAAC;QAC7E,MAAM,GAAG,sBAAsB,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CAAC;IACzD,CAAC;IAED,OAAO;QACL,MAAM;QACN,MAAM;QACN,QAAQ,EAAE,IAAI,CAAC,KAAK,CAAC,YAAY;QACjC,SAAS,EAAE,IAAI,CAAC,KAAK,CAAC,aAAa;KACpC,CAAC;AACJ,CAAC;AAED,8EAA8E;AAC9E,aAAa;AACb,8EAA8E;AAE9E;;;;;;;;;GASG;AACH,MAAM,CAAC,KAAK,UAAU,WAAW,CAC/B,QAAiE,EACjE,eAA8B,EAC9B,OAAsB;IAEtB,MAAM,UAAU,GAAG,OAAO,EAAE,UAAU,IAAI,mBAAmB,CAAC;IAC9D,MAAM,QAAQ,GAAG,OAAO,EAAE,QAAQ,IAAI,iBAAiB,CAAC;IACxD,MAAM,SAAS,GAAG,eAAe,IAAI,EAAE,CAAC;IAExC,yDAAyD;IACzD,IAAI,CAAC,SAAS,CAAC,IAAI,EAAE,EAAE,CAAC;QACtB,OAAO;YACL,UAAU,EAAE,QAAQ,CAAC,EAAE;YACvB,MAAM,EAAE,KAAK;YACb,WAAW,EAAE,aAAa;YAC1B,eAAe,EAAE,SAAS;YAC1B,WAAW,EAAE,QAAQ,CAAC,QAAQ;SAC/B,CAAC;IACJ,CAAC;IAED,uDAAuD;IACvD,MAAM,aAAa,GAAG,eAAe,CAAC,SAAS,CAAC,CAAC;IACjD,MAAM,YAAY,GAAG,eAAe,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC;IAExD,IAAI,aAAa,KAAK,YAAY,EAAE,CAAC;QACnC,OAAO;YACL,UAAU,EAAE,QAAQ,CAAC,EAAE;YACvB,MAAM,EAAE,IAAI;YACZ,WAAW,EAAE,aAAa;YAC1B,eAAe,EAAE,SAAS;YAC1B,WAAW,EAAE,QAAQ,CAAC,QAAQ;SAC/B,CAAC;IACJ,CAAC;IAED,yDAAyD;IACzD,oEAAoE;IACpE,+CAA+C;IAC/C,IAAI,oBAAoB,CAAC,aAAa,EAAE,YAAY,EAAE,QAAQ,CAAC,YAAY,CAAC,EAAE,CAAC;QAC7E,OAAO;YACL,UAAU,EAAE,QAAQ,CAAC,EAAE;YACvB,MAAM,EAAE,IAAI;YACZ,WAAW,EAAE,aAAa;YAC1B,eAAe,EAAE,SAAS;YAC1B,WAAW,EAAE,QAAQ,CAAC,QAAQ;SAC/B,CAAC;IACJ,CAAC;IAED,8CAA8C;IAC9C,MAAM,GAAG,GAAG,QAAQ,CAAC,QAAQ,CAAC,EAAE,EAAE,SAAS,EAAE,UAAU,CAAC,CAAC;IACzD,IAAI,CAAC,OAAO,EAAE,SAAS,EAAE,CAAC;QACxB,MAAM,MAAM,GAAG,SAAS,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC;QACxC,IAAI,MAAM,KAAK,IAAI,EAAE,CAAC;YACpB,OAAO,EAAE,GAAG,MAAM,EAAE,WAAW,EAAE,OAAO,EAAE,CAAC;QAC7C,CAAC;IACH,CAAC;IAED,8BAA8B;IAC9B,MAAM,MAAM,GAAG,aAAa,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC;IAC9C,MAAM,YAAY,GAAG,sBAAsB,EAAE,CAAC;IAC9C,MAAM,WAAW,GAAG,qBAAqB,CACvC,QAAQ,CAAC,YAAY,IAAI,QAAQ,CAAC,QAAQ,EAC1C,QAAQ,CAAC,QAAQ,EACjB,SAAS,CACV,CAAC;IAEF,MAAM,EAAE,MAAM,EAAE,MAAM,EAAE,QAAQ,EAAE,SAAS,EAAE,GAAG,MAAM,SAAS,CAC7D,YAAY,EACZ,WAAW,EACX,UAAU,EACV,MAAM,CACP,CAAC;IAEF,MAAM,OAAO,GACX,CAAC,QAAQ,GAAG,SAAS,CAAC,GAAG,uBAAuB;QAChD,CAAC,SAAS,GAAG,SAAS,CAAC,GAAG,wBAAwB,CAAC;IAErD,MAAM,MAAM,GAAgB;QAC1B,UAAU,EAAE,QAAQ,CAAC,EAAE;QACvB,MAAM;QACN,WAAW,EAAE,WAAW;QACxB,eAAe,EAAE,SAAS;QAC1B,WAAW,EAAE,QAAQ,CAAC,QAAQ;QAC9B,WAAW,EAAE,MAAM;QACnB,UAAU;QACV,aAAa,EAAE,QAAQ;QACvB,cAAc,EAAE,SAAS;QACzB,YAAY,EAAE,OAAO;KACtB,CAAC;IAEF,yEAAyE;IACzE,UAAU,CAAC,QAAQ,EAAE,GAAG,EAAE,MAAM,CAAC,CAAC;IAElC,OAAO,MAAM,CAAC;AAChB,CAAC;AAED,8EAA8E;AAC9E,eAAe;AACf,8EAA8E;AAE9E;;;;;;;;GAQG;AACH,KAAK,UAAU,QAAQ;IACrB,MAAM,IAAI,GAAG,qBAAqB,CAAC;IACnC,MAAM,IAAI,GAAG,qBAAqB,CAAC;IAEnC,IAAI,QAAQ,GAAG,CAAC,CAAC;IACjB,SAAS,KAAK,CAAC,KAAa,EAAE,SAAkB;QAC9C,IAAI,SAAS,EAAE,CAAC;YACd,OAAO,CAAC,GAAG,CAAC,KAAK,IAAI,KAAK,KAAK,EAAE,CAAC,CAAC;QACrC,CAAC;aAAM,CAAC;YACN,OAAO,CAAC,GAAG,CAAC,KAAK,IAAI,KAAK,KAAK,EAAE,CAAC,CAAC;YACnC,QAAQ,EAAE,CAAC;QACb,CAAC;IACH,CAAC;IAED,kDAAkD;IAClD,MAAM,WAAW,GAAG,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,MAAM,EAAE,EAAE,oBAAoB,IAAI,CAAC,GAAG,EAAE,EAAE,CAAC,CAAC;IAC7E,MAAM,QAAQ,GAAiB,EAAE,QAAQ,EAAE,WAAW,EAAE,CAAC;IAEzD,OAAO,CAAC,GAAG,CAAC,8BAA8B,CAAC,CAAC;IAE5C,2EAA2E;IAC3E,OAAO,CAAC,GAAG,CAAC,iCAAiC,CAAC,CAAC;IAC/C,KAAK,CAAC,4BAA4B,EAAE,eAAe,CAAC,KAAK,CAAC,KAAK,KAAK,CAAC,CAAC;IACtE,KAAK,CAAC,8BAA8B,EAAE,eAAe,CAAC,OAAO,CAAC,KAAK,KAAK,CAAC,CAAC;IAC1E,KAAK,CAAC,oCAAoC,EAAE,eAAe,CAAC,SAAS,CAAC,KAAK,OAAO,CAAC,CAAC;IACpF,KAAK,CAAC,6CAA6C,EAAE,eAAe,CAAC,cAAc,CAAC,KAAK,aAAa,CAAC,CAAC;IACxG,KAAK,CAAC,wBAAwB,EAAE,eAAe,CAAC,IAAI,CAAC,KAAK,EAAE,CAAC,CAAC;IAC9D,KAAK,CAAC,iCAAiC,EAAE,eAAe,CAAC,KAAK,CAAC,KAAK,eAAe,CAAC,KAAK,CAAC,CAAC,CAAC;IAE5F,yEAAyE;IACzE,OAAO,CAAC,GAAG,CAAC,oCAAoC,CAAC,CAAC;IAElD,MAAM,EAAE,GAAG,MAAM,WAAW,CAAC,EAAE,EAAE,EAAE,MAAM,EAAE,QAAQ,EAAE,KAAK,EAAE,EAAE,KAAK,EAAE,QAAQ,CAAC,CAAC;IAC/E,KAAK,CAAC,qDAAqD,EAAE,EAAE,CAAC,MAAM,IAAI,EAAE,CAAC,WAAW,KAAK,aAAa,CAAC,CAAC;IAE5G,MAAM,EAAE,GAAG,MAAM,WAAW,CAAC,EAAE,EAAE,EAAE,MAAM,EAAE,QAAQ,EAAE,KAAK,EAAE,EAAE,OAAO,EAAE,QAAQ,CAAC,CAAC;IACjF,KAAK,CAAC,sDAAsD,EAAE,EAAE,CAAC,MAAM,IAAI,EAAE,CAAC,WAAW,KAAK,aAAa,CAAC,CAAC;IAE7G,MAAM,EAAE,GAAG,MAAM,WAAW,CAAC,EAAE,EAAE,EAAE,MAAM,EAAE,QAAQ,EAAE,OAAO,EAAE,EAAE,SAAS,EAAE,QAAQ,CAAC,CAAC;IACrF,KAAK,CAAC,gEAAgE,EAAE,EAAE,CAAC,MAAM,IAAI,EAAE,CAAC,WAAW,KAAK,aAAa,CAAC,CAAC;IAEvH,MAAM,EAAE,GAAG,MAAM,WAAW,CAAC,EAAE,EAAE,EAAE,MAAM,EAAE,QAAQ,EAAE,aAAa,EAAE,EAAE,cAAc,EAAE,QAAQ,CAAC,CAAC;IAChG,KAAK,CAAC,6DAA6D,EAAE,EAAE,CAAC,MAAM,IAAI,EAAE,CAAC,WAAW,KAAK,aAAa,CAAC,CAAC;IAEpH,MAAM,KAAK,GAAG,MAAM,WAAW,CAAC,EAAE,EAAE,EAAE,SAAS,EAAE,QAAQ,EAAE,KAAK,EAAE,EAAE,IAAI,EAAE,QAAQ,CAAC,CAAC;IACpF,KAAK,CAAC,yCAAyC,EAAE,CAAC,KAAK,CAAC,MAAM,IAAI,KAAK,CAAC,WAAW,KAAK,aAAa,CAAC,CAAC;IAEvG,qDAAqD;IACrD,MAAM,MAAM,GAAG,CAAC,CAAC,CAAC,OAAO,CAAC,GAAG,CAAC,iBAAiB,EAAE,IAAI,EAAE,CAAC,CAAC;IACzD,IAAI,CAAC,MAAM,EAAE,CAAC;QACZ,OAAO,CAAC,GAAG,CAAC,6DAA6D,CAAC,CAAC;IAC7E,CAAC;SAAM,CAAC;QACN,OAAO,CAAC,GAAG,CAAC,4BAA4B,CAAC,CAAC;QAE1C,gDAAgD;QAChD,MAAM,EAAE,GAAG,MAAM,WAAW,CAC1B,EAAE,EAAE,EAAE,OAAO,EAAE,QAAQ,EAAE,OAAO,EAAE,EAClC,gCAAgC,EAChC,QAAQ,CACT,CAAC;QACF,KAAK,CACH,sEAAsE,EAAE,CAAC,WAAW,GAAG,EACvF,EAAE,CAAC,MAAM,CACV,CAAC;QAEF,0CAA0C;QAC1C,MAAM,EAAE,GAAG,MAAM,WAAW,CAC1B,EAAE,EAAE,EAAE,OAAO,EAAE,QAAQ,EAAE,SAAS,EAAE,EACpC,qBAAqB,EACrB,QAAQ,CACT,CAAC;QACF,KAAK,CACH,6DAA6D,EAAE,CAAC,WAAW,GAAG,EAC9E,CAAC,EAAE,CAAC,MAAM,CACX,CAAC;QAEF,MAAM,OAAO,GAAG,CAAC,EAAE,CAAC,YAAY,IAAI,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,YAAY,IAAI,CAAC,CAAC,CAAC;QAChE,OAAO,CAAC,GAAG,CAAC,YAAY,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK,CAAC,EAAE,CAAC,aAAa,IAAI,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,aAAa,IAAI,CAAC,CAAC,QAAQ,CAAC,EAAE,CAAC,cAAc,IAAI,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,cAAc,IAAI,CAAC,CAAC,OAAO,CAAC,CAAC;QAEpK,+BAA+B;QAC/B,OAAO,CAAC,GAAG,CAAC,4BAA4B,CAAC,CAAC;QAE1C,yCAAyC;QACzC,MAAM,EAAE,GAAG,MAAM,WAAW,CAC1B,EAAE,EAAE,EAAE,OAAO,EAAE,QAAQ,EAAE,OAAO,EAAE,EAClC,gCAAgC,EAChC,QAAQ,CACT,CAAC;QACF,KAAK,CAAC,iCAAiC,EAAE,EAAE,CAAC,WAAW,KAAK,OAAO,CAAC,CAAC;QACrE,KAAK,CAAC,sCAAsC,EAAE,EAAE,CAAC,MAAM,KAAK,EAAE,CAAC,MAAM,CAAC,CAAC;QAEvE,8CAA8C;QAC9C,MAAM,EAAE,GAAG,MAAM,WAAW,CAC1B,EAAE,EAAE,EAAE,OAAO,EAAE,QAAQ,EAAE,OAAO,EAAE,EAClC,gCAAgC,EAChC,EAAE,GAAG,QAAQ,EAAE,SAAS,EAAE,IAAI,EAAE,CACjC,CAAC;QACF,KAAK,CAAC,2CAA2C,EAAE,EAAE,CAAC,WAAW,KAAK,WAAW,CAAC,CAAC;IACrF,CAAC;IAED,yBAAyB;IACzB,IAAI,CAAC;QAAC,EAAE,CAAC,MAAM,CAAC,WAAW,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC;IAAC,CAAC;IAAC,MAAM,CAAC,CAAC,YAAY,CAAC,CAAC;IAExF,OAAO,CAAC,GAAG,CAAC,eAAe,QAAQ,KAAK,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,WAAW,QAAQ,gBAAgB,QAAQ,CAAC,CAAC;IACpG,IAAI,QAAQ,GAAG,CAAC;QAAE,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;AACpC,CAAC;AAED,mCAAmC;AACnC,MAAM,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAChC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,eAAe,CAAC;IACzC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,eAAe,CAAC,CAC1C,CAAC;AACF,IAAI,MAAM,EAAE,CAAC;IACX,QAAQ,EAAE,CAAC,KAAK,CAAC,CAAC,GAAG,EAAE,EAAE;QACvB,OAAO,CAAC,KAAK,CAAC,eAAe,EAAE,GAAG,CAAC,CAAC;QACpC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAClB,CAAC,CAAC,CAAC;AACL,CAAC"}

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

@@ -0,0 +1,87 @@
+/**
+ * GAIA Dataset Loader — ADR-133-PR1
+ *
+ * Authenticates to Hugging Face, downloads the `gaia-benchmark/GAIA`
+ * validation split, caches it under ~/.cache/swarmdo/gaia/, and exposes
+ * a typed `loadGaia()` API consumed by the capability-gaia subcommand.
+ *
+ * Token resolution order (mirrors performance-capability.ts ANTHROPIC_API_KEY pattern):
+ *   1. $HF_TOKEN env var
+ *   2. gcloud secrets versions access latest --secret=huggingface-token
+ *   3. Fail with a clear error message
+ *
+ * This file is deliberately a skeleton / PR-1 checkpoint.  The full
+ * dataset download is gated behind a real HF_TOKEN; a 5-question smoke
+ * fixture is provided for offline / CI-without-HF testing.
+ *
+ * Refs: ADR-133, #2156
+ */
+export type GaiaLevel = 1 | 2 | 3;
+export interface GaiaQuestion {
+    /** Unique identifier from the HF dataset. */
+    task_id: string;
+    /** Difficulty level: 1 (easiest) → 3 (hardest). */
+    level: GaiaLevel;
+    /** The question text sent to the agent. */
+    question: string;
+    /** Ground-truth final answer (string normalised, no surrounding whitespace). */
+    final_answer: string;
+    /** Optional file attachment filename; resolved to an absolute path by the loader. */
+    file_name: string | null;
+    /** Absolute path to the cached attachment, or null if no attachment. */
+    file_path: string | null;
+    /** Steps annotation (meta-data, not used by agent). */
+    annotator_metadata?: Record<string, unknown>;
+}
+export interface LoadGaiaOptions {
+    /** Level filter (default: 1). */
+    level?: GaiaLevel;
+    /** Maximum questions to return (default: all). */
+    limit?: number;
+    /** Skip HF download; use the built-in 5-question smoke fixture instead. */
+    smokeOnly?: boolean;
+    /** Override the cache directory (default: ~/.cache/swarmdo/gaia). */
+    cacheDir?: string;
+}
+/**
+ * Resolve a Hugging Face API token using two fallbacks:
+ *   1. $HF_TOKEN env var
+ *   2. `gcloud secrets versions access latest --secret=huggingface-token`
+ *
+ * Throws with a clear error if neither is available.
+ */
+export declare function resolveHfToken(): string;
+/**
+ * Export the default cache directory path (no side effects).
+ */
+export declare function getDefaultCacheDir(): string;
+/**
+ * Minimal offline fixture for smoke tests and CI environments without HF_TOKEN.
+ * Answers are verified by hand against publicly known facts.
+ * All questions are Level 1 (no file attachments, no multi-hop tool use required).
+ *
+ * IMPORTANT: Verify every answer key with `node -e 'console.log(…)'` before
+ * adding entries to this list — three answer-key bugs were caught in session #2156.
+ */
+export declare const SMOKE_FIXTURE: GaiaQuestion[];
+/**
+ * Download all attachment files referenced by the questions list.
+ * Mutates each question's `file_path` field in place.
+ * Skips questions without a file_name.
+ */
+export declare function resolveAttachments(questions: GaiaQuestion[], token: string, cacheDir: string): Promise<void>;
+/**
+ * Load GAIA validation questions.
+ *
+ * - With `smokeOnly: true` (or when HF_TOKEN is unavailable): returns the 5-question
+ *   smoke fixture — no network call, no token required.
+ * - Otherwise: authenticates to HF, downloads level N questions, caches locally.
+ *
+ * @throws if HF_TOKEN is missing and smokeOnly is false
+ */
+export declare function loadGaia(options?: LoadGaiaOptions): Promise<GaiaQuestion[]>;
+/**
+ * Returns the cache directory path (does not create it).
+ */
+export declare function getGaiaCacheDir(override?: string): string;
+//# sourceMappingURL=gaia-loader.d.ts.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"gaia-loader.d.ts","sourceRoot":"","sources":["../../../src/benchmarks/gaia-loader.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;GAiBG;AAaH,MAAM,MAAM,SAAS,GAAG,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;AAElC,MAAM,WAAW,YAAY;IAC3B,6CAA6C;IAC7C,OAAO,EAAE,MAAM,CAAC;IAChB,mDAAmD;IACnD,KAAK,EAAE,SAAS,CAAC;IACjB,2CAA2C;IAC3C,QAAQ,EAAE,MAAM,CAAC;IACjB,gFAAgF;IAChF,YAAY,EAAE,MAAM,CAAC;IACrB,qFAAqF;IACrF,SAAS,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,wEAAwE;IACxE,SAAS,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,uDAAuD;IACvD,kBAAkB,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CAC9C;AAED,MAAM,WAAW,eAAe;IAC9B,iCAAiC;IACjC,KAAK,CAAC,EAAE,SAAS,CAAC;IAClB,kDAAkD;IAClD,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,2EAA2E;IAC3E,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,qEAAqE;IACrE,QAAQ,CAAC,EAAE,MAAM,CAAC;CACnB;AAMD;;;;;;GAMG;AACH,wBAAgB,cAAc,IAAI,MAAM,CAkBvC;AAUD;;GAEG;AACH,wBAAgB,kBAAkB,IAAI,MAAM,CAE3C;AAcD;;;;;;;GAOG;AACH,eAAO,MAAM,aAAa,EAAE,YAAY,EAyCvC,CAAC;AA8LF;;;;GAIG;AACH,wBAAsB,kBAAkB,CACtC,SAAS,EAAE,YAAY,EAAE,EACzB,KAAK,EAAE,MAAM,EACb,QAAQ,EAAE,MAAM,GACf,OAAO,CAAC,IAAI,CAAC,CAQf;AAMD;;;;;;;;GAQG;AACH,wBAAsB,QAAQ,CAAC,OAAO,GAAE,eAAoB,GAAG,OAAO,CAAC,YAAY,EAAE,CAAC,CAarF;AAED;;GAEG;AACH,wBAAgB,eAAe,CAAC,QAAQ,CAAC,EAAE,MAAM,GAAG,MAAM,CAEzD"}