truthguard-ai 0.1.0

package/dist/Alerting/index.js

@@ -0,0 +1,76 @@
+"use strict";
+/**
+ * Alert Dispatchers — notify when grounding scores drop.
+ *
+ * Used by warn and block modes to send notifications
+ * through various channels (webhook, console, custom).
+ */
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.CallbackAlertDispatcher = exports.WebhookAlertDispatcher = exports.ConsoleAlertDispatcher = void 0;
+// ---------------------------------------------------------------------------
+// ConsoleAlertDispatcher
+// ---------------------------------------------------------------------------
+/** Logs alerts to stderr with color. */
+class ConsoleAlertDispatcher {
+    async dispatch(report, _trace) {
+        const score = (report.groundingScore * 100).toFixed(0);
+        const failures = report.detectedFailures
+            .map((f) => f.type)
+            .join(', ');
+        console.error(`[TruthGuard WARN] Trace ${report.traceId} scored ${score}% | ` +
+            `${report.claimsFailed} ungrounded claims | ` +
+            `Failures: ${failures || 'none'}`);
+    }
+}
+exports.ConsoleAlertDispatcher = ConsoleAlertDispatcher;
+/** Posts alert payloads to a webhook URL (Slack, Discord, PagerDuty, custom). */
+class WebhookAlertDispatcher {
+    constructor(config) {
+        this.config = config;
+    }
+    async dispatch(report, trace) {
+        const payload = {
+            event: 'truthguard.grounding.alert',
+            timestamp: new Date().toISOString(),
+            traceId: report.traceId,
+            groundingScore: report.groundingScore,
+            claimsTotal: report.claimsTotal,
+            claimsFailed: report.claimsFailed,
+            detectedFailures: report.detectedFailures.map((f) => ({
+                type: f.type,
+                confidence: f.confidence,
+                description: f.description,
+            })),
+            trace: {
+                traceId: trace.traceId,
+                stepsCount: trace.steps.length,
+            },
+        };
+        const res = await fetch(this.config.url, {
+            method: 'POST',
+            headers: {
+                'Content-Type': 'application/json',
+                ...this.config.headers,
+            },
+            body: JSON.stringify(payload),
+        });
+        if (!res.ok) {
+            console.error(`[TruthGuard] Webhook alert failed: ${res.status} ${res.statusText}`);
+        }
+    }
+}
+exports.WebhookAlertDispatcher = WebhookAlertDispatcher;
+// ---------------------------------------------------------------------------
+// CallbackAlertDispatcher — for custom integrations
+// ---------------------------------------------------------------------------
+/** Calls a user-provided function. */
+class CallbackAlertDispatcher {
+    constructor(fn) {
+        this.fn = fn;
+    }
+    async dispatch(report, trace) {
+        await this.fn(report, trace);
+    }
+}
+exports.CallbackAlertDispatcher = CallbackAlertDispatcher;
+//# sourceMappingURL=index.js.map

package/dist/Alerting/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../src/Alerting/index.ts"],"names":[],"mappings":";AAAA;;;;;GAKG;;;AAcH,8EAA8E;AAC9E,yBAAyB;AACzB,8EAA8E;AAE9E,wCAAwC;AACxC,MAAa,sBAAsB;IACjC,KAAK,CAAC,QAAQ,CAAC,MAAuB,EAAE,MAAa;QACnD,MAAM,KAAK,GAAG,CAAC,MAAM,CAAC,cAAc,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;QACvD,MAAM,QAAQ,GAAG,MAAM,CAAC,gBAAgB;aACrC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC;aAClB,IAAI,CAAC,IAAI,CAAC,CAAC;QACd,OAAO,CAAC,KAAK,CACX,2BAA2B,MAAM,CAAC,OAAO,WAAW,KAAK,MAAM;YAC7D,GAAG,MAAM,CAAC,YAAY,uBAAuB;YAC7C,aAAa,QAAQ,IAAI,MAAM,EAAE,CACpC,CAAC;IACJ,CAAC;CACF;AAZD,wDAYC;AAaD,iFAAiF;AACjF,MAAa,sBAAsB;IAGjC,YAAY,MAAqB;QAC/B,IAAI,CAAC,MAAM,GAAG,MAAM,CAAC;IACvB,CAAC;IAED,KAAK,CAAC,QAAQ,CAAC,MAAuB,EAAE,KAAY;QAClD,MAAM,OAAO,GAAG;YACd,KAAK,EAAE,4BAA4B;YACnC,SAAS,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;YACnC,OAAO,EAAE,MAAM,CAAC,OAAO;YACvB,cAAc,EAAE,MAAM,CAAC,cAAc;YACrC,WAAW,EAAE,MAAM,CAAC,WAAW;YAC/B,YAAY,EAAE,MAAM,CAAC,YAAY;YACjC,gBAAgB,EAAE,MAAM,CAAC,gBAAgB,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;gBACpD,IAAI,EAAE,CAAC,CAAC,IAAI;gBACZ,UAAU,EAAE,CAAC,CAAC,UAAU;gBACxB,WAAW,EAAE,CAAC,CAAC,WAAW;aAC3B,CAAC,CAAC;YACH,KAAK,EAAE;gBACL,OAAO,EAAE,KAAK,CAAC,OAAO;gBACtB,UAAU,EAAE,KAAK,CAAC,KAAK,CAAC,MAAM;aAC/B;SACF,CAAC;QAEF,MAAM,GAAG,GAAG,MAAM,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,GAAG,EAAE;YACvC,MAAM,EAAE,MAAM;YACd,OAAO,EAAE;gBACP,cAAc,EAAE,kBAAkB;gBAClC,GAAG,IAAI,CAAC,MAAM,CAAC,OAAO;aACvB;YACD,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC;SAC9B,CAAC,CAAC;QAEH,IAAI,CAAC,GAAG,CAAC,EAAE,EAAE,CAAC;YACZ,OAAO,CAAC,KAAK,CACX,sCAAsC,GAAG,CAAC,MAAM,IAAI,GAAG,CAAC,UAAU,EAAE,CACrE,CAAC;QACJ,CAAC;IACH,CAAC;CACF;AAzCD,wDAyCC;AAED,8EAA8E;AAC9E,oDAAoD;AACpD,8EAA8E;AAE9E,sCAAsC;AACtC,MAAa,uBAAuB;IAMlC,YACE,EAAmE;QAEnE,IAAI,CAAC,EAAE,GAAG,EAAE,CAAC;IACf,CAAC;IAED,KAAK,CAAC,QAAQ,CAAC,MAAuB,EAAE,KAAY;QAClD,MAAM,IAAI,CAAC,EAAE,CAAC,MAAM,EAAE,KAAK,CAAC,CAAC;IAC/B,CAAC;CACF;AAfD,0DAeC"}

package/dist/Auth/index.d.ts

@@ -0,0 +1,82 @@
+/**
+ * Auth — API Key management and local credential storage.
+ *
+ * Stores credentials in ~/.truthguard/config.json
+ * Provides login flow, key validation, and session management.
+ */
+/** Stored authentication configuration. */
+export interface AuthConfig {
+    /** API key for the TruthGuard cloud service. */
+    apiKey: string;
+    /** API base URL. */
+    apiUrl: string;
+    /** Whether telemetry is enabled (opt-in). */
+    telemetryEnabled: boolean;
+    /** ISO-8601 timestamp when the config was created. */
+    createdAt: string;
+    /** ISO-8601 timestamp when last authenticated. */
+    lastAuthAt?: string;
+}
+/** Response from the /auth/validate endpoint. */
+export interface AuthValidation {
+    valid: boolean;
+    plan?: 'free' | 'pro' | 'enterprise';
+    email?: string;
+    teamId?: string;
+    expiresAt?: string;
+}
+/** Benchmark data returned after login. */
+export interface BenchmarkData {
+    /** User's average score (from telemetry). */
+    userAvgScore?: number;
+    /** Global average score across all users. */
+    globalAvgScore: number;
+    /** Percentile rank (0–100). */
+    percentile?: number;
+    /** Top failure types seen globally. */
+    topFailures: string[];
+    /** Personalized recommendations. */
+    recommendations: string[];
+}
+/** Load the stored auth config, or null if not logged in. */
+export declare function loadAuthConfig(): AuthConfig | null;
+/** Save auth config to disk. */
+export declare function saveAuthConfig(config: AuthConfig): void;
+/** Remove stored credentials (logout). */
+export declare function clearAuthConfig(): boolean;
+/** Get the API key from config or TRUTHGUARD_API_KEY env var. */
+export declare function getApiKey(): string | null;
+/** Get the API URL from config or TRUTHGUARD_API_URL env var. */
+export declare function getApiUrl(): string;
+/** Check if telemetry is enabled. */
+export declare function isTelemetryEnabled(): boolean;
+/**
+ * Validate an API key against the cloud service.
+ * Returns validation result or null if the service is unreachable.
+ */
+export declare function validateApiKey(apiKey: string, apiUrl?: string): Promise<AuthValidation | null>;
+/**
+ * Fetch benchmark data after login.
+ * Returns benchmark or null if unavailable.
+ */
+export declare function fetchBenchmark(apiKey: string, apiUrl?: string): Promise<BenchmarkData | null>;
+/**
+ * Perform the login flow:
+ * 1. Save API key
+ * 2. Validate with cloud (if reachable)
+ * 3. Fetch benchmark (value exchange)
+ *
+ * Returns validation + benchmark, or null if completely offline.
+ */
+export declare function login(apiKey: string, options?: {
+    apiUrl?: string;
+    telemetry?: boolean;
+}): Promise<{
+    validation: AuthValidation | null;
+    benchmark: BenchmarkData | null;
+}>;
+/** Logout — remove stored credentials. */
+export declare function logout(): boolean;
+export declare const AUTH_CONFIG_DIR: string;
+export declare const AUTH_CONFIG_PATH: string;
+//# sourceMappingURL=index.d.ts.map

package/dist/Auth/index.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/Auth/index.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAoBH,2CAA2C;AAC3C,MAAM,WAAW,UAAU;IACzB,gDAAgD;IAChD,MAAM,EAAE,MAAM,CAAC;IACf,oBAAoB;IACpB,MAAM,EAAE,MAAM,CAAC;IACf,6CAA6C;IAC7C,gBAAgB,EAAE,OAAO,CAAC;IAC1B,sDAAsD;IACtD,SAAS,EAAE,MAAM,CAAC;IAClB,kDAAkD;IAClD,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB;AAED,iDAAiD;AACjD,MAAM,WAAW,cAAc;IAC7B,KAAK,EAAE,OAAO,CAAC;IACf,IAAI,CAAC,EAAE,MAAM,GAAG,KAAK,GAAG,YAAY,CAAC;IACrC,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB;AAED,2CAA2C;AAC3C,MAAM,WAAW,aAAa;IAC5B,6CAA6C;IAC7C,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,6CAA6C;IAC7C,cAAc,EAAE,MAAM,CAAC;IACvB,+BAA+B;IAC/B,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,uCAAuC;IACvC,WAAW,EAAE,MAAM,EAAE,CAAC;IACtB,oCAAoC;IACpC,eAAe,EAAE,MAAM,EAAE,CAAC;CAC3B;AAYD,6DAA6D;AAC7D,wBAAgB,cAAc,IAAI,UAAU,GAAG,IAAI,CAWlD;AAED,gCAAgC;AAChC,wBAAgB,cAAc,CAAC,MAAM,EAAE,UAAU,GAAG,IAAI,CAMvD;AAED,0CAA0C;AAC1C,wBAAgB,eAAe,IAAI,OAAO,CAMzC;AAED,iEAAiE;AACjE,wBAAgB,SAAS,IAAI,MAAM,GAAG,IAAI,CAMzC;AAED,iEAAiE;AACjE,wBAAgB,SAAS,IAAI,MAAM,CAMlC;AAED,qCAAqC;AACrC,wBAAgB,kBAAkB,IAAI,OAAO,CAO5C;AA0DD;;;GAGG;AACH,wBAAsB,cAAc,CAAC,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC,EAAE,MAAM,GAAG,OAAO,CAAC,cAAc,GAAG,IAAI,CAAC,CAkBpG;AAED;;;GAGG;AACH,wBAAsB,cAAc,CAAC,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC,EAAE,MAAM,GAAG,OAAO,CAAC,aAAa,GAAG,IAAI,CAAC,CAenG;AAED;;;;;;;GAOG;AACH,wBAAsB,KAAK,CACzB,MAAM,EAAE,MAAM,EACd,OAAO,CAAC,EAAE;IAAE,MAAM,CAAC,EAAE,MAAM,CAAC;IAAC,SAAS,CAAC,EAAE,OAAO,CAAA;CAAE,GACjD,OAAO,CAAC;IAAE,UAAU,EAAE,cAAc,GAAG,IAAI,CAAC;IAAC,SAAS,EAAE,aAAa,GAAG,IAAI,CAAA;CAAE,CAAC,CAoBjF;AAED,0CAA0C;AAC1C,wBAAgB,MAAM,IAAI,OAAO,CAEhC;AAMD,eAAO,MAAM,eAAe,QAAa,CAAC;AAC1C,eAAO,MAAM,gBAAgB,QAAc,CAAC"}

package/dist/Auth/index.js

@@ -0,0 +1,242 @@
+"use strict";
+/**
+ * Auth — API Key management and local credential storage.
+ *
+ * Stores credentials in ~/.truthguard/config.json
+ * Provides login flow, key validation, and session management.
+ */
+var __createBinding = (this && this.__createBinding) || (Object.create ? (function(o, m, k, k2) {
+    if (k2 === undefined) k2 = k;
+    var desc = Object.getOwnPropertyDescriptor(m, k);
+    if (!desc || ("get" in desc ? !m.__esModule : desc.writable || desc.configurable)) {
+      desc = { enumerable: true, get: function() { return m[k]; } };
+    }
+    Object.defineProperty(o, k2, desc);
+}) : (function(o, m, k, k2) {
+    if (k2 === undefined) k2 = k;
+    o[k2] = m[k];
+}));
+var __setModuleDefault = (this && this.__setModuleDefault) || (Object.create ? (function(o, v) {
+    Object.defineProperty(o, "default", { enumerable: true, value: v });
+}) : function(o, v) {
+    o["default"] = v;
+});
+var __importStar = (this && this.__importStar) || (function () {
+    var ownKeys = function(o) {
+        ownKeys = Object.getOwnPropertyNames || function (o) {
+            var ar = [];
+            for (var k in o) if (Object.prototype.hasOwnProperty.call(o, k)) ar[ar.length] = k;
+            return ar;
+        };
+        return ownKeys(o);
+    };
+    return function (mod) {
+        if (mod && mod.__esModule) return mod;
+        var result = {};
+        if (mod != null) for (var k = ownKeys(mod), i = 0; i < k.length; i++) if (k[i] !== "default") __createBinding(result, mod, k[i]);
+        __setModuleDefault(result, mod);
+        return result;
+    };
+})();
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.AUTH_CONFIG_PATH = exports.AUTH_CONFIG_DIR = void 0;
+exports.loadAuthConfig = loadAuthConfig;
+exports.saveAuthConfig = saveAuthConfig;
+exports.clearAuthConfig = clearAuthConfig;
+exports.getApiKey = getApiKey;
+exports.getApiUrl = getApiUrl;
+exports.isTelemetryEnabled = isTelemetryEnabled;
+exports.validateApiKey = validateApiKey;
+exports.fetchBenchmark = fetchBenchmark;
+exports.login = login;
+exports.logout = logout;
+const fs = __importStar(require("fs"));
+const path = __importStar(require("path"));
+const os = __importStar(require("os"));
+const https = __importStar(require("https"));
+const http = __importStar(require("http"));
+// ---------------------------------------------------------------------------
+// Constants
+// ---------------------------------------------------------------------------
+const CONFIG_DIR = path.join(os.homedir(), '.truthguard');
+const CONFIG_PATH = path.join(CONFIG_DIR, 'config.json');
+const DEFAULT_API_URL = 'https://api.truthguard.dev';
+// ---------------------------------------------------------------------------
+// Config persistence
+// ---------------------------------------------------------------------------
+function ensureConfigDir() {
+    if (!fs.existsSync(CONFIG_DIR)) {
+        fs.mkdirSync(CONFIG_DIR, { recursive: true, mode: 0o700 });
+    }
+}
+/** Load the stored auth config, or null if not logged in. */
+function loadAuthConfig() {
+    if (!fs.existsSync(CONFIG_PATH))
+        return null;
+    try {
+        const raw = fs.readFileSync(CONFIG_PATH, 'utf-8');
+        const parsed = JSON.parse(raw);
+        if (!parsed.apiKey)
+            return null;
+        return parsed;
+    }
+    catch {
+        return null;
+    }
+}
+/** Save auth config to disk. */
+function saveAuthConfig(config) {
+    ensureConfigDir();
+    fs.writeFileSync(CONFIG_PATH, JSON.stringify(config, null, 2), {
+        encoding: 'utf-8',
+        mode: 0o600, // owner read/write only
+    });
+}
+/** Remove stored credentials (logout). */
+function clearAuthConfig() {
+    if (fs.existsSync(CONFIG_PATH)) {
+        fs.unlinkSync(CONFIG_PATH);
+        return true;
+    }
+    return false;
+}
+/** Get the API key from config or TRUTHGUARD_API_KEY env var. */
+function getApiKey() {
+    const envKey = process.env['TRUTHGUARD_API_KEY'];
+    if (envKey)
+        return envKey;
+    const config = loadAuthConfig();
+    return config?.apiKey ?? null;
+}
+/** Get the API URL from config or TRUTHGUARD_API_URL env var. */
+function getApiUrl() {
+    const envUrl = process.env['TRUTHGUARD_API_URL'];
+    if (envUrl)
+        return envUrl;
+    const config = loadAuthConfig();
+    return config?.apiUrl ?? DEFAULT_API_URL;
+}
+/** Check if telemetry is enabled. */
+function isTelemetryEnabled() {
+    const envFlag = process.env['TRUTHGUARD_TELEMETRY'];
+    if (envFlag === '0' || envFlag === 'false')
+        return false;
+    if (envFlag === '1' || envFlag === 'true')
+        return true;
+    const config = loadAuthConfig();
+    return config?.telemetryEnabled ?? false;
+}
+function request(url, options) {
+    return new Promise((resolve, reject) => {
+        const parsed = new URL(url);
+        const transport = parsed.protocol === 'https:' ? https : http;
+        const req = transport.request({
+            hostname: parsed.hostname,
+            port: parsed.port || (parsed.protocol === 'https:' ? 443 : 80),
+            path: parsed.pathname + parsed.search,
+            method: options.method,
+            headers: {
+                'Content-Type': 'application/json',
+                'User-Agent': 'truthguard-cli/0.1.0',
+                ...options.headers,
+            },
+            timeout: 10000,
+        }, (res) => {
+            let data = '';
+            res.on('data', (chunk) => (data += chunk.toString()));
+            res.on('end', () => {
+                resolve({ statusCode: res.statusCode ?? 0, body: data });
+            });
+        });
+        req.on('error', reject);
+        req.on('timeout', () => {
+            req.destroy();
+            reject(new Error('Request timeout'));
+        });
+        if (options.body) {
+            req.write(options.body);
+        }
+        req.end();
+    });
+}
+// ---------------------------------------------------------------------------
+// Auth API calls
+// ---------------------------------------------------------------------------
+/**
+ * Validate an API key against the cloud service.
+ * Returns validation result or null if the service is unreachable.
+ */
+async function validateApiKey(apiKey, apiUrl) {
+    const baseUrl = apiUrl ?? getApiUrl();
+    try {
+        const resp = await request(`${baseUrl}/v1/auth/validate`, {
+            method: 'POST',
+            headers: { Authorization: `Bearer ${apiKey}` },
+            body: JSON.stringify({ key: apiKey }),
+        });
+        if (resp.statusCode === 200) {
+            return JSON.parse(resp.body);
+        }
+        return { valid: false };
+    }
+    catch {
+        // Service unreachable — allow offline usage
+        return null;
+    }
+}
+/**
+ * Fetch benchmark data after login.
+ * Returns benchmark or null if unavailable.
+ */
+async function fetchBenchmark(apiKey, apiUrl) {
+    const baseUrl = apiUrl ?? getApiUrl();
+    try {
+        const resp = await request(`${baseUrl}/v1/benchmark`, {
+            method: 'GET',
+            headers: { Authorization: `Bearer ${apiKey}` },
+        });
+        if (resp.statusCode === 200) {
+            return JSON.parse(resp.body);
+        }
+        return null;
+    }
+    catch {
+        return null;
+    }
+}
+/**
+ * Perform the login flow:
+ * 1. Save API key
+ * 2. Validate with cloud (if reachable)
+ * 3. Fetch benchmark (value exchange)
+ *
+ * Returns validation + benchmark, or null if completely offline.
+ */
+async function login(apiKey, options) {
+    const apiUrl = options?.apiUrl ?? DEFAULT_API_URL;
+    const telemetryEnabled = options?.telemetry ?? true;
+    // Save config immediately (works offline)
+    saveAuthConfig({
+        apiKey,
+        apiUrl,
+        telemetryEnabled,
+        createdAt: new Date().toISOString(),
+        lastAuthAt: new Date().toISOString(),
+    });
+    // Validate + fetch benchmark in parallel
+    const [validation, benchmark] = await Promise.all([
+        validateApiKey(apiKey, apiUrl),
+        fetchBenchmark(apiKey, apiUrl),
+    ]);
+    return { validation, benchmark };
+}
+/** Logout — remove stored credentials. */
+function logout() {
+    return clearAuthConfig();
+}
+// ---------------------------------------------------------------------------
+// Config directory path (for tests)
+// ---------------------------------------------------------------------------
+exports.AUTH_CONFIG_DIR = CONFIG_DIR;
+exports.AUTH_CONFIG_PATH = CONFIG_PATH;
+//# sourceMappingURL=index.js.map

package/dist/Auth/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../src/Auth/index.ts"],"names":[],"mappings":";AAAA;;;;;GAKG;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAoEH,wCAWC;AAGD,wCAMC;AAGD,0CAMC;AAGD,8BAMC;AAGD,8BAMC;AAGD,gDAOC;AA8DD,wCAkBC;AAMD,wCAeC;AAUD,sBAuBC;AAGD,wBAEC;AAtQD,uCAAyB;AACzB,2CAA6B;AAC7B,uCAAyB;AACzB,6CAA+B;AAC/B,2CAA6B;AAE7B,8EAA8E;AAC9E,YAAY;AACZ,8EAA8E;AAE9E,MAAM,UAAU,GAAG,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,OAAO,EAAE,EAAE,aAAa,CAAC,CAAC;AAC1D,MAAM,WAAW,GAAG,IAAI,CAAC,IAAI,CAAC,UAAU,EAAE,aAAa,CAAC,CAAC;AACzD,MAAM,eAAe,GAAG,4BAA4B,CAAC;AA2CrD,8EAA8E;AAC9E,qBAAqB;AACrB,8EAA8E;AAE9E,SAAS,eAAe;IACtB,IAAI,CAAC,EAAE,CAAC,UAAU,CAAC,UAAU,CAAC,EAAE,CAAC;QAC/B,EAAE,CAAC,SAAS,CAAC,UAAU,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,IAAI,EAAE,KAAK,EAAE,CAAC,CAAC;IAC7D,CAAC;AACH,CAAC;AAED,6DAA6D;AAC7D,SAAgB,cAAc;IAC5B,IAAI,CAAC,EAAE,CAAC,UAAU,CAAC,WAAW,CAAC;QAAE,OAAO,IAAI,CAAC;IAE7C,IAAI,CAAC;QACH,MAAM,GAAG,GAAG,EAAE,CAAC,YAAY,CAAC,WAAW,EAAE,OAAO,CAAC,CAAC;QAClD,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAe,CAAC;QAC7C,IAAI,CAAC,MAAM,CAAC,MAAM;YAAE,OAAO,IAAI,CAAC;QAChC,OAAO,MAAM,CAAC;IAChB,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,IAAI,CAAC;IACd,CAAC;AACH,CAAC;AAED,gCAAgC;AAChC,SAAgB,cAAc,CAAC,MAAkB;IAC/C,eAAe,EAAE,CAAC;IAClB,EAAE,CAAC,aAAa,CAAC,WAAW,EAAE,IAAI,CAAC,SAAS,CAAC,MAAM,EAAE,IAAI,EAAE,CAAC,CAAC,EAAE;QAC7D,QAAQ,EAAE,OAAO;QACjB,IAAI,EAAE,KAAK,EAAE,wBAAwB;KACtC,CAAC,CAAC;AACL,CAAC;AAED,0CAA0C;AAC1C,SAAgB,eAAe;IAC7B,IAAI,EAAE,CAAC,UAAU,CAAC,WAAW,CAAC,EAAE,CAAC;QAC/B,EAAE,CAAC,UAAU,CAAC,WAAW,CAAC,CAAC;QAC3B,OAAO,IAAI,CAAC;IACd,CAAC;IACD,OAAO,KAAK,CAAC;AACf,CAAC;AAED,iEAAiE;AACjE,SAAgB,SAAS;IACvB,MAAM,MAAM,GAAG,OAAO,CAAC,GAAG,CAAC,oBAAoB,CAAC,CAAC;IACjD,IAAI,MAAM;QAAE,OAAO,MAAM,CAAC;IAE1B,MAAM,MAAM,GAAG,cAAc,EAAE,CAAC;IAChC,OAAO,MAAM,EAAE,MAAM,IAAI,IAAI,CAAC;AAChC,CAAC;AAED,iEAAiE;AACjE,SAAgB,SAAS;IACvB,MAAM,MAAM,GAAG,OAAO,CAAC,GAAG,CAAC,oBAAoB,CAAC,CAAC;IACjD,IAAI,MAAM;QAAE,OAAO,MAAM,CAAC;IAE1B,MAAM,MAAM,GAAG,cAAc,EAAE,CAAC;IAChC,OAAO,MAAM,EAAE,MAAM,IAAI,eAAe,CAAC;AAC3C,CAAC;AAED,qCAAqC;AACrC,SAAgB,kBAAkB;IAChC,MAAM,OAAO,GAAG,OAAO,CAAC,GAAG,CAAC,sBAAsB,CAAC,CAAC;IACpD,IAAI,OAAO,KAAK,GAAG,IAAI,OAAO,KAAK,OAAO;QAAE,OAAO,KAAK,CAAC;IACzD,IAAI,OAAO,KAAK,GAAG,IAAI,OAAO,KAAK,MAAM;QAAE,OAAO,IAAI,CAAC;IAEvD,MAAM,MAAM,GAAG,cAAc,EAAE,CAAC;IAChC,OAAO,MAAM,EAAE,gBAAgB,IAAI,KAAK,CAAC;AAC3C,CAAC;AAWD,SAAS,OAAO,CACd,GAAW,EACX,OAA4E;IAE5E,OAAO,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE;QACrC,MAAM,MAAM,GAAG,IAAI,GAAG,CAAC,GAAG,CAAC,CAAC;QAC5B,MAAM,SAAS,GAAG,MAAM,CAAC,QAAQ,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC;QAE9D,MAAM,GAAG,GAAG,SAAS,CAAC,OAAO,CAC3B;YACE,QAAQ,EAAE,MAAM,CAAC,QAAQ;YACzB,IAAI,EAAE,MAAM,CAAC,IAAI,IAAI,CAAC,MAAM,CAAC,QAAQ,KAAK,QAAQ,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC;YAC9D,IAAI,EAAE,MAAM,CAAC,QAAQ,GAAG,MAAM,CAAC,MAAM;YACrC,MAAM,EAAE,OAAO,CAAC,MAAM;YACtB,OAAO,EAAE;gBACP,cAAc,EAAE,kBAAkB;gBAClC,YAAY,EAAE,sBAAsB;gBACpC,GAAG,OAAO,CAAC,OAAO;aACnB;YACD,OAAO,EAAE,KAAK;SACf,EACD,CAAC,GAAG,EAAE,EAAE;YACN,IAAI,IAAI,GAAG,EAAE,CAAC;YACd,GAAG,CAAC,EAAE,CAAC,MAAM,EAAE,CAAC,KAAa,EAAE,EAAE,CAAC,CAAC,IAAI,IAAI,KAAK,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC;YAC9D,GAAG,CAAC,EAAE,CAAC,KAAK,EAAE,GAAG,EAAE;gBACjB,OAAO,CAAC,EAAE,UAAU,EAAE,GAAG,CAAC,UAAU,IAAI,CAAC,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC,CAAC;YAC3D,CAAC,CAAC,CAAC;QACL,CAAC,CACF,CAAC;QAEF,GAAG,CAAC,EAAE,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC;QACxB,GAAG,CAAC,EAAE,CAAC,SAAS,EAAE,GAAG,EAAE;YACrB,GAAG,CAAC,OAAO,EAAE,CAAC;YACd,MAAM,CAAC,IAAI,KAAK,CAAC,iBAAiB,CAAC,CAAC,CAAC;QACvC,CAAC,CAAC,CAAC;QAEH,IAAI,OAAO,CAAC,IAAI,EAAE,CAAC;YACjB,GAAG,CAAC,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;QAC1B,CAAC;QACD,GAAG,CAAC,GAAG,EAAE,CAAC;IACZ,CAAC,CAAC,CAAC;AACL,CAAC;AAED,8EAA8E;AAC9E,iBAAiB;AACjB,8EAA8E;AAE9E;;;GAGG;AACI,KAAK,UAAU,cAAc,CAAC,MAAc,EAAE,MAAe;IAClE,MAAM,OAAO,GAAG,MAAM,IAAI,SAAS,EAAE,CAAC;IACtC,IAAI,CAAC;QACH,MAAM,IAAI,GAAG,MAAM,OAAO,CAAC,GAAG,OAAO,mBAAmB,EAAE;YACxD,MAAM,EAAE,MAAM;YACd,OAAO,EAAE,EAAE,aAAa,EAAE,UAAU,MAAM,EAAE,EAAE;YAC9C,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC,EAAE,GAAG,EAAE,MAAM,EAAE,CAAC;SACtC,CAAC,CAAC;QAEH,IAAI,IAAI,CAAC,UAAU,KAAK,GAAG,EAAE,CAAC;YAC5B,OAAO,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,IAAI,CAAmB,CAAC;QACjD,CAAC;QAED,OAAO,EAAE,KAAK,EAAE,KAAK,EAAE,CAAC;IAC1B,CAAC;IAAC,MAAM,CAAC;QACP,4CAA4C;QAC5C,OAAO,IAAI,CAAC;IACd,CAAC;AACH,CAAC;AAED;;;GAGG;AACI,KAAK,UAAU,cAAc,CAAC,MAAc,EAAE,MAAe;IAClE,MAAM,OAAO,GAAG,MAAM,IAAI,SAAS,EAAE,CAAC;IACtC,IAAI,CAAC;QACH,MAAM,IAAI,GAAG,MAAM,OAAO,CAAC,GAAG,OAAO,eAAe,EAAE;YACpD,MAAM,EAAE,KAAK;YACb,OAAO,EAAE,EAAE,aAAa,EAAE,UAAU,MAAM,EAAE,EAAE;SAC/C,CAAC,CAAC;QAEH,IAAI,IAAI,CAAC,UAAU,KAAK,GAAG,EAAE,CAAC;YAC5B,OAAO,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,IAAI,CAAkB,CAAC;QAChD,CAAC;QACD,OAAO,IAAI,CAAC;IACd,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,IAAI,CAAC;IACd,CAAC;AACH,CAAC;AAED;;;;;;;GAOG;AACI,KAAK,UAAU,KAAK,CACzB,MAAc,EACd,OAAkD;IAElD,MAAM,MAAM,GAAG,OAAO,EAAE,MAAM,IAAI,eAAe,CAAC;IAClD,MAAM,gBAAgB,GAAG,OAAO,EAAE,SAAS,IAAI,IAAI,CAAC;IAEpD,0CAA0C;IAC1C,cAAc,CAAC;QACb,MAAM;QACN,MAAM;QACN,gBAAgB;QAChB,SAAS,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;QACnC,UAAU,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;KACrC,CAAC,CAAC;IAEH,yCAAyC;IACzC,MAAM,CAAC,UAAU,EAAE,SAAS,CAAC,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC;QAChD,cAAc,CAAC,MAAM,EAAE,MAAM,CAAC;QAC9B,cAAc,CAAC,MAAM,EAAE,MAAM,CAAC;KAC/B,CAAC,CAAC;IAEH,OAAO,EAAE,UAAU,EAAE,SAAS,EAAE,CAAC;AACnC,CAAC;AAED,0CAA0C;AAC1C,SAAgB,MAAM;IACpB,OAAO,eAAe,EAAE,CAAC;AAC3B,CAAC;AAED,8EAA8E;AAC9E,oCAAoC;AACpC,8EAA8E;AAEjE,QAAA,eAAe,GAAG,UAAU,CAAC;AAC7B,QAAA,gBAAgB,GAAG,WAAW,CAAC"}

package/dist/Baseline/index.d.ts

@@ -0,0 +1,43 @@
+/**
+ * Baseline Comparison
+ *
+ * Save dataset run results as a baseline snapshot, then compare future runs
+ * against it to detect regressions.
+ *
+ * Usage:
+ * ```ts
+ * // Save baseline after a known-good run
+ * const snapshot = createSnapshot(runResult, 'main');
+ * saveBaseline('.ai-rcp-baseline.json', snapshot);
+ *
+ * // Later, compare a new run against the baseline
+ * const comparison = compareToBaseline(newRunResult, snapshot, 0.02);
+ * if (!comparison.withinTolerance) { /* fail CI * / }
+ * ```
+ */
+import type { BaselineComparison, BaselineSnapshot, DatasetRunResult } from '../types';
+/**
+ * Create a baseline snapshot from a dataset run result.
+ */
+export declare function createSnapshot(runResult: DatasetRunResult, label: string): BaselineSnapshot;
+/**
+ * Save a baseline snapshot to a JSON file.
+ */
+export declare function saveBaseline(filePath: string, snapshot: BaselineSnapshot): void;
+/**
+ * Load a baseline snapshot from a JSON file.
+ */
+export declare function loadBaseline(filePath: string): BaselineSnapshot;
+/**
+ * Compare a current dataset run against a baseline snapshot.
+ *
+ * @param runResult - The current run to evaluate.
+ * @param baseline - The baseline snapshot to compare against.
+ * @param tolerance - Maximum allowed score drop (e.g. 0.02 = 2%). Default: 0.02.
+ */
+export declare function compareToBaseline(runResult: DatasetRunResult, baseline: BaselineSnapshot, tolerance?: number): BaselineComparison;
+/**
+ * Format a baseline comparison as a human-readable text block.
+ */
+export declare function formatBaselineReport(comparison: BaselineComparison): string;
+//# sourceMappingURL=index.d.ts.map

package/dist/Baseline/index.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/Baseline/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;GAgBG;AAGH,OAAO,KAAK,EACV,kBAAkB,EAClB,gBAAgB,EAChB,gBAAgB,EACjB,MAAM,UAAU,CAAC;AAMlB;;GAEG;AACH,wBAAgB,cAAc,CAC5B,SAAS,EAAE,gBAAgB,EAC3B,KAAK,EAAE,MAAM,GACZ,gBAAgB,CAmBlB;AAMD;;GAEG;AACH,wBAAgB,YAAY,CAAC,QAAQ,EAAE,MAAM,EAAE,QAAQ,EAAE,gBAAgB,GAAG,IAAI,CAE/E;AAED;;GAEG;AACH,wBAAgB,YAAY,CAAC,QAAQ,EAAE,MAAM,GAAG,gBAAgB,CAS/D;AAMD;;;;;;GAMG;AACH,wBAAgB,iBAAiB,CAC/B,SAAS,EAAE,gBAAgB,EAC3B,QAAQ,EAAE,gBAAgB,EAC1B,SAAS,GAAE,MAAa,GACvB,kBAAkB,CA0DpB;AAMD;;GAEG;AACH,wBAAgB,oBAAoB,CAAC,UAAU,EAAE,kBAAkB,GAAG,MAAM,CA0C3E"}

package/dist/Baseline/index.js

@@ -0,0 +1,195 @@
+"use strict";
+/**
+ * Baseline Comparison
+ *
+ * Save dataset run results as a baseline snapshot, then compare future runs
+ * against it to detect regressions.
+ *
+ * Usage:
+ * ```ts
+ * // Save baseline after a known-good run
+ * const snapshot = createSnapshot(runResult, 'main');
+ * saveBaseline('.ai-rcp-baseline.json', snapshot);
+ *
+ * // Later, compare a new run against the baseline
+ * const comparison = compareToBaseline(newRunResult, snapshot, 0.02);
+ * if (!comparison.withinTolerance) { /* fail CI * / }
+ * ```
+ */
+var __createBinding = (this && this.__createBinding) || (Object.create ? (function(o, m, k, k2) {
+    if (k2 === undefined) k2 = k;
+    var desc = Object.getOwnPropertyDescriptor(m, k);
+    if (!desc || ("get" in desc ? !m.__esModule : desc.writable || desc.configurable)) {
+      desc = { enumerable: true, get: function() { return m[k]; } };
+    }
+    Object.defineProperty(o, k2, desc);
+}) : (function(o, m, k, k2) {
+    if (k2 === undefined) k2 = k;
+    o[k2] = m[k];
+}));
+var __setModuleDefault = (this && this.__setModuleDefault) || (Object.create ? (function(o, v) {
+    Object.defineProperty(o, "default", { enumerable: true, value: v });
+}) : function(o, v) {
+    o["default"] = v;
+});
+var __importStar = (this && this.__importStar) || (function () {
+    var ownKeys = function(o) {
+        ownKeys = Object.getOwnPropertyNames || function (o) {
+            var ar = [];
+            for (var k in o) if (Object.prototype.hasOwnProperty.call(o, k)) ar[ar.length] = k;
+            return ar;
+        };
+        return ownKeys(o);
+    };
+    return function (mod) {
+        if (mod && mod.__esModule) return mod;
+        var result = {};
+        if (mod != null) for (var k = ownKeys(mod), i = 0; i < k.length; i++) if (k[i] !== "default") __createBinding(result, mod, k[i]);
+        __setModuleDefault(result, mod);
+        return result;
+    };
+})();
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.createSnapshot = createSnapshot;
+exports.saveBaseline = saveBaseline;
+exports.loadBaseline = loadBaseline;
+exports.compareToBaseline = compareToBaseline;
+exports.formatBaselineReport = formatBaselineReport;
+const fs = __importStar(require("fs"));
+// ---------------------------------------------------------------------------
+// Snapshot creation
+// ---------------------------------------------------------------------------
+/**
+ * Create a baseline snapshot from a dataset run result.
+ */
+function createSnapshot(runResult, label) {
+    return {
+        createdAt: new Date().toISOString(),
+        label,
+        averageScore: runResult.averageScore,
+        totalFailures: runResult.entries.reduce((sum, e) => sum + e.report.detectedFailures.length, 0),
+        passRate: runResult.totalEntries > 0
+            ? runResult.passedEntries / runResult.totalEntries
+            : 1.0,
+        entryScores: runResult.entries.map((e) => ({
+            name: e.name,
+            score: e.report.groundingScore,
+            failureCount: e.report.detectedFailures.length,
+        })),
+    };
+}
+// ---------------------------------------------------------------------------
+// Persistence
+// ---------------------------------------------------------------------------
+/**
+ * Save a baseline snapshot to a JSON file.
+ */
+function saveBaseline(filePath, snapshot) {
+    fs.writeFileSync(filePath, JSON.stringify(snapshot, null, 2), 'utf-8');
+}
+/**
+ * Load a baseline snapshot from a JSON file.
+ */
+function loadBaseline(filePath) {
+    const content = fs.readFileSync(filePath, 'utf-8');
+    const parsed = JSON.parse(content);
+    if (!parsed.label || typeof parsed.averageScore !== 'number') {
+        throw new Error('Invalid baseline file: missing required fields');
+    }
+    return parsed;
+}
+// ---------------------------------------------------------------------------
+// Comparison
+// ---------------------------------------------------------------------------
+/**
+ * Compare a current dataset run against a baseline snapshot.
+ *
+ * @param runResult - The current run to evaluate.
+ * @param baseline - The baseline snapshot to compare against.
+ * @param tolerance - Maximum allowed score drop (e.g. 0.02 = 2%). Default: 0.02.
+ */
+function compareToBaseline(runResult, baseline, tolerance = 0.02) {
+    const currentScore = runResult.averageScore;
+    const currentFailures = runResult.entries.reduce((sum, e) => sum + e.report.detectedFailures.length, 0);
+    const currentPassRate = runResult.totalEntries > 0
+        ? runResult.passedEntries / runResult.totalEntries
+        : 1.0;
+    const scoreDelta = currentScore - baseline.averageScore;
+    const failureDelta = currentFailures - baseline.totalFailures;
+    const passRateDelta = currentPassRate - baseline.passRate;
+    // Within tolerance if score didn't drop more than allowed
+    const withinTolerance = scoreDelta >= -tolerance;
+    // Per-entry comparison
+    const baselineMap = new Map(baseline.entryScores.map((e) => [e.name, e]));
+    const regressions = [];
+    const improvements = [];
+    for (const entry of runResult.entries) {
+        const base = baselineMap.get(entry.name);
+        if (!base)
+            continue; // New entry, no comparison possible
+        const delta = entry.report.groundingScore - base.score;
+        if (delta < -0.001) {
+            regressions.push({
+                name: entry.name,
+                baselineScore: base.score,
+                currentScore: entry.report.groundingScore,
+                delta,
+            });
+        }
+        else if (delta > 0.001) {
+            improvements.push({
+                name: entry.name,
+                baselineScore: base.score,
+                currentScore: entry.report.groundingScore,
+                delta,
+            });
+        }
+    }
+    return {
+        baselineLabel: baseline.label,
+        scoreDelta,
+        failureDelta,
+        passRateDelta,
+        withinTolerance,
+        tolerance,
+        regressions,
+        improvements,
+    };
+}
+// ---------------------------------------------------------------------------
+// Text report
+// ---------------------------------------------------------------------------
+/**
+ * Format a baseline comparison as a human-readable text block.
+ */
+function formatBaselineReport(comparison) {
+    const lines = [];
+    const sign = (n) => (n >= 0 ? '+' : '');
+    lines.push('── Baseline Comparison ─────────────────────────────');
+    lines.push(`Compared to : ${comparison.baselineLabel}`);
+    lines.push(`Score delta  : ${sign(comparison.scoreDelta)}${(comparison.scoreDelta * 100).toFixed(2)}%` +
+        ` (tolerance: ±${(comparison.tolerance * 100).toFixed(1)}%)`);
+    lines.push(`Failure delta: ${sign(comparison.failureDelta)}${comparison.failureDelta}`);
+    lines.push(`Pass rate Δ  : ${sign(comparison.passRateDelta)}${(comparison.passRateDelta * 100).toFixed(2)}%`);
+    const icon = comparison.withinTolerance ? '✅' : '❌';
+    const label = comparison.withinTolerance
+        ? 'WITHIN TOLERANCE'
+        : 'REGRESSION DETECTED';
+    lines.push(`Result       : ${icon} ${label}`);
+    if (comparison.regressions.length > 0) {
+        lines.push('');
+        lines.push(`Regressions (${comparison.regressions.length}):`);
+        for (const r of comparison.regressions) {
+            lines.push(`  ↓ ${r.name}: ${(r.baselineScore * 100).toFixed(1)}% → ${(r.currentScore * 100).toFixed(1)}% (${(r.delta * 100).toFixed(2)}%)`);
+        }
+    }
+    if (comparison.improvements.length > 0) {
+        lines.push('');
+        lines.push(`Improvements (${comparison.improvements.length}):`);
+        for (const imp of comparison.improvements) {
+            lines.push(`  ↑ ${imp.name}: ${(imp.baselineScore * 100).toFixed(1)}% → ${(imp.currentScore * 100).toFixed(1)}% (+${(imp.delta * 100).toFixed(2)}%)`);
+        }
+    }
+    return lines.join('\n');
+}
+//# sourceMappingURL=index.js.map