@thinkhive/sdk 3.1.1 → 4.0.0

package/dist/api/deterministic-graders.js

@@ -0,0 +1,191 @@
+"use strict";
+/**
+ * ThinkHive SDK v3.0 - Deterministic Graders API
+ *
+ * API for running deterministic (code-based) evaluations
+ */
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.deterministicGraders = void 0;
+exports.createRegexRule = createRegexRule;
+exports.createContainsRule = createContainsRule;
+exports.createLengthRule = createLengthRule;
+exports.createJsonSchemaRule = createJsonSchemaRule;
+exports.allRulesPassed = allRulesPassed;
+exports.getFailedRules = getFailedRules;
+exports.calculateAverageScore = calculateAverageScore;
+const client_1 = require("../core/client");
+// ============================================================================
+// DETERMINISTIC GRADERS API CLIENT
+// ============================================================================
+/**
+ * Deterministic Graders API client for code-based evaluations
+ */
+exports.deterministicGraders = {
+    /**
+     * Run deterministic evaluation on a single trace
+     *
+     * @example
+     * ```typescript
+     * const result = await deterministicGraders.evaluate({
+     *   traceId: 'trace_123',
+     *   criterionId: 'criterion_456',
+     * });
+     * console.log(`Passed: ${result.passed}, Score: ${result.score}`);
+     * ```
+     */
+    async evaluate(options) {
+        return (0, client_1.apiRequestWithData)('/deterministic-graders/evaluate', {
+            method: 'POST',
+            body: options,
+            apiVersion: 'none',
+        });
+    },
+    /**
+     * Run deterministic evaluations on multiple traces
+     *
+     * @example
+     * ```typescript
+     * const { results, summary } = await deterministicGraders.bulkEvaluate({
+     *   evaluations: [
+     *     { traceId: 'trace_1', criterionId: 'criterion_456' },
+     *     { traceId: 'trace_2', criterionId: 'criterion_456' },
+     *     { traceId: 'trace_3', criterionId: 'criterion_456' },
+     *   ],
+     * });
+     * console.log(`Pass rate: ${summary.passRate * 100}%`);
+     * ```
+     */
+    async bulkEvaluate(options) {
+        return (0, client_1.apiRequestWithData)('/deterministic-graders/bulk-evaluate', {
+            method: 'POST',
+            body: options,
+            apiVersion: 'none',
+        });
+    },
+    /**
+     * Get available rule types with descriptions
+     *
+     * @example
+     * ```typescript
+     * const ruleTypes = await deterministicGraders.getRuleTypes();
+     * for (const type of ruleTypes) {
+     *   console.log(`${type.name}: ${type.description}`);
+     * }
+     * ```
+     */
+    async getRuleTypes() {
+        return (0, client_1.apiRequestWithData)('/deterministic-graders/rule-types', { apiVersion: 'none' });
+    },
+    /**
+     * Get rule templates
+     *
+     * @example
+     * ```typescript
+     * const templates = await deterministicGraders.getTemplates();
+     * const noPiiTemplate = templates.find(t => t.id === 'no_pii');
+     * ```
+     */
+    async getTemplates() {
+        return (0, client_1.apiRequestWithData)('/deterministic-graders/templates', { apiVersion: 'none' });
+    },
+};
+// ============================================================================
+// HELPER FUNCTIONS
+// ============================================================================
+/**
+ * Create a regex rule configuration
+ *
+ * @param pattern - Regular expression pattern
+ * @param flags - Regex flags (default: 'gi')
+ * @returns Rule configuration object
+ *
+ * @example
+ * ```typescript
+ * const config = createRegexRule('\\b(error|fail)\\b', 'gi');
+ * ```
+ */
+function createRegexRule(pattern, flags = 'gi') {
+    return { pattern, flags };
+}
+/**
+ * Create a contains rule configuration
+ *
+ * @param values - Strings to check for
+ * @param caseSensitive - Whether comparison is case-sensitive
+ * @returns Rule configuration object
+ *
+ * @example
+ * ```typescript
+ * const config = createContainsRule(['hello', 'hi', 'hey'], false);
+ * ```
+ */
+function createContainsRule(values, caseSensitive = false) {
+    return { values, caseSensitive };
+}
+/**
+ * Create a length rule configuration
+ *
+ * @param min - Minimum length (optional)
+ * @param max - Maximum length (optional)
+ * @returns Rule configuration object
+ *
+ * @example
+ * ```typescript
+ * const config = createLengthRule(50, 1000);
+ * ```
+ */
+function createLengthRule(min, max) {
+    return { min, max };
+}
+/**
+ * Create a JSON schema rule configuration
+ *
+ * @param schema - JSON Schema object
+ * @returns Rule configuration object
+ *
+ * @example
+ * ```typescript
+ * const config = createJsonSchemaRule({
+ *   type: 'object',
+ *   required: ['name', 'email'],
+ *   properties: {
+ *     name: { type: 'string' },
+ *     email: { type: 'string', format: 'email' },
+ *   },
+ * });
+ * ```
+ */
+function createJsonSchemaRule(schema) {
+    return { schema };
+}
+/**
+ * Check if all rule results passed
+ *
+ * @param results - Array of rule results
+ * @returns Whether all rules passed
+ */
+function allRulesPassed(results) {
+    return results.every(r => r.passed);
+}
+/**
+ * Get failed rules from results
+ *
+ * @param results - Array of rule results
+ * @returns Array of failed rule results
+ */
+function getFailedRules(results) {
+    return results.filter(r => !r.passed);
+}
+/**
+ * Calculate average score from rule results
+ *
+ * @param results - Array of rule results
+ * @returns Average score (0-100)
+ */
+function calculateAverageScore(results) {
+    if (results.length === 0)
+        return 0;
+    const sum = results.reduce((acc, r) => acc + r.score, 0);
+    return sum / results.length;
+}
+//# sourceMappingURL=data:application/json;base64,{"version":3,"file":"deterministic-graders.js","sourceRoot":"","sources":["../../src/api/deterministic-graders.ts"],"names":[],"mappings":";AAAA;;;;GAIG;;;AAsLH,0CAEC;AAcD,gDAKC;AAcD,4CAEC;AAoBD,oDAEC;AAQD,wCAEC;AAQD,wCAEC;AAQD,sDAIC;AA/QD,2CAAoD;AA8EpD,+EAA+E;AAC/E,mCAAmC;AACnC,+EAA+E;AAE/E;;GAEG;AACU,QAAA,oBAAoB,GAAG;IAClC;;;;;;;;;;;OAWG;IACH,KAAK,CAAC,QAAQ,CAAC,OAAwB;QACrC,OAAO,IAAA,2BAAkB,EAA0B,iCAAiC,EAAE;YACpF,MAAM,EAAE,MAAM;YACd,IAAI,EAAE,OAAO;YACb,UAAU,EAAE,MAAM;SACnB,CAAC,CAAC;IACL,CAAC;IAED;;;;;;;;;;;;;;OAcG;IACH,KAAK,CAAC,YAAY,CAAC,OAA4B;QAC7C,OAAO,IAAA,2BAAkB,EAAqB,sCAAsC,EAAE;YACpF,MAAM,EAAE,MAAM;YACd,IAAI,EAAE,OAAO;YACb,UAAU,EAAE,MAAM;SACnB,CAAC,CAAC;IACL,CAAC;IAED;;;;;;;;;;OAUG;IACH,KAAK,CAAC,YAAY;QAChB,OAAO,IAAA,2BAAkB,EACvB,mCAAmC,EACnC,EAAE,UAAU,EAAE,MAAM,EAAE,CACvB,CAAC;IACJ,CAAC;IAED;;;;;;;;OAQG;IACH,KAAK,CAAC,YAAY;QAChB,OAAO,IAAA,2BAAkB,EACvB,kCAAkC,EAClC,EAAE,UAAU,EAAE,MAAM,EAAE,CACvB,CAAC;IACJ,CAAC;CACF,CAAC;AAEF,+EAA+E;AAC/E,mBAAmB;AACnB,+EAA+E;AAE/E;;;;;;;;;;;GAWG;AACH,SAAgB,eAAe,CAAC,OAAe,EAAE,KAAK,GAAG,IAAI;IAC3D,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC;AAC5B,CAAC;AAED;;;;;;;;;;;GAWG;AACH,SAAgB,kBAAkB,CAChC,MAAgB,EAChB,aAAa,GAAG,KAAK;IAErB,OAAO,EAAE,MAAM,EAAE,aAAa,EAAE,CAAC;AACnC,CAAC;AAED;;;;;;;;;;;GAWG;AACH,SAAgB,gBAAgB,CAAC,GAAY,EAAE,GAAY;IACzD,OAAO,EAAE,GAAG,EAAE,GAAG,EAAE,CAAC;AACtB,CAAC;AAED;;;;;;;;;;;;;;;;;GAiBG;AACH,SAAgB,oBAAoB,CAAC,MAA+B;IAClE,OAAO,EAAE,MAAM,EAAE,CAAC;AACpB,CAAC;AAED;;;;;GAKG;AACH,SAAgB,cAAc,CAAC,OAAqB;IAClD,OAAO,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC;AACtC,CAAC;AAED;;;;;GAKG;AACH,SAAgB,cAAc,CAAC,OAAqB;IAClD,OAAO,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC;AACxC,CAAC;AAED;;;;;GAKG;AACH,SAAgB,qBAAqB,CAAC,OAAqB;IACzD,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,CAAC,CAAC;IACnC,MAAM,GAAG,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC;IACzD,OAAO,GAAG,GAAG,OAAO,CAAC,MAAM,CAAC;AAC9B,CAAC","sourcesContent":["/**\n * ThinkHive SDK v3.0 - Deterministic Graders API\n *\n * API for running deterministic (code-based) evaluations\n */\n\nimport { apiRequestWithData } from '../core/client';\n\n// ============================================================================\n// TYPES\n// ============================================================================\n\nexport type RuleType =\n  | 'regex'\n  | 'contains'\n  | 'not_contains'\n  | 'json_valid'\n  | 'json_schema'\n  | 'length'\n  | 'pii_check'\n  | 'sentiment'\n  | 'latency'\n  | 'token_count';\n\nexport interface DeterministicEvalResult {\n  passed: boolean;\n  score: number;\n  reasoning: string;\n  ruleResults?: RuleResult[];\n  metadata?: Record<string, unknown>;\n}\n\nexport interface RuleResult {\n  ruleId: string;\n  ruleName: string;\n  ruleType: RuleType;\n  passed: boolean;\n  score: number;\n  details?: string;\n}\n\nexport interface EvaluateOptions {\n  traceId: string;\n  criterionId: string;\n}\n\nexport interface BulkEvaluateOptions {\n  evaluations: Array<{\n    traceId: string;\n    criterionId: string;\n  }>;\n}\n\nexport interface BulkEvaluateResult {\n  results: Array<{\n    traceId: string;\n    criterionId: string;\n    passed: boolean;\n    score: number;\n    error?: string;\n  }>;\n  summary: {\n    total: number;\n    passed: number;\n    failed: number;\n    passRate: number;\n  };\n}\n\nexport interface RuleTypeInfo {\n  id: RuleType;\n  name: string;\n  description: string;\n  configFields: string[];\n}\n\nexport interface RuleTemplate {\n  id: string;\n  name: string;\n  description: string;\n  ruleType: RuleType;\n  config: Record<string, unknown>;\n}\n\n// ============================================================================\n// DETERMINISTIC GRADERS API CLIENT\n// ============================================================================\n\n/**\n * Deterministic Graders API client for code-based evaluations\n */\nexport const deterministicGraders = {\n  /**\n   * Run deterministic evaluation on a single trace\n   *\n   * @example\n   * ```typescript\n   * const result = await deterministicGraders.evaluate({\n   *   traceId: 'trace_123',\n   *   criterionId: 'criterion_456',\n   * });\n   * console.log(`Passed: ${result.passed}, Score: ${result.score}`);\n   * ```\n   */\n  async evaluate(options: EvaluateOptions): Promise<DeterministicEvalResult> {\n    return apiRequestWithData<DeterministicEvalResult>('/deterministic-graders/evaluate', {\n      method: 'POST',\n      body: options,\n      apiVersion: 'none',\n    });\n  },\n\n  /**\n   * Run deterministic evaluations on multiple traces\n   *\n   * @example\n   * ```typescript\n   * const { results, summary } = await deterministicGraders.bulkEvaluate({\n   *   evaluations: [\n   *     { traceId: 'trace_1', criterionId: 'criterion_456' },\n   *     { traceId: 'trace_2', criterionId: 'criterion_456' },\n   *     { traceId: 'trace_3', criterionId: 'criterion_456' },\n   *   ],\n   * });\n   * console.log(`Pass rate: ${summary.passRate * 100}%`);\n   * ```\n   */\n  async bulkEvaluate(options: BulkEvaluateOptions): Promise<BulkEvaluateResult> {\n    return apiRequestWithData<BulkEvaluateResult>('/deterministic-graders/bulk-evaluate', {\n      method: 'POST',\n      body: options,\n      apiVersion: 'none',\n    });\n  },\n\n  /**\n   * Get available rule types with descriptions\n   *\n   * @example\n   * ```typescript\n   * const ruleTypes = await deterministicGraders.getRuleTypes();\n   * for (const type of ruleTypes) {\n   *   console.log(`${type.name}: ${type.description}`);\n   * }\n   * ```\n   */\n  async getRuleTypes(): Promise<RuleTypeInfo[]> {\n    return apiRequestWithData<RuleTypeInfo[]>(\n      '/deterministic-graders/rule-types',\n      { apiVersion: 'none' }\n    );\n  },\n\n  /**\n   * Get rule templates\n   *\n   * @example\n   * ```typescript\n   * const templates = await deterministicGraders.getTemplates();\n   * const noPiiTemplate = templates.find(t => t.id === 'no_pii');\n   * ```\n   */\n  async getTemplates(): Promise<RuleTemplate[]> {\n    return apiRequestWithData<RuleTemplate[]>(\n      '/deterministic-graders/templates',\n      { apiVersion: 'none' }\n    );\n  },\n};\n\n// ============================================================================\n// HELPER FUNCTIONS\n// ============================================================================\n\n/**\n * Create a regex rule configuration\n *\n * @param pattern - Regular expression pattern\n * @param flags - Regex flags (default: 'gi')\n * @returns Rule configuration object\n *\n * @example\n * ```typescript\n * const config = createRegexRule('\\\\b(error|fail)\\\\b', 'gi');\n * ```\n */\nexport function createRegexRule(pattern: string, flags = 'gi'): { pattern: string; flags: string } {\n  return { pattern, flags };\n}\n\n/**\n * Create a contains rule configuration\n *\n * @param values - Strings to check for\n * @param caseSensitive - Whether comparison is case-sensitive\n * @returns Rule configuration object\n *\n * @example\n * ```typescript\n * const config = createContainsRule(['hello', 'hi', 'hey'], false);\n * ```\n */\nexport function createContainsRule(\n  values: string[],\n  caseSensitive = false\n): { values: string[]; caseSensitive: boolean } {\n  return { values, caseSensitive };\n}\n\n/**\n * Create a length rule configuration\n *\n * @param min - Minimum length (optional)\n * @param max - Maximum length (optional)\n * @returns Rule configuration object\n *\n * @example\n * ```typescript\n * const config = createLengthRule(50, 1000);\n * ```\n */\nexport function createLengthRule(min?: number, max?: number): { min?: number; max?: number } {\n  return { min, max };\n}\n\n/**\n * Create a JSON schema rule configuration\n *\n * @param schema - JSON Schema object\n * @returns Rule configuration object\n *\n * @example\n * ```typescript\n * const config = createJsonSchemaRule({\n *   type: 'object',\n *   required: ['name', 'email'],\n *   properties: {\n *     name: { type: 'string' },\n *     email: { type: 'string', format: 'email' },\n *   },\n * });\n * ```\n */\nexport function createJsonSchemaRule(schema: Record<string, unknown>): { schema: Record<string, unknown> } {\n  return { schema };\n}\n\n/**\n * Check if all rule results passed\n *\n * @param results - Array of rule results\n * @returns Whether all rules passed\n */\nexport function allRulesPassed(results: RuleResult[]): boolean {\n  return results.every(r => r.passed);\n}\n\n/**\n * Get failed rules from results\n *\n * @param results - Array of rule results\n * @returns Array of failed rule results\n */\nexport function getFailedRules(results: RuleResult[]): RuleResult[] {\n  return results.filter(r => !r.passed);\n}\n\n/**\n * Calculate average score from rule results\n *\n * @param results - Array of rule results\n * @returns Average score (0-100)\n */\nexport function calculateAverageScore(results: RuleResult[]): number {\n  if (results.length === 0) return 0;\n  const sum = results.reduce((acc, r) => acc + r.score, 0);\n  return sum / results.length;\n}\n"]}

package/dist/api/eval-health.d.ts

@@ -0,0 +1,250 @@
+/**
+ * ThinkHive SDK v3.0 - Evaluation Health API
+ *
+ * API for eval saturation monitoring, regression detection, and health reports
+ */
+export type SaturationType = 'ceiling' | 'floor' | 'healthy';
+export type HealthStatus = 'healthy' | 'warning' | 'critical';
+export type RegressionSeverity = 'minor' | 'moderate' | 'severe';
+export interface EvalHealthSnapshot {
+    id: string;
+    companyId: string;
+    agentId: string;
+    criterionId?: string;
+    snapshotDate: string;
+    passRate?: string;
+    evalCount?: number;
+    meanScore?: string;
+    saturationType?: SaturationType;
+    daysAtSaturation?: number;
+    trendDirection?: string;
+    trendStrength?: string;
+    healthStatus?: HealthStatus;
+    healthScore?: string;
+    createdAt: string;
+}
+export interface EvalRegression {
+    id: string;
+    companyId: string;
+    agentId: string;
+    criterionId?: string;
+    severity: RegressionSeverity;
+    baselinePassRate: string;
+    currentPassRate: string;
+    delta: string;
+    deltaPercent?: string;
+    baselinePeriodStart: string;
+    baselinePeriodEnd: string;
+    currentPeriodStart: string;
+    currentPeriodEnd: string;
+    baselineEvalCount?: number;
+    currentEvalCount?: number;
+    suspectedCauses?: unknown[];
+    isSignificant?: boolean;
+    isResolved: boolean;
+    isAcknowledged: boolean;
+    resolvedAt?: string;
+    resolvedBy?: string;
+    resolutionType?: string;
+    resolutionNotes?: string;
+    acknowledgedAt?: string;
+    acknowledgedBy?: string;
+    detectedAt: string;
+    createdAt: string;
+}
+export interface HealthReport {
+    agentId: string;
+    generatedAt: string;
+    overallHealth: HealthStatus;
+    overallScore: number;
+    passRate: number;
+    evalCount: number;
+    saturationStatus: {
+        type: SaturationType;
+        daysAtSaturation: number;
+        recommendation: string;
+    };
+    regressionCount: number;
+    activeRegressions: EvalRegression[];
+    trend: {
+        direction: 'improving' | 'stable' | 'declining';
+        strength: number;
+        description: string;
+    };
+    recommendations: string[];
+}
+export interface CreateSnapshotOptions {
+    agentId: string;
+    criterionId?: string;
+    snapshotDate: string;
+    passRate?: string;
+    evalCount?: number;
+    meanScore?: string;
+    saturationType?: SaturationType;
+    daysAtSaturation?: number;
+    trendDirection?: string;
+    trendStrength?: string;
+    healthStatus?: HealthStatus;
+    healthScore?: string;
+}
+export interface CreateRegressionOptions {
+    agentId: string;
+    criterionId?: string;
+    severity: RegressionSeverity;
+    baselinePassRate: string;
+    currentPassRate: string;
+    delta: string;
+    deltaPercent?: string;
+    baselinePeriodStart: string;
+    baselinePeriodEnd: string;
+    currentPeriodStart: string;
+    currentPeriodEnd: string;
+    baselineEvalCount?: number;
+    currentEvalCount?: number;
+    suspectedCauses?: unknown[];
+    isSignificant?: boolean;
+    detectedAt?: string;
+}
+export interface ResolveRegressionOptions {
+    resolutionType: string;
+    notes?: string;
+}
+export interface GetSnapshotsOptions {
+    agentId: string;
+    criterionId?: string;
+    startDate?: string;
+    endDate?: string;
+}
+/**
+ * Evaluation Health API client for monitoring eval quality and detecting regressions
+ */
+export declare const evalHealth: {
+    /**
+     * Get comprehensive health report for an agent
+     *
+     * @example
+     * ```typescript
+     * const report = await evalHealth.getReport('agent_123');
+     * console.log(`Overall health: ${report.overallHealth}`);
+     * console.log(`Active regressions: ${report.regressionCount}`);
+     * ```
+     */
+    getReport(agentId: string): Promise<HealthReport>;
+    /**
+     * Get historical health snapshots
+     *
+     * @example
+     * ```typescript
+     * const snapshots = await evalHealth.getSnapshots({
+     *   agentId: 'agent_123',
+     *   startDate: '2024-01-01T00:00:00Z',
+     *   endDate: '2024-01-31T23:59:59Z',
+     * });
+     * ```
+     */
+    getSnapshots(options: GetSnapshotsOptions): Promise<EvalHealthSnapshot[]>;
+    /**
+     * Get latest health snapshot
+     *
+     * @example
+     * ```typescript
+     * const snapshot = await evalHealth.getLatestSnapshot('agent_123');
+     * ```
+     */
+    getLatestSnapshot(agentId: string, criterionId?: string): Promise<EvalHealthSnapshot | null>;
+    /**
+     * Record a health snapshot
+     *
+     * @example
+     * ```typescript
+     * const snapshot = await evalHealth.recordSnapshot({
+     *   agentId: 'agent_123',
+     *   snapshotDate: new Date().toISOString(),
+     *   passRate: '0.85',
+     *   evalCount: 150,
+     *   healthStatus: 'healthy',
+     * });
+     * ```
+     */
+    recordSnapshot(options: CreateSnapshotOptions): Promise<EvalHealthSnapshot>;
+    /**
+     * Get unresolved regressions for an agent
+     *
+     * @example
+     * ```typescript
+     * const regressions = await evalHealth.getRegressions('agent_123');
+     * for (const regression of regressions) {
+     *   console.log(`${regression.severity}: ${regression.delta}% drop`);
+     * }
+     * ```
+     */
+    getRegressions(agentId: string): Promise<EvalRegression[]>;
+    /**
+     * Record a new regression
+     *
+     * @example
+     * ```typescript
+     * const regression = await evalHealth.recordRegression({
+     *   agentId: 'agent_123',
+     *   severity: 'moderate',
+     *   baselinePassRate: '0.92',
+     *   currentPassRate: '0.78',
+     *   delta: '-0.14',
+     *   baselinePeriodStart: '2024-01-01T00:00:00Z',
+     *   baselinePeriodEnd: '2024-01-15T23:59:59Z',
+     *   currentPeriodStart: '2024-01-16T00:00:00Z',
+     *   currentPeriodEnd: '2024-01-31T23:59:59Z',
+     * });
+     * ```
+     */
+    recordRegression(options: CreateRegressionOptions): Promise<EvalRegression>;
+    /**
+     * Resolve a regression
+     *
+     * @example
+     * ```typescript
+     * await evalHealth.resolveRegression('regression_123', {
+     *   resolutionType: 'fixed',
+     *   notes: 'Updated prompt template to address quality issues',
+     * });
+     * ```
+     */
+    resolveRegression(regressionId: string, options: ResolveRegressionOptions): Promise<void>;
+    /**
+     * Acknowledge a regression
+     *
+     * @example
+     * ```typescript
+     * await evalHealth.acknowledgeRegression('regression_123');
+     * ```
+     */
+    acknowledgeRegression(regressionId: string): Promise<void>;
+};
+/**
+ * Check if health status indicates an issue
+ *
+ * @param status - Health status to check
+ * @returns Whether the status indicates a problem
+ */
+export declare function hasHealthIssue(status: HealthStatus): boolean;
+/**
+ * Get severity level as numeric value for sorting
+ *
+ * @param severity - Regression severity
+ * @returns Numeric severity (1-3, higher is worse)
+ */
+export declare function getSeverityLevel(severity: RegressionSeverity): number;
+/**
+ * Check if evaluation is saturated
+ *
+ * @param snapshot - Health snapshot to check
+ * @returns Whether evaluation is at ceiling or floor
+ */
+export declare function isSaturated(snapshot: EvalHealthSnapshot): boolean;
+/**
+ * Get recommendation for saturation type
+ *
+ * @param saturationType - Type of saturation
+ * @returns Recommendation string
+ */
+export declare function getSaturationRecommendation(saturationType: SaturationType): string;