@dooor-ai/toolkit/src/evals/contextual-recall.ts

Guards, Evals & Observability for AI applications - works seamlessly with LangChain/LangGraph

import { Eval } from "./base";
import { EvalResult, EvalConfig } from "../core/types";
import { getCortexDBClient, getGlobalProviderName } from "../observability/cortexdb-client";

export interface ContextualRecallConfig extends EvalConfig {
  /** Expected answer or ground truth */
  expectedAnswer?: string;
  /** Retrieved context/documents */
  context?: string;
}

/**
 * ContextualRecallEval - Measures if the retrieval system found all relevant information
 *
 * Evaluates whether the retrieved context contains all the information needed to
 * answer the question correctly. High recall means nothing important was missed.
 *
 * Example:
 * ```typescript
 * const eval = new ContextualRecallEval({
 *   threshold: 0.8,
 *   expectedAnswer: "Paris is the capital and was founded in 3rd century BC.",
 *   context: "Paris is the capital of France."
 * });
 * const result = await eval.evaluate(
 *   "What is the capital of France and when was it founded?",
 *   "The capital is Paris."
 * );
 * // result.score = 0.5 (missed founding date info), result.passed = false
 * ```
 */
export class ContextualRecallEval extends Eval {
  private expectedAnswer?: string;
  private context?: string;

  constructor(config: ContextualRecallConfig = {}) {
    super(config);
    this.expectedAnswer = config.expectedAnswer;
    this.context = config.context;
  }

  get name(): string {
    return "ContextualRecallEval";
  }

  /**
   * Set expected answer dynamically
   */
  setExpectedAnswer(answer: string): void {
    this.expectedAnswer = answer;
  }

  /**
   * Set context dynamically
   */
  setContext(context: string): void {
    this.context = context;
  }

  async evaluate(
    input: string,
    output: string,
    metadata?: Record<string, any>
  ): Promise<EvalResult> {
    const startTime = Date.now();

    const context = this.context || metadata?.context || metadata?.retrievedDocs;
    const expected = this.expectedAnswer || metadata?.expectedAnswer || metadata?.groundTruth;

    if (!context || !expected) {
      return {
        name: this.name,
        score: 0.5,
        passed: false,
        details: "Missing required data for contextual recall. Provide 'context' and 'expectedAnswer' via config or metadata.",
        metadata: {
          latency: Date.now() - startTime,
        },
        timestamp: new Date(),
      };
    }

    try {
      const cortexClient = getCortexDBClient();
      const providerName = getGlobalProviderName();

      const prompt = this.buildPrompt(input, expected, context);

      const response = await cortexClient.invokeAI({
        prompt,
        usage: "evaluation",
        providerName: providerName || undefined,
        temperature: 0.0,
        maxTokens: 300,
      });

      const score = this.parseScore(response.text);
      const passed = score >= this.getThreshold();

      return {
        name: this.name,
        score,
        passed,
        details: `Contextual recall score: ${score.toFixed(2)}. ${passed ? "PASSED" : "FAILED"} (threshold: ${this.getThreshold()})`,
        metadata: {
          latency: Date.now() - startTime,
          judgeResponse: response.text,
          contextLength: context.length,
        },
        timestamp: new Date(),
      };
    } catch (error) {
      console.error("ContextualRecallEval failed:", error);

      return {
        name: this.name,
        score: 0.5,
        passed: false,
        details: `Eval failed: ${error instanceof Error ? error.message : "Unknown error"}`,
        metadata: {
          error: String(error),
          latency: Date.now() - startTime,
        },
        timestamp: new Date(),
      };
    }
  }

  private buildPrompt(question: string, expectedAnswer: string, context: string): string {
    return `You are an expert evaluator. Your task is to measure RECALL: did the retrieved context contain all the information needed to produce the expected answer?

Question: "${question}"

Expected Answer (ground truth): """
${expectedAnswer}
"""

Retrieved Context: """
${context}
"""

Evaluate recall (how complete is the context):
- 1.0 = Context contains ALL information from expected answer
- 0.7-0.9 = Context contains most information, minor gaps
- 0.4-0.6 = Context missing significant information
- 0.0-0.3 = Context missing most/all key information

Output ONLY a JSON object in this exact format:
{
  "score": 0.85,
  "reasoning": "What information is present vs missing from context"
}`;
  }

  private parseScore(response: string): number {
    try {
      const jsonMatch = response.match(/\{[\s\S]*\}/);
      if (jsonMatch) {
        const parsed = JSON.parse(jsonMatch[0]);
        if (typeof parsed.score === "number") {
          return Math.max(0, Math.min(1, parsed.score));
        }
      }

      const numberMatch = response.match(/\b0?\.\d+\b|\b1\.0\b|\b[01]\b/);
      if (numberMatch) {
        return Math.max(0, Math.min(1, parseFloat(numberMatch[0])));
      }

      console.warn("Could not parse score from response:", response);
      return 0.5;
    } catch (error) {
      console.error("Error parsing score:", error);
      return 0.5;
    }
  }
}