added mock of llms and full pipeline test

src/providers/mock_provider.mjs

@@ -0,0 +1,34 @@
+// src/providers/mock_provider.mjs
+// Simple deterministic provider for tests and local dry-runs.
+// Controlled via GENERATOR_PROVIDER/VERIFIER_PROVIDER/REWARD_PROVIDER/QUESTION_PROVIDER=mock
+import { BaseProvider } from './base.mjs';
+
+export class MockProvider extends BaseProvider {
+  constructor(stage = 'generator') {
+    super();
+    this.stage = stage;
+  }
+
+  async generate(prompt) {
+    switch (this.stage) {
+      case 'question':
+        // Return JSON so the question parser can pick it up directly
+        return JSON.stringify({
+          questions: [
+            'What is the main idea?',
+            'How does the text justify its claim?',
+          ],
+        });
+      case 'generator':
+        return `<think>mock reasoning</think>Here is a grounded answer.`;
+      case 'verifier':
+        return 'yes\nmock verifier rationale';
+      case 'reward':
+        return '0.9 good';
+      default:
+        return 'ok';
+    }
+  }
+}
+
+export default MockProvider;

src/providers/provider.mjs

@@ -2,6 +2,7 @@
 import { OllamaProvider } from "./ollama_provider.mjs";
 import { OpenAIProvider } from "./openai_provider.mjs";
 import { HttpProvider } from "./http_provider.mjs";
+import { MockProvider } from "./mock_provider.mjs";
 
 /**
  * Load provider for a specific pipeline stage:
@@ -16,6 +17,8 @@ export function loadProviderFor(stage) {
   const type = process.env[key] || process.env.PROVIDER_TYPE || "ollama";
 
   switch (type) {
+    case "mock":
+      return new MockProvider(stage);
     case "ollama":
       return new OllamaProvider(stage);
 

tests/pipeline.full.mock.test.mjs

@@ -0,0 +1,59 @@
+import { describe, it, expect, vi, beforeEach, afterEach } from 'vitest';
+import os from 'os';
+import path from 'path';
+
+/**
+ * Full pipeline test using mock providers + JSONL chunks.
+ * Ensures we can run a question-first batch without real LLMs or ES.
+ */
+describe('full pipeline (mock providers)', () => {
+  beforeEach(() => {
+    vi.resetModules();
+    process.env.PIPELINE_CHUNK_SOURCE = 'jsonl';
+    process.env.GENERATOR_PROVIDER = 'mock';
+    process.env.VERIFIER_PROVIDER = 'mock';
+    process.env.REWARD_PROVIDER = 'mock';
+    process.env.QUESTION_PROVIDER = 'mock';
+  });
+
+  afterEach(() => {
+    delete process.env.PIPELINE_CHUNK_SOURCE;
+    delete process.env.GENERATOR_PROVIDER;
+    delete process.env.VERIFIER_PROVIDER;
+    delete process.env.REWARD_PROVIDER;
+    delete process.env.QUESTION_PROVIDER;
+    vi.restoreAllMocks();
+    vi.unmock('../src/retrieval/jsonl_chunks.mjs');
+  });
+
+  it('processes mock chunks end-to-end and writes accepted samples', async () => {
+    // Mock chunk loader to avoid filesystem dependency
+    vi.doMock('../src/retrieval/jsonl_chunks.mjs', () => ({
+      loadRagChunks: vi.fn(async () => [
+        { id: 'chunk-1', content: 'Mock content A' },
+        { id: 'chunk-2', content: 'Mock content B' },
+      ]),
+    }));
+
+    // Ensure no accidental ES calls
+    vi.doMock('../src/retrieval/retrieval.mjs', () => ({
+      fetchChunksFromIndex: vi.fn(),
+    }));
+
+    const { runPipelineBatch } = await import('../src/pipeline/batch.mjs');
+
+    const outPath = path.join(os.tmpdir(), `mock-pipeline-${Date.now()}.jsonl`);
+    const result = await runPipelineBatch({
+      seedMode: 'question-first',
+      limit: 3, // question cap
+      outPath,
+      verbose: false,
+      logger: { log() {}, error() {} },
+    });
+
+    expect(result.mode).toBe('question-first');
+    expect(result.processed).toBe(3);
+    expect(result.accepted).toBe(3);
+    expect(result.statusCounts.accepted).toBe(3);
+  });
+});