wisp/src/lib/learn/summary.test.ts

import { describe, it, expect } from 'vitest';
import { summarize } from './summary';
import type { Segment } from '../types';

const seg = (start: number, text: string, id?: string): Segment => ({
  id,
  start,
  end: start + 1,
  text,
});

describe('summarize', () => {
  it('returns [] for no segments', () => {
    expect(summarize([])).toEqual([]);
  });

  it('returns [] when maxSentences <= 0', () => {
    expect(summarize([seg(0, 'Neural networks learn.')], { maxSentences: 0 })).toEqual(
      [],
    );
  });

  it('picks the top-N highest-frequency sentences', () => {
    // "neural" + "networks" dominate; the off-topic sentence should be dropped.
    const segs = [
      seg(0, 'Neural networks process data.'),
      seg(1, 'Neural networks learn from neural networks.'),
      seg(2, 'The weather today is sunny.'),
    ];
    const out = summarize(segs, { maxSentences: 2 });
    expect(out.length).toBe(2);
    const texts = out.map((s) => s.text);
    expect(texts).not.toContain('The weather today is sunny.');
  });

  it('returns sentences in chronological order with the right start + id', () => {
    const segs = [
      seg(10, 'Gradient descent optimizes the gradient.', 'segB'),
      seg(2, 'Gradient descent uses the gradient slope.', 'segA'),
    ];
    const out = summarize(segs, { maxSentences: 2 });
    // Both kept; chronological order means start 2 comes before start 10.
    expect(out.map((s) => s.start)).toEqual([2, 10]);
    expect(out[0]!.segmentId).toBe('segA');
    expect(out[1]!.segmentId).toBe('segB');
  });

  it('flattens multi-sentence segments and tags them with the segment start', () => {
    const segs = [seg(5, 'Alpha beta gamma. Alpha beta delta.', 'multi')];
    const out = summarize(segs, { maxSentences: 5 });
    expect(out.length).toBe(2);
    for (const s of out) {
      expect(s.start).toBe(5);
      expect(s.segmentId).toBe('multi');
    }
  });
});