/**
 * llmindex.io — worker unit tests (cost gate, judge protocol invariants)
 * Author:  Simon-Pierre Boucher
 * Contact: contact@spboucher.ai
 * License: Proprietary — © Simon-Pierre Boucher, all rights reserved
 */
import { describe, expect, it } from 'vitest';
import { DEGRADED_FAILURE_RATE, EST_TOKENS, estimateBatchCostUsd } from './eval-runner';
import {
  JudgeConfigError,
  eligibleJudges,
  loadJudgeConfig,
  parseVerdict,
} from './judges/config';

describe('cost guardrails', () => {
  it('estimates batch cost from pricing and k samples', () => {
    const cost = estimateBatchCostUsd(200, 1, { promptPerM: 3, completionPerM: 15 });
    expect(cost).toBeCloseTo(
      (200 * (EST_TOKENS.prompt * 3 + EST_TOKENS.completion * 15)) / 1_000_000,
      8,
    );
    const withK = estimateBatchCostUsd(200, 5, { promptPerM: 3, completionPerM: 15 });
    expect(withK).toBeCloseTo(cost * 5, 8);
  });
  it('degraded threshold matches §6 (2%)', () => {
    expect(DEGRADED_FAILURE_RATE).toBe(0.02);
  });
});

describe('judge config', () => {
  it('requires ≥2 judges across ≥2 providers', () => {
    expect(() => loadJudgeConfig({ JUDGE_MODELS: 'a/x' } as NodeJS.ProcessEnv)).toThrow(
      JudgeConfigError,
    );
    expect(() => loadJudgeConfig({ JUDGE_MODELS: 'a/x,a/y' } as NodeJS.ProcessEnv)).toThrow(
      JudgeConfigError,
    );
    const ok = loadJudgeConfig({ JUDGE_MODELS: 'a/x, b/y' } as NodeJS.ProcessEnv);
    expect(ok.slugs).toEqual(['a/x', 'b/y']);
  });
  it('a model never judges its own duels', () => {
    const config = { slugs: ['a/x', 'b/y', 'c/z'] };
    expect(eligibleJudges(config, 'a/x', 'q/r')).toEqual(['b/y', 'c/z']);
    expect(eligibleJudges(config, 'a/x', 'b/y')).toEqual(['c/z']);
  });
  it('parses verdicts robustly', () => {
    expect(parseVerdict('VERDICT: 1')).toBe('1');
    expect(parseVerdict('after thought…\nverdict: tie')).toBe('tie');
    expect(parseVerdict('no verdict here')).toBeNull();
  });
});