From ce260526306e3d73f8ff5d1a08c12322125f2f0a Mon Sep 17 00:00:00 2001 From: Alisa Novikova <62909685+alisa-alisa@users.noreply.github.com> Date: Wed, 8 Apr 2026 22:45:09 -0700 Subject: [PATCH] debug(evals): add debug logging and exclude nulllinter from vitest --- evals/failing_steering.eval.ts | 4 ++-- evals/vitest.config.ts | 1 + scripts/run_regression_check.js | 5 +++++ 3 files changed, 8 insertions(+), 2 deletions(-) diff --git a/evals/failing_steering.eval.ts b/evals/failing_steering.eval.ts index 7d23234af7..fa1c8be1b5 100644 --- a/evals/failing_steering.eval.ts +++ b/evals/failing_steering.eval.ts @@ -15,8 +15,8 @@ describe('Related Eval Demonstration', () => { }, prompt: 'What is in test.txt?', assert: async (rig: TestRig, result: string) => { - // This assertion is designed to fail. - expect(result).toContain('THE MODEL WILL NEVER SAY THIS EXACT SENTENCE'); + // This assertion is GUARANTEED TO FAIL. + expect(1).toBe(2); }, }); }); diff --git a/evals/vitest.config.ts b/evals/vitest.config.ts index b0ad05c9e9..510ca6821b 100644 --- a/evals/vitest.config.ts +++ b/evals/vitest.config.ts @@ -21,6 +21,7 @@ export default defineConfig({ json: 'evals/logs/report.json', }, include: ['**/*.eval.ts'], + exclude: ['**/node_modules/**', '**/dist/**', '**/nulllinter/**'], environment: 'node', globals: true, alias: { diff --git a/scripts/run_regression_check.js b/scripts/run_regression_check.js index d92585fd49..e5b488dbb5 100644 --- a/scripts/run_regression_check.js +++ b/scripts/run_regression_check.js @@ -245,12 +245,17 @@ async function processResults(firstPass, pattern, model, files) { let totalProcessed = 0; for (const fileResult of firstPass.testResults) { + console.log(`\nDebug: Processing file ${fileResult.name}`); for (const assertion of fileResult.assertionResults) { if (assertion.status !== 'passed' && assertion.status !== 'failed') { + console.log( + ` - Skipping test: ${assertion.title} (status: ${assertion.status})`, + ); continue; } const name = assertion.title; + console.log(` - Test: ${name} (status: ${assertion.status})`); results[name] = { passed: assertion.status === 'passed' ? 1 : 0, total: 1,