Skip to content

WorldLine

WorldLine is counterfactual debugging + institutional memory for multi-agent AI reliability.

Code Demo Video Cerebral Valley

Core trick

Forks a multi-agent run at each decision, injects a counterfactual output, and re-simulates downstream to find which decision flips the result.

Tags

dev toolsagents

Description

WorldLine is counterfactual debugging + institutional memory for multi-agent AI reliability. When an agent pipeline returns the wrong answer, WorldLine has Claude intervention-test every decision in parallel — forking the run, injecting a counterfactual, and re-simulating the downstream live — to find the decision that actually CAUSED the failure, ruling out plausible decoys (even the last-touch agent that looks guilty). It then authors a repair and proves it flips the outcome with a deterministic code assertion. Every verified fix becomes a durable lesson in a fleet knowledge graph. A different agent about to repeat the same failure class — even worded differently — is caught and fixed pre-emptively, and a pre-ship CI gate blocks any change that would reintroduce a known failure. Tracing shows what happened. WorldLine finds the cause, proves the fix, and makes your agent fleet stop repeating itself.
Murad-Huseynli/claude-build-day★ 00 forksTypeScriptCreated Jun 13, 2026 · last push Jun 14, 2026

README excerpt

<div align="center">

# 🌐 WorldLine

### Counterfactual debugging **+ institutional memory** for multi-agent AI reliability

**Tracing shows what _happened_. WorldLine finds the decision that _caused_ the failure, proves the fix, and makes your agent fleet stop repeating it.**

[![Live Demo](https://img.shields.io/badge/▶_Live_Demo-claude--build--day--alpha.vercel.app-5b8cff?style=for-the-badge)](https://claude-build-day-alpha.vercel.app)
&nbsp;
[![1-min Video](https://img.shields.io/badge/▶_1--min_Demo-YouTube-ff5a52?style=for-the-badge&logo=youtube&logoColor=white)](https://youtu.be/2jSkWQZ-EOE)
&nbsp;
[![Opus 4.8](https://img.shields.io/badge/Built_with-Claude_Opus_4.8-3ddc84?style=for-the-badge)](https://www.anthropic.com)

<sub>Built at **Claude Build Day 2026** · `claude-opus-4-8` · Next.js 16 + react-three-fiber on Vercel</sub>

<br/>

![WorldLine — counterfactual fork: the failing run flips red→green when the culprit decision is corrected](./docs/simulator.png)

</div>

---

## The problem

Multi-agent systems fail in non-deterministic, multi-step ways. When a 7-step pipeline returns the wrong answer, *which decision* caused it? Today you read traces and guess. Tracing shows **what happened**; replay **repeats** it; forking lets you **explore one alternate path** — but you still pick the checkpoint by hand, and **nothing proves the fix**. And when the fix lands, nothing stops a *different* agent from shipping the same class of bug next week.

## The reliability lifecy

Team