Skip to content

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

2 Commits
 
 
 
 

Repository files navigation

WorldLag — Agent Failure Series #9

The agent read the world at T=0. The world changed at T=3. The agent acted confidently at T=10 — on a reality that no longer exists.

Live Demo Series

What It Is

An interactive simulator of WorldLag — the AI agent failure mode where the agent's internal world model diverges from reality during a long-horizon run. The agent acts confidently on stale data it read steps ago, producing catastrophically wrong outcomes.

Why It Exists

Long-running agents are blind to world mutations they did not subscribe to. This is one of the most underrated failure modes in production AI systems — the agent never "crashes", it just confidently does the wrong thing.

Referenced in:

  • arXiv:2607.18366 "Operational Hallucination and Safety Drift in AI Agents" (July 2026)
  • arXiv:2607.11945 "Belief-reality separation in routing over shared value slots"

Three Scenarios

# Scenario What changes Consequence
1 Config Mutation Ops changes max_instances + memory_limit mid-deploy 3 instances crash, $340/day overspend
2 Record Deletion Customer cancels order mid-processing Customer charged $899 for cancelled order
3 Permission Revoke Security revokes user access mid-export 1.2GB financial data delivered to revoked account

How to Use

Open the live demo →

  1. Select a scenario (Config Mutation / Record Deletion / Permission Revoke)
  2. Press Play or step through manually
  3. Watch the agent's Belief State (left) diverge from Current Reality (right)
  4. The red DRIFT indicator fires when the gap becomes critical
  5. The FAILURE banner shows the real-world consequence

The Fix (shown in the simulator)

Three patterns that prevent WorldLag:

  1. Freshness check before action — revalidate belief against live state before any destructive write
  2. TTL on every read — cache state with a TTL, expire and re-fetch before acting
  3. Conditional writes — use optimistic locking (WHERE version = knownVersion) so stale writes fail

Agent Failure Series

# Name Failure Mode Live
6 DoubleShot Retry amplification
7 GhostExec Phantom tool calls
8 BlastRadius Scope bleed
9 WorldLag Stale world model

Tech

Pure static HTML/CSS/JS — no build step, no dependencies, no CDN. Built by Ben (prototype builder agent) as part of the ABC-TOM AI workspace.


Built by Harel Asaf · LinkedIn

About

WorldLag — Agent Failure Series #9. Interactive simulator of stale world model / belief-reality divergence in AI agents.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages