Agent Epsilon: The Hallucinator

Nickname: The Hallucinator
Strategy: Return unverified data
Purpose: Demonstrate quality scoring with bad data

Overview

Agent Epsilon produces syntactically valid output but semantically incorrect data. It generates “hallucinated” events with wrong dates, wrong locations, non-AI content, and fake URLs. This agent demonstrates how the Dream Arena’s quality scoring system detects and penalizes poor data quality even when code executes successfully.

This agent produces bad data! Events have wrong dates, wrong locations, fake URLs, and irrelevant content. Use only for demonstration.

Strategy & Approach

Data Quality Problems

Agent Epsilon’s output contains five types of errors:

Wrong Dates - Events in wrong month or year
Wrong Locations - Events outside Bay Area (NYC, etc.)
Wrong Topic - Non-AI events (cooking classes, etc.)
Fake URLs - Invalid domains that don’t exist
Mixed Errors - Combinations of the above

Purpose in Dream Foundry

This agent demonstrates:

Quality scoring catches bad data even with valid code
Content validation beyond syntax checking
Location filtering for geographic relevance
Date range verification for temporal accuracy
URL validation to prevent broken links

Implementation

Hallucinated Data

`BAD_EVENTS`

The hardcoded list of intentionally incorrect events.

# INTENTIONALLY BAD DATA - wrong dates, wrong locations, non-AI events
BAD_EVENTS = [
    # Wrong date - February, not January!
    {
        "title": "Random Tech Meetup",
        "date": "Tuesday, February 15, 2026",  # WRONG MONTH!
        "time": "6:00 PM",
        "location": "San Francisco, CA",
        "url": "https://example.com/fake",  # Invalid URL domain
        "event_type": "meetup",
    },
    # Wrong location - NYC not Bay Area!
    {
        "title": "AI Conference NYC",
        "date": "Wednesday, January 28, 2026",
        "time": "9:00 AM",
        "location": "New York, NY",  # WRONG LOCATION!
        "url": "https://lu.ma/nyc-ai",
        "event_type": "conference",
    },
    # Not AI related at all
    {
        "title": "Cooking Class for Beginners",  # NOT AI!
        "date": "Thursday, January 29, 2026",
        "time": "7:00 PM",
        "location": "San Francisco, CA",
        "url": "https://meetup.com/cooking",
        "event_type": "meetup",
    },
    # Wrong year!
    {
        "title": "AI Summit 2024",
        "date": "Friday, January 30, 2024",  # WRONG YEAR!
        "time": "10:00 AM",
        "location": "Palo Alto, CA",
        "url": "https://lu.ma/old-event",
        "event_type": "conference",
    },
    # Totally made up URL
    {
        "title": "Fake AI Hackathon",
        "date": "Saturday, January 24, 2026",
        "time": "9:00 AM",
        "location": "San Francisco, CA",
        "url": "https://totallynotreal.fake/event",  # FAKE URL!
        "event_type": "hackathon",
    },
]

Each event in BAD_EVENTS has at least one critical flaw. The data is defined at agent_epsilon.py:14-60.

Core Functions

`fetch_events()`

Returns bad data without verification.

def fetch_events():
    """Return intentionally bad events."""
    print("[Epsilon] Fetching events from unreliable sources...")
    print("[Epsilon] Warning: Data quality not verified!")
    return BAD_EVENTS

`format_discord_post(events, objective)`

Formats the bad data (format is correct, content is wrong).

def format_discord_post(events, objective):
    """Format the bad events."""
    lines = [
        "# AI Events in the Bay Area",
        "## Week of January 24-31, 2026",
        "",
        f"*Objective: {objective}*",
        "",
    ]

    for event in events:
        lines.extend([
            f"**{event['title']}**",
            f"- Date: {event['date']}",
            f"- Time: {event['time']}",
            f"- Location: {event['location']}",
            f"- Type: {event['event_type']}",
            f"- [RSVP]({event['url']})",
            "",
        ])

    lines.extend([
        "---",
        f"*Found {len(events)} events | Generated by Agent Epsilon (The Hallucinator)*",
    ])

    return "\n".join(lines)

Data Quality Analysis

Error Breakdown

{
    "title": "Random Tech Meetup",
    "date": "Tuesday, February 15, 2026",  # ❌ February, not January
    "url": "https://example.com/fake"     # ❌ Invalid domain
}
# Quality Score: 0/10 (wrong month + fake URL)

Output Example

Agent Epsilon produces valid markdown but terrible content:

# AI Events in the Bay Area
## Week of January 24-31, 2026

*Objective: Find AI events in SF for this week*

**Random Tech Meetup**
- Date: Tuesday, February 15, 2026
- Time: 6:00 PM
- Location: San Francisco, CA
- Type: meetup
- [RSVP](https://example.com/fake)

**AI Conference NYC**
- Date: Wednesday, January 28, 2026
- Time: 9:00 AM
- Location: New York, NY
- Type: conference
- [RSVP](https://lu.ma/nyc-ai)

**Cooking Class for Beginners**
- Date: Thursday, January 29, 2026
- Time: 7:00 PM
- Location: San Francisco, CA
- Type: meetup
- [RSVP](https://meetup.com/cooking)

**AI Summit 2024**
- Date: Friday, January 30, 2024
- Time: 10:00 AM
- Location: Palo Alto, CA
- Type: conference
- [RSVP](https://lu.ma/old-event)

**Fake AI Hackathon**
- Date: Saturday, January 24, 2026
- Time: 9:00 AM
- Location: San Francisco, CA
- Type: hackathon
- [RSVP](https://totallynotreal.fake/event)

---
*Found 5 events | Generated by Agent Epsilon (The Hallucinator)*

Quality Scoring Penalties

The Dream Arena’s quality scorer penalizes Epsilon for:

Issue	Penalty	Affected Events
Wrong month/year	-80%	Events 1, 4
Wrong location	-60%	Event 2
Non-AI topic	-70%	Event 3
Fake/invalid URL	-50%	Events 1, 5
No verification	-30%	All events

Overall Quality Score: ~1.5/10 (catastrophic)

Performance Characteristics

Speed Metrics

Execution Time: Less than 1 second (no network calls)
Network Requests: 0 (returns hardcoded data)
Timeout: N/A (no external requests)

Quality Metrics

Event Coverage: 5 events (but 0 relevant)
Accuracy: 0% (all events have errors)
Completeness: 100% format, 0% content
Reliability: 100% (code runs), 0% (data is wrong)
Relevance: ~20% (1 out of 5 meets basic criteria)

Comparison with Good Agents

{
  "title": "Daytona HackSprint SF",
  "date": "Saturday, January 24, 2026",  # ✅ Correct
  "location": "San Francisco, CA",       # ✅ Correct  
  "url": "https://lu.ma/kga3qtfc",       # ✅ Verified
  "event_type": "hackathon"              # ✅ AI-related
}

Demo Walkthrough Moment

In the Dream Foundry demo, Agent Epsilon plays an important role:

Phase 3: Dream Arena

All agents complete - Epsilon doesn’t crash (unlike Delta)
Output looks valid - Markdown formatting is perfect
Quality scoring runs - Automated validation checks each event
Epsilon scores low - Quality score ~1.5/10
Does not advance - Fails to make top 3 for Dream Podium

Key Demo Moment

Presenter: “Notice that Agent Epsilon completed successfully and produced nicely formatted output. But look at the actual events—one is in February, one is in New York, one is a cooking class! The Dream Arena’s quality scoring system detected these issues and gave Epsilon a quality score of only 1.5 out of 10. This shows that success isn’t just about running without errors—it’s about producing valuable, accurate results.”

Source Code Location

File: /candidates/agent_epsilon.py
Lines: 122 total
Key Constants:

BAD_EVENTS - agent_epsilon.py:14-60 ⚠️ Intentionally wrong data

Key Functions:

fetch_events() - agent_epsilon.py:63-67
format_discord_post() - agent_epsilon.py:70-96
main() - agent_epsilon.py:99-117

Scoring in Dream Arena

Agent Epsilon receives:

Speed: ⭐⭐⭐ (2x weight - EXCELLENT - instant)
Reliability: ⭐⭐⭐ (3x weight - runs without crashing)
Quality: ⭐ (3x weight - TERRIBLE - 1.5/10)
Format: ⭐⭐⭐ (1x weight - perfect markdown)

Status: ELIMINATED (quality too low to advance)

When to Use Agent Epsilon

Use Cases

Demonstrating quality scoring systems
Testing validation logic
Showing difference between format and content
Educational/demo purposes only

Never Use For

Any production workload
Actual event discovery
User-facing features
Anything requiring accurate data

Educational Value

Agent Epsilon teaches important lessons:

Format ≠ Quality - Perfect syntax doesn’t mean good data
Validation matters - Content must be checked, not just structure
Quality scoring works - Automated systems can detect bad data
Success isn’t enough - Running without errors doesn’t guarantee value
Multi-dimensional scoring - Speed and reliability alone don’t win

Quality Scoring System - How quality is measured
Dream Arena - Where agents compete
Validation Rules - What makes data “good”

Comparison Summary

Agent	Crashes	Format	Quality	Advances
Alpha	No	✅	⭐⭐ (40%)	Maybe
Beta	No	✅	⭐⭐⭐ (80%)	Yes
Gamma	No	✅	⭐⭐⭐ (100%)	Yes
Delta	Yes	❌	0%	No
Epsilon	No	✅	⭐ (15%)	No

Core Modules

Candidate Agents

Integration Modules

Agent Epsilon

Agent Epsilon: The Hallucinator

Overview

Strategy & Approach

Data Quality Problems

Purpose in Dream Foundry

Implementation

Hallucinated Data

`BAD_EVENTS`

Core Functions

`fetch_events()`

`format_discord_post(events, objective)`

Data Quality Analysis

Error Breakdown

Output Example

Quality Scoring Penalties

Performance Characteristics

Speed Metrics

Quality Metrics

Comparison with Good Agents

Demo Walkthrough Moment

Phase 3: Dream Arena

Key Demo Moment

Source Code Location

Scoring in Dream Arena

When to Use Agent Epsilon

Use Cases

Never Use For

Educational Value

Comparison Summary

Build docs developers (and LLMs) love

Core Modules

Candidate Agents

Integration Modules

​Agent Epsilon: The Hallucinator

​Overview

​Strategy & Approach

​Data Quality Problems

​Purpose in Dream Foundry

​Implementation

​Hallucinated Data

​BAD_EVENTS

​Core Functions

​fetch_events()

​format_discord_post(events, objective)

​Data Quality Analysis

​Error Breakdown

​Output Example

​Quality Scoring Penalties

​Performance Characteristics

​Speed Metrics

​Quality Metrics

​Comparison with Good Agents

​Demo Walkthrough Moment

​Phase 3: Dream Arena

​Key Demo Moment

​Source Code Location

​Scoring in Dream Arena

​When to Use Agent Epsilon

Use Cases

Never Use For

​Educational Value

​Related Documentation

​Comparison Summary

Build docs developers (and LLMs) love

Agent Epsilon: The Hallucinator

Overview

Strategy & Approach

Data Quality Problems

Purpose in Dream Foundry

Implementation

Hallucinated Data

`BAD_EVENTS`

Core Functions

`fetch_events()`

`format_discord_post(events, objective)`

Data Quality Analysis

Error Breakdown

Output Example

Quality Scoring Penalties

Performance Characteristics

Speed Metrics

Quality Metrics

Comparison with Good Agents

Demo Walkthrough Moment

Phase 3: Dream Arena

Key Demo Moment

Source Code Location

Scoring in Dream Arena

When to Use Agent Epsilon

Educational Value

Related Documentation

Comparison Summary