The Metaphor

Every context layer adds to the input stream. Signal is context aligned with your goal—it amplifies the right outputs. Noise is conflicting context—it drowns out what you want.

The Goal

Maximize your signal-to-noise ratio. The higher the ratio, the more probability mass concentrates on your desired output.

Signal
Noise

Click any layer to see signal vs noise effects →

Click a context layer to see
how it contributes signal or noise

System Context

Signal

What It Is

Platform capabilities, current date, base behavioral rules. The foundational layer you can't edit.

As Signal ✓

When your request matches system capabilities, this layer amplifies valid responses. The system "knows what it can do" and boosts those outputs.

Example

Asking for code explanation when coding is enabled → system context reinforces helpful, accurate responses.

System Context

Noise

What It Is

Platform capabilities, current date, base behavioral rules. The foundational layer you can't edit.

As Noise ✗

When you ask for things outside capabilities, system context generates interference—refusals, hedging, or hallucinated workarounds compete with helpful responses.

Example

"What's happening in the news right now?" without web search → system adds noise by struggling between declining and guessing.

Memory

Signal

What It Is

Persistent knowledge about you—preferences, background, past projects. Carries across sessions.

As Signal ✓

When memory aligns with your current task, it pre-tunes the response. You don't have to re-explain context—it's already amplifying the right direction.

Example

Memory knows you're building a React app. You ask about state management → responses naturally assume React context.

Memory

Noise

What It Is

Persistent knowledge about you—preferences, background, past projects. Carries across sessions.

As Noise ✗

Outdated or context-inappropriate memory creates interference. It pulls responses toward patterns that don't fit your current need.

Example

Memory learned you prefer technical depth. Today you need a simple summary for a non-technical audience → responses keep drifting toward complexity.

Conversation

Signal

What It Is

The accumulated back-and-forth in this session. Every exchange adds to it. You control it.

As Signal ✓

A focused conversation builds compounding signal. Each exchange reinforces shared understanding, making subsequent responses more precisely tuned.

Example

Five exchanges refining a business proposal. By message six, the model deeply understands your goals, audience, and constraints.

Conversation

Noise

What It Is

The accumulated back-and-forth in this session. Every exchange adds to it. You control it.

As Noise ✗

Topic pivots and tangents accumulate as noise. Old context interferes with new requests, pulling responses toward outdated frames.

Example

You discussed vacation planning, then pivoted to debugging code. The casual, planning-oriented tone bleeds into technical responses.

Your Message

Signal

What It Is

The immediate instruction—what you type right now. Your highest-leverage input.

As Signal ✓

A clear, specific message that aligns with upstream context is pure signal. When everything points the same direction, your message has maximum amplification.

Example

Clean conversation history + relevant memory + clear request = your message cuts through cleanly with no competing noise.

Your Message

Noise

What It Is

The immediate instruction—what you type right now. Your highest-leverage input.

As Noise ✗

Even a perfect message becomes noise if it contradicts established context. You're adding signal, but upstream noise drowns it out.

Example

"Be brief" after 10 verbose exchanges. Your instruction adds signal, but it's fighting against the established pattern—net effect is noisy, inconsistent output.

✓ High Signal

Context reinforces intent

System
Strong
Memory
Strong
Conversation
Strong
Your Message
Strong
Combined Output
Clean Output
Predictable, focused
SNR
High

✗ High Noise

Context conflicts with intent

System
Weak
Memory
Weak
Conversation
Weak
Your Message
Weak
Combined Output
Noisy Output
Unpredictable, scattered
SNR
Low