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.
Click any layer to see signal vs noise effects →
Click a context layer to see
how it contributes signal or noise
System Context
SignalWhat 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
NoiseWhat 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
SignalWhat 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
NoiseWhat 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Outdated or context-inappropriate memory creates interference.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
SignalWhat 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
NoiseWhat 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
SignalWhat 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
NoiseWhat 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
✗ High Noise
Context conflicts with intent