Skip to content

😱 Amygdala — Emotional Valence

Biological Analog: The amygdala is the brain's emotional processor. It assigns emotional significance to experiences — fear, joy, anger, relief — which profoundly influences how memories are encoded, stored, and retrieved. Emotionally charged memories are remembered more vividly and last longer.

The Concept

Every memory in Spector carries a valence score — a single byte (-128 to +127) representing its emotional coloring:

Range Meaning Examples
-128 to -50 Strongly negative Critical errors, data loss, security breaches
-50 to -10 Mildly negative Warnings, slow performance, minor bugs
-10 to +10 Neutral Factual information, routine operations
+10 to +50 Mildly positive Successful deployments, optimizations
+50 to +127 Strongly positive Major breakthroughs, user praise, goals achieved

How It Works

The valence tracker computes emotional coloring from two signals:

flowchart TD
    TEXT["Memory text content"] --> SENTIMENT["Content-based sentiment<br/><i>keyword analysis</i>"]
    SOURCE["Memory source type"] --> BIAS["Source-based bias<br/><i>e.g., errors → negative</i>"]

    SENTIMENT --> FUSE["Fused valence score<br/><b>clamped to [-128, +127]</b>"]
    BIAS --> FUSE

    FUSE --> NEG["Negative valence<br/><i>errors, failures, warnings</i>"]
    FUSE --> ZERO["Neutral valence<br/><i>factual, routine</i>"]
    FUSE --> POS["Positive valence<br/><i>successes, breakthroughs</i>"]

    style NEG fill:#e74c3c,color:white
    style ZERO fill:#95a5a6,color:white
    style POS fill:#2ecc71,color:white

Valence-Filtered Recall

The most powerful use of valence is in recall filtering. Agents can filter by emotional range to answer different types of questions:

"What went wrong?" — Negative Memories

memory.recall("database connection",
    topK: 10,
    maxValence: -10,       // Only negative memories
    tags: ["database", "error"])

"What worked well?" — Positive Memories

memory.recall("deployment strategy",
    topK: 5,
    minValence: +10,       // Only positive memories
    tags: ["deployment"])

Full Emotional Range (Default)

By default, no valence filter is applied — the agent sees the full emotional spectrum. The valence still influences recall indirectly because the flashbulb policy pins emotionally intense memories at higher importance.

Where It Fits in the Pipeline

Valence filtering happens at Phase 3 of the 6-phase scorer — before the expensive SIMD vector math:

flowchart LR
    P1["Phase 1<br/>Tombstone"] --> P2["Phase 2<br/>Tag Gate"]
    P2 --> P3["Phase 3<br/><b>Valence Filter</b><br/><i>~2 cycles</i>"]
    P3 --> P4["Phase 4<br/>Importance"]
    P4 --> P5["Phase 5<br/>SIMD L2<br/><i>~200 cycles</i>"]
    P5 --> P6["Phase 6<br/>Fused Score"]

    style P3 fill:#e74c3c,color:white
    style P5 fill:#0984e3,color:white

Cost: 2 CPU cycles — a single byte read and two comparisons. Records outside the valence range are eliminated before Phase 5's ~200-cycle SIMD computation.

Storage

Valence is stored in the 64-byte synaptic header as a single signed byte:

Offset 30: [1B valence] — signed byte [-128 to +127]

This costs exactly 1 byte per memory — negligible overhead for a powerful filtering dimension.

Next Steps