Signals

What BCI data the system reads — and what it cannot tell us.

Purpose

This document defines the exact data inputs available to consumer BCI systems. Every signal listed here comes from consumer neurotechnology devices (EEG headbands, neurofeedback systems, etc.). No signal in this list constitutes a medical measurement.

A developer reading this page should know:

1. What can be read from the hardware
2. What can be derived with reasonable confidence
3. What must never be inferred from this data

Supported Signals

Signal Characteristics

Pattern-based, not state-reading

Brainwave signals show patterns, not mental states. High alpha waves correlate with relaxation in research—but cannot definitively prove "you are relaxed right now."

Personal baseline required

A frequency band value (e.g., "alpha power = 45 microvolts") means nothing without context. The same value could indicate calm for one person and agitation for another. Only personal baseline comparisons are meaningful.

Noisy and artifact-prone

Consumer EEG is extremely sensitive to:

• Electrode placement and contact quality
• Muscle tension (jaw clenching, eyebrow movement)
• Eye movements and blinks
• Electrical noise (phones, computers, power lines)
• Movement and position changes

Single-moment readings are noise. Systems must average over time windows and require multi-day baselines.

Delayed and indirect

Brain signals don't directly encode thoughts. They reflect underlying neural activity that correlates with—but doesn't causally determine—mental states and behaviors.

Individual variability

Research shows population-level correlations (e.g., "alpha increases during meditation"), but individual patterns vary widely. Some people have high baseline alpha; others low. Population norms are not decision inputs.

What BCI Signals Cannot Tell Us

Population Averages Are Not Decision Inputs

The system explicitly rejects population averages as primary decision criteria.

Why:

• "Normal alpha range" varies by age, genetics, meditation experience, device calibration
• Telling someone their beta is "low" (compared to population) creates anxiety without actionable insight
• The only meaningful comparison is you vs. your own baseline

Rule: Population data may be used for educational context ("research shows alpha often increases during meditation") but must never drive agent recommendations or pattern notifications.

Signal Processing Requirements

Artifact Detection

Before using EEG data for neurofeedback, filter out:

• Eye blinks (sharp, large amplitude signals)
• Muscle tension (high-frequency noise)
• Movement artifacts (sudden large deflections)
• Electrode disconnection (flatline or extreme amplitude)

Time Windows

• Real-time neurofeedback: 1-5 second windows (with artifact rejection)
• Pattern recognition: 10-60 second windows
• Baseline calculation: 30-90 days of daily recordings

Multi-Signal Confirmation

Never trigger high-confidence feedback from a single signal. Require:

• Multiple frequency bands (e.g., alpha + theta for meditation)
• Temporal consistency (pattern sustained for X seconds)
• Baseline deviation threshold (e.g., >20% change)

BCI-Specific Data Types

Raw EEG

What it is: Voltage measurements from scalp electrodes (~10-100 microvolts)

What it's good for: Computing frequency bands, detecting artifacts, research

What it's NOT: Readable thought or emotion data

Frequency Band Power

What it is: Amount of activity in alpha/beta/theta/delta ranges

What it's good for: Comparing to personal baseline, neurofeedback

What it's NOT: Absolute indicators of mental states

Derived Metrics (Focus, Relaxation, etc.)

What they are: Algorithmic combinations of frequency bands and patterns

What they're good for: Simplified feedback to users (when based on personal baseline)

What they're NOT: Cognitive ability scores or medical assessments

Consumer Devices vs. Medical-Grade EEG

Implication: Consumer BCI data is fundamentally different from clinical EEG. Systems must not claim medical-grade accuracy.

Developer Guidance

DO

• Compare signals to the user's personal baseline (30-90 days)
• Require multi-signal, multi-timepoint confirmation before high-confidence feedback
• Treat single-moment readings as noise
• Display trends and patterns, not absolute values
• Provide artifact rejection and signal quality indicators
• Explain that EEG shows patterns, not thoughts

DON'T

• Use population "normal ranges" for decision-making
• Generate insights without stable personal baseline
• Show raw numbers without context (e.g., "alpha = 45 μV" means nothing alone)
• Claim to know thoughts, emotions, or cognitive states with certainty
• Present consumer EEG as medically accurate
• Use brain data to assess intelligence or make diagnoses

Signal-to-Insight Pathway

HARDWARE
  ↓
Raw EEG voltage
  ↓
Artifact rejection & filtering
  ↓
Frequency band extraction (alpha, beta, theta, delta)
  ↓
Personal baseline comparison (requires 30-90 days)
  ↓
Pattern matching (similarity to past states)
  ↓
Multi-signal confirmation
  ↓
NEUROFEEDBACK / OBSERVATION
  ↓
User sees: "Your alpha pattern is similar to past relaxed states"
  ↓
User DOES NOT see: "You are relaxed" (too definitive)

Privacy Considerations for Neural Signals

Brain data is the most sensitive biometric.

Why it's sensitive:

• Reflects mental activity (even if we can't read specific thoughts)
• Could potentially reveal medical conditions (epilepsy signatures)
• Highly personal (everyone's baseline is different)
• Permanent biometric (brain patterns are stable over time)

Required protections:

• Explicit informed consent before any EEG collection
• End-to-end encryption for storage and transmission
• User control (delete all data anytime)
• No third-party sharing without explicit permission
• Minimal retention (30-90 days for baseline, then rolling deletion)

Example: Alpha Wave Interpretation

❌ Wrong Interpretation

if (alphaPower > 50) {
  return "You are relaxed and calm."; // Definitive mental state claim
}

Problems:

• Uses absolute threshold (population norm)
• Claims certainty about mental state
• Ignores personal baseline

✅ Correct Interpretation

if (user.bciBaselineStatus !== 'STABLE') {
  return null; // No feedback without baseline
}
 
const alphaRatio = alphaPower / user.personalBaseline.alphaPower;
 
if (alphaRatio > 1.2 && sustainedFor > 5seconds) {
  return {
    message: "Your alpha wave activity is elevated compared to your recent baseline. This pattern is similar to your past meditation sessions.",
    disclaimer: "Based on your personal brain patterns. Brain signals show patterns, not definitive mental states."
  };
}

Why this is correct:

• Requires personal baseline
• Uses relative comparison (ratio, not absolute)
• Temporal consistency check (sustained pattern)
• Observational language ("pattern is similar to")
• Disclaimer about limitations

Summary

BCI signals provide:

• ✅ Brain activity patterns (frequency bands)
• ✅ Similarity to past personal states
• ✅ Trends over time (with baseline)
• ✅ Neurofeedback for meditation/focus practice

BCI signals do NOT provide:

• ❌ Specific thoughts or mental content
• ❌ Definitive emotion states
• ❌ Cognitive ability measurements
• ❌ Medical diagnoses
• ❌ Lie detection or intention reading

Bottom line: These signals are pattern observation tools for wellness neurofeedback, not mind-reading devices or medical instruments.