Baseline

How BCI systems learn "normal" brain patterns for each person

Purpose

The personal baseline is the foundation of responsible BCI data systems. It is the learned model of an individual's typical brain activity patterns across different states (focused, relaxed, asleep, etc.).

Without a stable baseline, the system cannot provide meaningful insights.

This document defines:

• What a baseline is
• Why it's mandatory
• How long it takes to establish
• What can (and cannot) be done during the learning period

What is a Personal Baseline?

A personal baseline is a statistical model of your typical brain activity built from 30-90 days of EEG data across various states and times of day.

Components

A complete BCI baseline includes:

Why Personal?

Population averages are meaningless for BCI data:

• One person's "high alpha" (60 μV) = another person's "low alpha" (20 μV)
• Meditation experience changes baseline brain patterns
• Age, genetics, and neural architecture create huge individual differences
• Device placement and calibration add person-specific variations

The only valid comparison is YOU vs. YOUR OWN BASELINE.

Baseline Learning Timeline

Phase 1: Initial Data Collection (Days 1-30)

Status: LEARNING
System behavior: Silent observation, no neurofeedback or recommendations

Day 1-7:   📊 Collecting initial patterns
Day 8-14:  📊 Observing state variations
Day 15-21: 📊 Building preliminary model
Day 22-30: 📊 Refining baseline estimates

User sees: "Building your personal brain pattern baseline. This takes 30-90 days."

System does:

• Collects raw EEG across different times of day
• Detects and categorizes different states (focus, relaxation, rest)
• Calculates preliminary statistical models
• BUT: Provides no pattern-based feedback yet (insufficient data)

Phase 2: Baseline Stabilization (Days 31-90)

Status: STABILIZING
System behavior: Early, low-confidence feedback begins

Day 31-45: 📊 Baseline becoming more stable
Day 46-60: 📊 Confidence increasing
Day 61-75: 📊 Nearly stable
Day 76-90: ✅ Baseline stable

User sees: "Your baseline is forming. Early insights may be available, but confidence is still building."

System does:

• Continues refining baseline models
• May provide gentle, low-confidence pattern observations
• Still learning individual variability and state transitions

Phase 3: Stable Baseline (Day 90+)

Status: STABLE
System behavior: Full neurofeedback and pattern recognition active

User sees: Normal neurofeedback and pattern observations

System does:

• Compares current brain activity to well-established baseline
• Provides high-confidence neurofeedback during meditation/focus
• Detects meaningful deviations from typical patterns
• Continuously updates baseline as new data arrives (rolling window)

Baseline Requirements Before Feedback

Hard Requirement

All agent-generated insights must check baseline status:

function generateNeurofeedback(user, eegData) {
  // MANDATORY GATE
  if (user.bciBaselineStatus !== 'STABLE') {
    return null; // No feedback during learning phase
  }
  
  // Only after baseline is stable can we proceed
  const alphaDeviation = compareToBaseline(eegData, user.personalBaseline);
  
  if (significantDeviation(alphaDeviation)) {
    return generatePatternObservation(alphaDeviation, user);
  }
}

Why This Is Non-Negotiable

Without baseline:

• No context for whether a brainwave value is high/low/normal for this person
• Cannot distinguish signal from noise
• Risk of false positives (claiming relaxation when it's just that person's typical state)
• Population comparisons would be used instead (forbidden)

Baseline Calculation Methodology

Statistical Approach

For each frequency band (alpha, beta, theta, delta):

1. Collect: 30-90 days of data across states
2. Segment: Separate by state (focused, relaxed, resting, etc.)
3. Calculate: Mean, median, standard deviation, percentiles for each state
4. Model: Distribution of typical values (not just a single number)
5. Update: Rolling window update as new data arrives

Example: Alpha Band Baseline

user.personalBaseline.alpha = {
  restingState: {
    mean: 42.3,
    stdDev: 8.1,
    p25: 36.2,
    median: 41.8,
    p75: 48.1,
    dataPoints: 120 // from 30+ sessions
  },
  focusedState: {
    mean: 28.5,
    stdDev: 6.3,
    p25: 24.1,
    median: 28.0,
    p75: 32.7,
    dataPoints: 85
  },
  relaxedState: {
    mean: 56.7,
    stdDev: 11.2,
    p25: 48.3,
    median: 55.9,
    p75: 64.2,
    dataPoints: 95
  }
};

State Detection

The baseline must include state models—brain patterns typical of different activities:

• Focused: High beta, moderate alpha, low theta
• Relaxed/meditation: High alpha, moderate theta, low beta
• Drowsy: High theta, moderate alpha
• Deep sleep: High delta, low everything else

This enables pattern similarity matching: "Your current pattern resembles your past focused states."

Minimum Data Requirements

If minimum requirements not met: System remains in LEARNING status and provides no feedback.

Handling Special Cases

Case 1: New Meditation Practice

If a user starts meditating regularly during baseline learning:

Problem: Meditation changes brain patterns (increases alpha, shifts baseline)

Solution:

• Detect the shift (new pattern emerging)
• Extend baseline period to capture new equilibrium
• Segment data (pre-meditation baseline vs. post-meditation baseline)

Case 2: Lifestyle Changes

Major life changes (new job, travel, illness) can shift baselines:

Solution:

• Allow baseline to adapt (rolling window update)
• If dramatic shift detected, flag for baseline recalibration
• Never lock baseline permanently—always evolving

Case 3: Illness or Medication

Brain patterns may change during illness or medication use:

Solution:

• Do NOT interpret as meaningful deviations
• Display: "Your brain patterns have changed significantly from baseline. This might be due to illness, medication, or other factors. Patterns may normalize when you return to your usual state."
• Do NOT diagnose or make medical claims

Baseline Updates (After Initial Establishment)

Once a stable baseline exists (90+ days), it should continuously update using a rolling window:

Rolling Window Update

const BASELINE_WINDOW = 90 days;
 
function updateBaseline(user, newEegData) {
  // Add new data
  user.baselineData.push(newEegData);
  
  // Remove data older than window
  user.baselineData = user.baselineData.filter(d => 
    d.timestamp > now() - BASELINE_WINDOW
  );
  
  // or recalculate statistics
  user.personalBaseline = calculateBaselineStats(user.baselineData);
}

Why Continuous Updates?

• Brain patterns evolve with age, lifestyle, meditation practice
• Seasonal variations (winter vs. summer)
• Device recalibration or replacement
• Long-term stability requires adaptation

Communicating Baseline Status to Users

During Learning Phase

Good:

Building Your Brain Pattern Baseline

We're learning your unique brain patterns across different states 
(focused, relaxed, resting). This takes 30-90 days.

Progress: Day 24 of 90
  - Focused state: 18 sessions recorded ✓
  - Relaxed state: 14 sessions recorded
  - Sleep: 21 nights recorded ✓

Keep using the device regularly to improve baseline accuracy.

Bad: ❌ "Your brain is still calibrating" (sounds mechanical/diagnostic)
❌ "Baseline creation: 26% complete" (progress bar implies it's just processing, not learning individual patterns)

When Baseline is Stable

Good:

Baseline Established ✓

We've learned your personal brain patterns! 

Now you'll receive neurofeedback when your brain activity 
shows patterns similar to your focused or relaxed states.

Your baseline will continue updating as you use the device.

Developer Checklist

Before implementing any BCI feature that provides user insights:

• Does it check user.bciBaselineStatus === 'STABLE' before activating?
• Does it compare to personal baseline, not population norms?
• Does it require minimum data thresholds (30+ days, 15+ sessions per state)?
• Does the baseline update continuously (rolling window)?
• Are users informed when they're in the learning phase?
• Is there a clear communication about why baseline is necessary?

Summary

Personal baseline is mandatory because:

1. Brain patterns vary hugely between individuals
2. Population averages are meaningless
3. Without baseline, "high alpha" or "low beta" has no context
4. Comparing to self is the only valid approach

Baseline requirements:

• 30-90 days of data
• Multiple states (focused, relaxed, resting, sleep)
• Continuous rolling updates
• No feedback until baseline is stable

Rule: If baselineStatus !== 'STABLE', no pattern-based insights. Period.