Deviation Engine

When and why BCI neurofeedback systems activate

Purpose

The deviation engine determines when a BCI system provides neurofeedback or pattern observations to the user.

Core principle: Only notify when brain activity meaningfully deviates from personal baseline—not for every fluctuation.

This document defines:

• When neurofeedback is appropriate
• How to detect meaningful deviations
• What "meaningful" means for brain patterns
• When to stay silent

The Core Question

"Should the system say something to the user right now?"

This depends on:

1. Is baseline stable? (30-90 days of data)
2. Is current pattern different enough from baseline to matter?
3. Is the deviation sustained (not just noise)?
4. Can the user take any action based on this information?
5. Is this the right time to provide feedback?

If any answer is NO → stay silent.

Deviation Detection Logic

Step 1: Baseline Gate

function shouldProvideFeedback(user, eegData) {
  // MANDATORY FIRST CHECK
  if (user.bciBaselineStatus !== 'STABLE') {
    return false; // Silent during learning phase (30-90 days)
  }
  
  // Continue to step 2...
}

No baseline = no feedback. Ever.

Step 2: Calculate Deviation from Personal Baseline

For each frequency band, compare current activity to personal baseline:

const alphaDeviation = {
  current: eegData.alphaPower,
  baseline: user.personalBaseline.alpha.currentState.median,
  ratio: eegData.alphaPower / user.personalBaseline.alpha.currentState.median,
  percentChange: ((eegData.alphaPower / user.personalBaseline.alpha.currentState.median) - 1) * 100,
  zScore: (eegData.alphaPower - user.personalBaseline.alpha.currentState.mean) / 
          user.personalBaseline.alpha.currentState.stdDev
};

Step 3: Define "Meaningful" Deviation

Not every deviation warrants feedback. Use thresholds:

Also consider z-scores:

• z < 1.0: Within normal variation
• z = 1.0-2.0: Moderate deviation
• z > 2.0: Significant deviation

Step 4: Temporal Consistency Requirement

Single-moment spikes are noise. Require sustained patterns:

function isPatternSustained(eegStream, deviationThreshold, minDuration) {
  const windowSize = minDuration; // e.g., 10 seconds
  const samples = eegStream.last(windowSize);
  
  // Check if >80% of samples in window exceed threshold
  const exceedingThreshold = samples.filter(s => 
    Math.abs(s.deviation) > deviationThreshold
  );
  
  return (exceedingThreshold.length / samples.length) > 0.8;
}

Minimum sustain durations:

• Real-time neurofeedback: 3-5 seconds
• Pattern notifications: 10-30 seconds
• State change detection: 60+ seconds

Step 5: Actionability Filter

Only provide feedback if the user can do something with the information.

Deviation Categories and Responses

Category 1: Positive Reinforcement (Neurofeedback)

When: User is meditating/practicing and achieves desired brain state

Trigger:

• Alpha power >20% above baseline (meditation)
• Beta/theta ratio matches "focused" baseline state (concentration practice)
• Pattern sustained for 5+ seconds

Response:

return {
  type: "positive_feedback",
  message: "Nice! Your alpha waves are elevated—you're entering a relaxed state.",
  visual: "calm_indicator", // e.g., green light, expanding circle
  tone: "encouraging"
};

Purpose: Real-time reinforcement to help user learn what the desired state feels like.

Category 2: Pattern Observation (Informational)

When: User's brain pattern deviates significantly from typical in a noticeable way

Trigger:

• Any frequency band >30% different from baseline
• Pattern sustained for 30+ seconds
• Outside active neurofeedback session (general awareness)

Response:

return {
  type: "pattern_update",
  message: "Your beta wave activity has been elevated compared to your recent baseline. This pattern sometimes appears during focused or alert states.",
  actionSuggestion: null, // Informational only
  tone: "neutral"
};

Purpose: Help user notice patterns they might not be aware of.

Category 3: State Change Detection

When: Brain activity transitions from one characteristic state to another

Trigger:

• Pattern similarity shifts from "focused baseline" to "relaxed baseline" (or vice versa)
• Transition sustained for 60+ seconds

Response:

return {
  type: "state_transition",
  message: "Your brain pattern has shifted from your typical focused state toward a more relaxed pattern.",
  implication: "If you're trying to concentrate, this might be a sign to take a break or re-engage.",
  tone: "informative"
};

Purpose: Awareness of natural state transitions.

Category 4: Anomaly / Sensor Issue

When: Extreme or impossible values that suggest sensor problems

Trigger:

• Any frequency band >100% different from baseline
• Flatline (all zeros)
• Extreme amplitude (sensor disconnection)
• Multiple bands simultaneously at extremes

Response:

return {
  type: "sensor_check",
  message: "We're seeing an unusual pattern that might be due to sensor placement or connection. Try adjusting the device.",
  possibleCauses: ["Sensor not making good contact", "Device shifted position", "Electrical interference"],
  tone: "calm_troubleshooting"
};

Purpose: Help user troubleshoot hardware issues without alarming them.

When to Stay Silent

The system should NOT provide feedback when:

1. Baseline Not Stable

• User has <30 days of data
• Learning individual patterns still ongoing

2. Deviation Too Small

• <15% change from baseline
• Within normal fluctuation range

3. Pattern Not Sustained

• Brief spike or dip (<3 seconds)
• Intermittent pattern (not consistent)

4. User Cannot Act

• During sleep
• While driving
• Not engaged with the app

5. Too Frequent

• Last feedback was <5 minutes ago
• Risk of notification fatigue

6. State is Expected

• Brain patterns match current activity (focused during work = expected)
• No deviation from baseline for current context

Deviation Engine Configuration

Sensitivity Settings

Allow users to control feedback frequency:

Context-Aware Activation

const deviationRules = {
  duringMeditation: {
    enabled: true,
    threshold: 0.15, // 15% deviation
    sustainDuration: 5, // seconds
    frequencyLimit: "none" // Real-time ok
  },
  
  duringGeneralUse: {
    enabled: true,
    threshold: 0.30, // 30% deviation
    sustainDuration: 30,
    frequencyLimit: "max_1_per_hour"
  },
  
  duringSleep: {
    enabled: false // Never interrupt sleep
  },
  
  duringCriticalTasks: {
    enabled: false // Never distract during driving, etc.
  }
};

Multi-Signal Confirmation

Never trigger feedback from a single signal. Require multiple confirming signals:

Example: Meditation State Detection

function detectMeditationState(eeg, user) {
  const alphaElevated = (eeg.alphaPower / user.baseline.alpha.median) > 1.20;
  const betaReduced = (eeg.betaPower / user.baseline.beta.median) < 0.85;
  const thetaModerate = (eeg.thetaPower / user.baseline.theta.median) > 0.90;
  
  const meditationScore = 
    (alphaElevated ? 40 : 0) +
    (betaReduced ? 30 : 0) +
    (thetaModerate ? 30 : 0);
  
  // Require combined evidence, not just one signal
  return meditationScore > 60; // At least 2 of 3 signals present
}

Deviation Presentation to Users

For Real-Time Neurofeedback

Visual indicator preferred over text (less distracting during meditation):

{
  visual: "expanding_circle", // Size corresponds to alpha power
  color: "gradient_blue_to_green", // Calm palette
  audio: "soft_chime", // Optional subtle sound
  text: null // Minimal or no text during active practice
}

For Pattern Observations

Text notification appropriate:

{
  title: "Brain Pattern Update",
  body: `Your beta wave activity has been higher than your recent baseline. 
         This pattern sometimes appears during alert or focused states.`,
  timestamp: new Date(),
  dismissable: true
}

Avoiding Notification Fatigue

Frequency Limits

const notificationLimits = {
  maxPerHour: 2, // General pattern updates
  maxPerDay: 10,
  minIntervalBetween: 15 * 60 // 15 minutes
};
 
function shouldAllowNotification(user) {
  const recentNotifications = user.notifications.filter(n => 
    n.timestamp > now() - 1hour
  );
  
  return recentNotifications.length < notificationLimits.maxPerHour;
}

Smart Grouping

If multiple deviations detected in short time:

// Instead of 3 separate notifications:
// ❌ "Alpha elevated"
// ❌ "Beta reduced"
// ❌ "Theta moderate"
 
// Combine into one:
✅ "Your brain pattern has shifted toward a relaxed, meditative state 
    (elevated alpha, reduced beta). This is similar to your past 
    meditation sessions."

Testing Deviation Logic

Test Cases

1. Baseline not stable: Should return null regardless of current patterns
2. Noise (5% deviation): Should ignore
3. Brief spike (25% for 1 second): Should ignore (not sustained)
4. Sustained deviation (25% for 20 seconds): Should trigger observation
5. Sensor disconnection (flatline): Should suggest sensor check, not pattern observation
6. Meditation state (multi-signal confirmation): Should provide real-time feedback

Example Test

describe('Deviation Engine', () => {
  it('should not provide feedback without stable baseline', () => {
    const user = { bciBaselineStatus: 'LEARNING' };
    const eeg = { alphaPower: 100 }; // Extreme deviation
    
    const result = deviationEngine.check(user, eeg);
    expect(result).toBeNull(); // No feedback during learning
  });
  
  it('should ignore brief fluctuations', () => {
    const user = { bciBaselineStatus: 'STABLE', baseline: { alpha: { median: 50 } } };
    const eeg = [
      { alphaPower: 65, timestamp: 0 },    // 30% spike
      { alphaPower: 52, timestamp: 1 },    // Back to baseline
    ];
    
    const result = deviationEngine.check(user, eeg);
    expect(result).toBeNull(); // Not sustained
  });
});

Summary

The deviation engine activates when:

1. ✅ Baseline is stable (30-90 days)
2. ✅ Deviation is meaningful (>15-30%)
3. ✅ Pattern is sustained (5-30 seconds)
4. ✅ User can take action
5. ✅ Not too frequent (prevents fatigue)

Purpose: Provide signal when the signal matters—stay quiet when it doesn't.

Philosophy: Less is more. Quality over quantity. Only speak when you have something useful to say.