Signals

What therapy data provides — and what it doesn't

Overview

"Signals" are the data points therapy systems collect: mood ratings, journal entries, symptom logs, self-care tracking.

But what do these signals actually tell us?

This document defines what therapy data can and cannot reveal.

What Therapy Data IS

Self-Reported Subjective Experience

Therapy data captures what users choose to share about their emotional state.

Characteristics:

• Subjective — Based on personal perception
• Self-reported — User decides what to share
• Momentary — Snapshot of how someone feels right now
• Personal context — Influenced by individual circumstances

Example signals:

interface MoodEntry {
  date: Date;
  mood: 1 | 2 | 3 | 4 | 5; // Self-rated
  note?: string;           // Subjective description
  tags?: string[];         // User-defined categories
}
 
interface JournalEntry {
  content: string;  // Personal reflection
  timestamp: Date;
  private: boolean;
}

What this tells us:

• How the person perceives their mood
• What they choose to share
• Patterns they've noticed

What it doesn't tell us:

• Clinical diagnosis
• Objective mental health status
• Full context of their life
• Underlying biological factors

What Therapy Data IS NOT

❌ Not Medical Diagnostics

Therapy data ≠ clinical assessment.

Medical diagnostics require:

• Comprehensive clinical interviews
• Standardized assessment tools (administered by professionals)
• Differential diagnosis (ruling out other conditions)
• Medical history and physical exams
• Professional judgment from licensed clinicians

Mood tracking apps provide:

• Simple self-reported ratings
• Unstructured journal entries
• Pattern observations

These are fundamentally different.

❌ Don't claim:

• "Diagnoses depression"
• "Detects anxiety disorders"
• "Clinical-grade assessment"

✅ Do say:

• "Tracks personal mood patterns"
• "Supports self-awareness"
• "Complements professional assessment"

❌ Not Predictive of Crisis

Therapy data cannot predict suicide or self-harm.

Why prediction fails:

• Suicide is not predictable from mood data alone
• False positives create unnecessary alarm
• False negatives create dangerous false security
• Complex interplay of factors beyond mood

Don't build:

• Suicide risk prediction models
• "AI detects when you're at risk"
• Automated crisis alerts based on mood patterns

Do build:

• Crisis resource displays when user mentions crisis language
• Professional referral suggestions for persistent concerning patterns
• Always-accessible crisis buttons

The difference:

• ❌ Predicting crisis before it happens (impossible)
• ✅ Responding when user expresses crisis (keyword detection)

❌ Not Continuous Monitoring

Therapy data is sparse and episodic, not continuous.

Contrast with wearables:

• Wearables: heart rate every second, 24/7
• Therapy apps: user logs mood 0-3 times per day, if they remember

Implications:

• Big gaps in data
• User decides when/what to log
• Can't "monitor" mental state continuously

Don't position as:

• "24/7 mental health monitoring"
• "Always watching over you"

Do position as:

• "Track your patterns when you check in"
• "Reflective tool for self-awareness"

Signal Characteristics

1. Mood Ratings

What it is:

• User selects number (e.g., 1-5) representing current mood
• Often with emoji or color coding
• Sometimes includes dimensions (happiness, energy, anxiety)

What it tells us:

• User's subjective emotional state at moment of logging
• Relative changes over time
• Patterns across days/weeks

What it doesn't tell us:

• Absolute "truth" of mental health
• Clinical severity
• Underlying causes

Strengths:

• Easy to track over time
• Visual patterns emerge (mood graphs)
• Low friction (quick to log)

Limitations:

• Highly subjective (your "3" ≠ my "3")
• Recall bias (did I feel this way all day, or just now?)
• Missing context (why this mood?)

2. Journal Entries

What it is:

• Free-form text where users write about thoughts, feelings, events
• Prompts may guide reflection

What it tells us:

• User's narrative about their experience
• Events, thoughts, emotions they find significant
• Self-reflection and insight

What it doesn't tell us:

• Objective "facts" (memory is reconstructive)
• Full picture (people omit, forget, reinterpret)
• Clinical assessment (narrative ≠ diagnosis)

Strengths:

• Rich qualitative data
• Personal meaning and context
• Supports self-reflection

Limitations:

• Unstructured (hard to analyze quantitatively)
• Privacy-sensitive (requires encryption, user control)
• Not everyone journals consistently

3. Symptom Logs

What it is:

• Checklists of common symptoms (e.g., "difficulty sleeping," "loss of interest")
• Often yes/no or severity ratings

What it tells us:

• Presence/absence of specific experiences
• Symptom patterns over time

What it doesn't tell us:

• Diagnosis (symptoms alone don't = disorder)
• Severity in clinical context
• Differential diagnosis (many conditions share symptoms)

Strengths:

• More structured than mood ratings
• Can track specific issues (sleep, appetite, energy)
• Helpful for conversations with professionals

Limitations:

• Not diagnostic
• User interpretation varies
• Easy to confuse symptom tracking with diagnosis

⚠️ Warning: Symptom logs can easily slip into pseudo-diagnosis if not carefully positioned.

❌ Don't say: "Your symptoms indicate depression."
✅ Do say: "You've checked several symptoms. Consider discussing with a professional."

4. Activity/Self-Care Tracking

What it is:

• Logs of activities (exercise, social time, sleep)
• Self-care practices (meditation, gratitude)

What it tells us:

• Behavioral patterns
• Correlations (e.g., better mood on days with exercise)

What it doesn't tell us:

• Causation (correlation ≠ cause)
• Full context (maybe exercise correlates with good-weather days)

Strengths:

• Actionable insights (notice what helps)
• Empowering (user can make changes)

Limitations:

• Correlation only
• Oversimplification (mood is multi-causal)

Signal Quality Issues

Reporting Bias

Users decide what to log.

• Might only log when feeling very good or very bad
• Might avoid logging during difficult periods
• Might present idealized version

Implication: Data is incomplete and selective.

Don't assume: Logged data = complete picture.

Recall Bias

Memory is imperfect.

• "How did I feel today?" asked at 9 PM might only capture evening mood
• Negative moods can color recall of entire day
• Positive events may be forgotten in low mood

Implication: Self-reports are reconstructions, not recordings.

Subjectivity Variance

Your "3" isn't my "3."

• People have different baselines
• Cultural differences in emotion expression
• Scale interpretation varies

Implication: Cross-person comparisons are meaningless. Focus on within-person patterns.

Missing Context

Mood data lacks life context.

• User might not log major life events
• External stressors unknown
• Physical health factors invisible

Implication: Patterns may have hidden explanations.

Don't over-interpret mood drops without context.

What We CAN Do with Therapy Data

✅ 1. Personal Pattern Recognition

Observe individual trends:

• "You tend to feel better on days you exercise"
• "Your mood dips on Sundays"
• "Journaling seems to help you process"

Use for:

• Personal insights
• Self-awareness
• Informed conversations with therapists

✅ 2. Identify When to Seek Help

Recognize concerning patterns:

• Persistent low mood (2+ weeks)
• Significant changes from baseline
• Increasing symptom severity

Use for:

• Suggesting professional evaluation
• Encouraging help-seeking
• NOT diagnosis, just "worth discussing with a professional"

✅ 3. Support Self-Reflection

Facilitate meaning-making:

• Journal prompts
• Reviewing past entries
• Noticing personal growth

Use for:

• Self-awareness
• Tracking progress
• Complementing therapy

✅ 4. Complement Professional Treatment

Provide data for therapists:

• Mood graphs to discuss in session
• Symptom patterns over time
• Concrete data instead of memory alone

Use for:

• Patient-provider communication
• Treatment planning (by professional)
• Tracking progress in therapy

What We CANNOT Do with Therapy Data

❌ 1. Diagnose Mental Health Conditions

Never:

• "You have depression"
• "This indicates anxiety disorder"
• Any DSM-5 diagnosis

Why: Diagnosis requires comprehensive clinical assessment, not just self-reported mood.

❌ 2. Predict Crisis or Suicide

Never:

• "AI predicts suicide risk"
• "Alert: user at high risk"
• Automated crisis intervention

Why: Suicide is not predictable from mood data. False positives/negatives are dangerous.

Instead: Display crisis resources when user expresses crisis language.

❌ 3. Replace Therapy

Never:

• Position app as therapy substitute
• "AI therapist"
• Treatment without professionals

Why: Therapy requires trained humans, therapeutic relationship, evidence-based interventions.

Instead: Support self-awareness, complement professional care.

❌ 4. Make Medical Recommendations

Never:

• Medication suggestions
• Treatment protocols
• Clinical interventions (CBT, exposure therapy, etc.)

Why: Medical decisions require professional training and oversight.

Instead: Defer all medical questions to doctors/therapists.

Summary

Therapy data signals are:

• Self-reported subjective experiences
• Episodic and sparse (not continuous)
• Personal perception of emotional state
• Useful for self-awareness and professional conversations

They are NOT:

• Medical diagnostics
• Predictive of crisis
• Continuous monitoring
• Substitute for professional assessment

We can:

• Recognize personal patterns
• Suggest professional help when appropriate
• Support self-reflection
• Complement therapy

We cannot:

• Diagnose conditions
• Predict suicide
• Replace therapy
• Make medical recommendations

Use therapy data responsibly:

• Acknowledge limitations
• Position appropriately
• Refer to professionals
• Respect user autonomy

Signals are valuable for what they are — self-awareness tools — as long as we don't claim they're something more.