Agentprivacy Pretext Measurement

PVM-V5.2 Role Skill — Pretext Measurement

Source: Privacy Value Model V5.2 + First Person Spellbook Act XXVIII (The Ceremony Engine) Target context: Browser extension developers, privacy engineers, measurement specialists Architecture: agentprivacy.ai · Sync: sync.soulbis.com · Contact: mage@agentprivacy.ai

What this is

Pretext measurement is a technique for extracting text properties without accessing DOM content. The measurement touches the shadow, not the surface. It knows the weight without reading the words.

The shadow tells you the height without climbing the tower. The orb measures the text without reading the words.

This enables privacy-preserving interactions where:

• Text properties are needed (length, structure, patterns)
• Text content must remain private
• DOM access would leak information

The Measurement Problem

Traditional Approach (Privacy-Leaking)

// This reads the content
const text = document.getElementById('content').innerText;
const length = text.length;
const wordCount = text.split(' ').length;
// Content is now in memory, accessible, loggable

Pretext Approach (Privacy-Preserving)

// This measures without reading
const pretext = measurePretext(element);
// pretext.length exists
// pretext.wordCount exists
// pretext.content does NOT exist

Shadow Measurement Technique

Principle

The DOM casts shadows through:

• Computed styles (width, height)
• Scroll dimensions
• Selection ranges
• Intersection observations

These shadows reveal properties without exposing content.

Implementation Sketch

function measurePretext(element) {
  // Create measurement probe
  const probe = document.createElement('div');
  probe.style.cssText = `
    position: absolute;
    visibility: hidden;
    white-space: pre-wrap;
    font: inherit;
  `;

  // Mirror element's text styling
  const computed = getComputedStyle(element);
  probe.style.width = computed.width;
  probe.style.font = computed.font;
  probe.style.letterSpacing = computed.letterSpacing;

  // Measure shadow (scroll dimensions indicate content size)
  // WITHOUT reading innerText
  const metrics = {
    scrollHeight: element.scrollHeight,
    scrollWidth: element.scrollWidth,
    lineHeight: parseFloat(computed.lineHeight),
    charWidth: measureCharWidth(computed.font)
  };

  // Derive properties from shadow
  return {
    estimatedLines: Math.ceil(metrics.scrollHeight / metrics.lineHeight),
    estimatedChars: Math.ceil(metrics.scrollWidth / metrics.charWidth),
    hasOverflow: element.scrollHeight > element.clientHeight,
    // Content is NEVER accessed
  };
}

The Pretext Orb

The orb is the visual representation of pretext measurement:

     ╭──────────╮
    ╱   📊      ╲
   │  Metrics   │
   │  without   │
   │  Content   │
    ╲   🔮      ╱
     ╰──────────╯

Orb Properties

Orb Limitations

The orb cannot determine:

• Actual text content
• Specific words
• Semantic meaning
• Named entities

This is by design. The limitation IS the feature.

Use Cases

1. Progressive Disclosure Check

Before revealing content, check if the reader has spent enough time:

const pretext = measurePretext(article);
const estimatedReadTime = pretext.estimatedChars / 200; // 200 wpm
const actualTimeSpent = getTimeOnPage();

if (actualTimeSpent >= estimatedReadTime * 0.8) {
  // Reader likely read the content
  unlockNextSection();
}

2. Mana Calculation

Mana earned depends on content engagement without reading content:

const pretext = measurePretext(spellbookPage);
const engagementScore = calculateEngagement(pretext, scrollDepth, timeSpent);
const manaEarned = engagementScore * MANA_RATE;

3. Ceremony Verification

Verify understanding ceremony without accessing submission:

const originalPretext = measurePretext(originalContent);
const responsePretext = measurePretext(userResponse);

// Check structural similarity without reading either
if (responsePretext.complexity >= originalPretext.complexity * 0.7) {
  // Response demonstrates engagement
  proceedToCeremony();
}

Security Properties

What Pretext Protects Against

• Content logging by measurement code
• Content exfiltration through metrics API
• Side-channel content reconstruction (within limits)

What Pretext Does NOT Protect Against

• Timing attacks on specific content patterns
• High-precision scroll position analysis
• Intersection observer abuse with fine targets

Threat Model

Pretext assumes:

• Measurement code is potentially untrusted
• Content must remain in DOM (visible to user)
• Approximate properties are acceptable

Mapping to PVM-V5.2

Proverb

"The shadow tells you the height without climbing the tower. The orb measures the text without reading the words."

Emoji Spell

📐⊥DOM → 📏(text) → weight(shadow) → measure·¬touch → 🔮(orb)

Open Problems

1. Measurement Precision: How precise can shadow measurement get before it leaks content?
2. Cross-Browser Consistency: Do shadow measurements work consistently across browsers?
3. Dynamic Content: How to handle content that changes during measurement?
4. Adversarial Content: Can content be crafted to fool pretext measurement?
5. Formal Guarantees: Can we prove information-theoretic bounds on pretext leakage?

Verify: agentprivacy.ai · spellweb.ai · github.com/mitchuski/agentprivacy-docs