Voice Integrity Score (VIS) – Technical Appendix v1.2 (Public Version)

Aligned with internal VIS v1.2 flow & rubric.
Utility-only. No price language.

1. Purpose

The Voice Integrity Score (VIS) is the ranking and integrity layer of REAX.me.
It measures the trustworthiness and quality of a post — and, over time, the creator behind it — using six fixed components.

VIS ensures that:

• Verified humans are prioritized

• High-integrity posts rise in the feed

• Abusive, synthetic, or deceptive patterns are caught early

• The network remains human-first, not bot-first
  
  

VIS does not evaluate political views, identity attributes, or personal beliefs.

2. High-Level Flow (Public-Safe Summary)

The full internal workflow is more detailed, but this is the accurate public representation.

Step 1 — Capture

User records a video or voice reaction.

Step 2 — Attestation Layer (Pre-publish checks)

We apply a set of verification checks:

• C2PA content credentials

• Liveness check

• Voice-match confidence

• Synthetic/clone/replay detection

• Device + account consistency
  
  

Step 3 — Gate

• Pass → Proceed to VIS calculation

• Low-confidence or suspicious → Human-in-the-loop (HITL) review queue

  
  

Step 4 — VIS Compute (0–100)

Score is computed using six factors with fixed weights.

Step 5 — Publish

Post displays:

• “Human-Verified” badge

• VIS band (High / Mid / Low)
  
  

Step 6 — Logs

System stores:

• Proof bundle

• VIS components

• Model version

• Reviewer actions (if any)
  
  
  

Step 7 — Dispute / Appeal

User can request a re-check. Human adjudication included.

3. Fixed Formula (Exact from VIS v1.2)

VIS = 0.25·Credibility

• 0.20·Originality

• 0.15·Delivery

• 0.15·Impact

• 0.15·CommunityEcho

  − 0.10·(PolicyFlag)**

• All positive factors are scored 0–100

• PolicyFlag is 0 or 1

• VIS is capped between 0 and 100, then mapped to bands
  
  

4. Component Definitions (Accurate to VIS v1.2)

Credibility (0–100)

Inputs :

• Profile verified
  
  
  

• C2PA present

• Liveness pass

• Voice-match confidence

• Device/account history

• Zero-flags

• Moderator score
  
  
  

Scoring model (exact shape, publicly safe wording):

• Base: 50

• +10 C2PA

• +10 liveness

• +10 high voice-match

• +10 verified ID/brand

• +5 device/account consistency

• +5 strong moderator score

• Cap: 100
  
  

Meaning:
Credibility measures whether the post is coming from a consistent, verified, and trustworthy human.

Originality (0–100)

Inputs:

• Content similarity

• Repost ratio

• Synthetic likelihood

• Use of cited sources

• Novel media
  
  

Scoring:

• Base: 70
  
  
  

• Minus similarity band

• Minus synthetic likelihood band

• +5 cited source

• +5 novel media

• Floor 0, cap 100
  
  
  

Meaning:
Originality measures newness: is this creator adding something meaningful rather than reposting or using generated material?

Delivery (0–100)

Inputs:

• Cadence stability

• Completion (audio clarity, coherence)

• On-topic cohesion
  
  

Scoring:

• Base: 60
  
  

• +10 consistent weekly posting

• −10 spam bursts

• −10 long inactivity

• +5 clear audio

• +5 concise structure
  
  

Meaning:
Delivery measures clarity, structure, and whether the creator maintains a steady, human posting pattern.

Impact (0–100)

Inputs:

• Verified replies

• Saves

• Time-watched %

• Downstream remixes
  
  

Scoring:

• Weighted engagement quality

  • 60% from verified replies

  • 40% from shares/remixes

    
    

• Normalized by reach
  
  
  

Meaning:
Impact measures whether other verified humans meaningfully engage with the post, not just raw view counts.

CommunityEcho (0–100)

Inputs:

• Ratio of verified reactions

• Community trust rating

• Past VIS stability
  
  

Scoring:

• (Verified reactions ÷ Total reactions) × 100

• Blended with trust panel data
  
  

Meaning:

CommunityEcho measures how real humans respond, trust, and interact with the creator over time.

PolicyFlag (0 or 1)

Inputs:

• Safety violations (policy-defined)

• Harassment, hate, incitement

• Coordinated spam or manipulation

• Attempts to evade trust & safety
  
  

Effect:

• 1 → applies −10 penalty and triggers a moderation flow

• 0 → no penalty applied
  
  
  

Meaning:

PolicyFlag is the only negative component and is used only for serious violations.

5. How VIS Is Used (Public-Safe)

VIS impacts:

Ranking

Higher VIS → higher placement in human-first feeds.

Eligibility

Certain features or reward programs may require:

• Verified account

• Minimum VIS band

  
  

Safety

Low VIS + repeated PolicyFlags → extra review or limits.

VIS never evaluates:

• Political alignment

• Demographics

• Personal identity

• Opinion conformity
  
  

Only behavior and integrity signals.

6. Post-Level vs. Creator-Level VIS

Accurate to internal rubric.

• Post-level VIS:
  Calculated per post using the full 6-factor model.

  
  

• Creator-level VIS:

  
  Aggregated from multiple posts.
  Incorporates historical VIS stability and avoids overreacting to spikes.
  
  

Creator-level VIS updates more slowly to ensure stability and reduce manipulation

7. Versioning & Transparency

This appendix describes VIS v1.2 from the internal rubric.

Public commitments:

• Fixed weights remain public.

• Component definitions remain public.

• Internal detection thresholds and adversarial rules remain private (security reasons).

• Major version changes will be documented publicly.
  
  

VIS bands and explanations will be visible in-app through a simple breakdown UI.