Understanding Reports | AI Verification Help

Report Overview

Every completed verification generates a comprehensive report analyzing the submitted content across multiple dimensions. Reports combine automated AI analysis with structured scoring to provide actionable insights for administrators reviewing copyright registrations.

Understanding how to read and interpret these reports is essential for making informed approval or rejection decisions. This guide explains each section of the report and how to use the information effectively.

Report Structure

Verification reports are organized into distinct sections, each providing specific insights:

1. Executive Summary

The executive summary appears at the top of every report and provides a high-level overview of the verification results. This section is designed for quick assessment and decision-making.

Overall Risk Score: Single aggregated score from 0-100
Recommendation: Approve, reject, or investigate further
Key Findings: Most significant concerns identified
Processing Time: How long verification took
Providers Used: Which AI providers contributed

2. Individual Confidence Scores

Each of the four verification types produces its own confidence score. These scores indicate how strongly the AI detected specific issues.

Infringement Risk: 0-100 scale, higher means more similarity detected
Laundering Risk: 0-100 scale, higher means manipulation detected
Authenticity Score: 0-100 scale, higher means more authentic
Duplicate Probability: 0-100 scale, higher means duplicate likely

3. Detailed Findings

Specific evidence and reasoning supporting the confidence scores. This section provides the "why" behind the numbers.

Exact matches or similarities found
Specific manipulation techniques detected
AI generation indicators identified
Duplicate records in database
Supporting evidence and excerpts

4. Provider-Specific Results

Individual results from each AI provider that participated in the analysis. Different providers may identify different concerns.

OpenAI GPT-4o text and audio analysis
Anthropic Claude reasoning and context
Google Cloud vision and transcription
Custom ML model fingerprinting

5. Technical Details

Metadata and technical information about the verification process itself.

Verification ID and timestamp
Content type and file formats analyzed
Processing duration and queue time
Provider API versions and models used
Any errors or warnings encountered

Understanding Risk Scores

All risk scores in the verification system use a consistent 0-100 scale. Understanding how to interpret these scores is crucial for making appropriate decisions.

Low Risk (0-30)

Content appears legitimate with minimal concerns detected. Generally safe to approve, but always review context and other factors.

Medium Risk (31-70)

Some concerns identified that warrant further investigation. Manual review recommended before making final decision.

High Risk (71-100)

Significant issues detected. Recommendation is typically rejection unless strong evidence supports legitimacy.

Overall Risk

Weighted combination of all individual scores. Most important metric for quick assessment and decision-making.

Interpreting Confidence Scores

Infringement Risk Score

Indicates similarity to existing copyrighted works. Higher scores suggest potential infringement.

0-20: No significant similarity found, unique content
21-40: Minor similarities that may be coincidental
41-60: Notable similarities requiring investigation
61-80: Strong similarities suggesting possible infringement
81-100: Very high similarity, likely infringement

Laundering Risk Score

Detects attempts to disguise copyrighted content through modifications. Higher scores indicate manipulation.

0-20: No manipulation indicators detected
21-40: Minor modifications that appear legitimate
41-60: Suspicious patterns requiring closer examination
61-80: Clear manipulation techniques identified
81-100: Strong evidence of copyright laundering

Authenticity Score

Measures likelihood that content is genuine human-created work. Note: Higher is better for authenticity.

0-20: Almost certainly AI-generated or fake
21-40: Strong AI generation indicators present
41-60: Uncertain authenticity, mixed signals
61-80: Likely authentic with minor concerns
81-100: High confidence in human authenticity

Duplicate Probability

Likelihood that content matches existing registrations. Higher scores indicate duplication.

0-20: No duplicates found, unique work
21-40: Some similarity but likely distinct work
41-60: Notable similarity to existing registrations
61-80: High similarity, likely duplicate
81-100: Near-certain duplicate registration

Reading Detailed Findings

The detailed findings section provides specific evidence supporting the confidence scores. Understanding how to interpret this information helps validate AI conclusions.

Types of Evidence Provided

Text Matches: Specific lyric phrases or passages matching known works
Audio Signatures: Waveform fingerprints with similarity percentages
Melodic Patterns: Musical sequences identified in analysis
Metadata Anomalies: Inconsistencies in file metadata
AI Artifacts: Specific indicators of AI generation
Database Matches: Registration IDs of similar works

Example: Infringement Finding

Similarity detected with "Song Title" by Artist Name
- Registration ID: abc-123-xyz
- Similarity Score: 78%
- Matching Elements:
  * Chorus melody: 85% match
  * Chord progression: 92% match
  * Lyrical phrase: "specific phrase here" (exact match)
- Analysis: High structural similarity suggests derivative work

Example: Laundering Detection

Manipulation indicators detected:
- Pitch shifted by +2 semitones
- Tempo increased by 8%
- Spectral signature matches known work after normalization
- File metadata shows recent conversion
- Recommendation: Likely laundered content

Example: AI Detection

AI generation indicators found:
- Vocal synthesis markers consistent with AI voice generation
- Instrumental patterns match common ML model outputs
- Lack of natural recording artifacts
- Metadata inconsistent with claimed recording method
- Confidence: 89% AI-generated

Provider-Specific Analysis

Different AI providers contribute unique perspectives to the verification. Understanding what each provider focuses on helps interpret conflicting results.

OpenAI Results

Text analysis, audio transcription, visual content recognition. Strong at natural language understanding and contextual analysis.

Anthropic Claude Results

Deep reasoning, legal context, complex pattern recognition. Excellent at explaining "why" behind detections.

Google Cloud Results

OCR, image labeling, speech-to-text. Specialized in visual and audio document processing.

Custom ML Results

Audio fingerprinting, waveform analysis, spectral comparison. Technical analysis of audio characteristics.

When Providers Disagree

If different providers produce conflicting results, consider these factors:

Each provider analyzes different aspects of the content
Check which provider specializes in the relevant area
Look for consensus among majority of providers
Consider the strength of evidence from each provider
Review detailed findings for specific evidence
When in doubt, err on side of caution with rejection

Making Decisions Based on Reports

Approval Guidelines

Consider approving when:

Overall risk score below 30
All individual scores in low-risk range
No specific high-confidence concerns identified
Provider consensus supports approval
Detailed findings show no significant issues
Authenticity score above 70

Rejection Guidelines

Consider rejecting when:

Overall risk score above 70
Any individual score in high-risk range (71+)
Clear evidence of infringement or laundering
Authenticity score below 40 (AI-generated)
Duplicate probability above 80
Strong provider consensus on issues

Investigation Guidelines

Require further investigation when:

Overall risk score between 31-70
Mixed individual scores (some high, some low)
Providers disagree on findings
Evidence is unclear or ambiguous
High-value registration with medium risk
Applicant can provide additional documentation

Important Considerations

AI reports are tools to assist decisions, not replacement for human judgment
Always read detailed findings, not just overall scores
Document reasoning for decisions that contradict AI recommendations
Consider broader context including registration history and applicant reputation

Common Report Scenarios

Scenario 1: Clear Approval

Overall Risk: 8
Infringement: 5, Laundering: 3, Authenticity: 95, Duplicates: 0
Action: Approve - All indicators positive, no concerns detected

Scenario 2: Clear Rejection

Overall Risk: 87
Infringement: 92, Laundering: 78, Authenticity: 88, Duplicates: 15
Action: Reject - High infringement and laundering risk

Scenario 3: AI-Generated

Overall Risk: 82
Infringement: 15, Laundering: 8, Authenticity: 22, Duplicates: 0
Action: Reject - AI-generated content not eligible

Scenario 4: Needs Investigation

Overall Risk: 48
Infringement: 52, Laundering: 28, Authenticity: 85, Duplicates: 12
Action: Investigate - Medium infringement risk requires review

Understanding Verification Reports