The AI Debate Platform uses a comprehensive 4-criterion scoring system with Evidence Quality weighted at 40% as the primary factor. This ensures that debaters presenting facts, citations, and irrefutable arguments receive the highest scores.
The most important scoring criterion. Awards high scores for fact-backed arguments.
High Score (8-10/10):
- Extensive citations from peer-reviewed research
- Well-known statements and established facts
- Specific, verifiable data and statistics
- References to recognized experts and institutions
- Clear sourcing of claims
- Multiple independent sources cited
- Arguments supported by irrefutable facts
Medium Score (5-7/10):
- Some citations and references
- Mix of factual and opinion-based statements
- General references without specific citations
- Some verifiable data
- Partially substantiated claims
Low Score (0-4/10):
- No citations or references
- Unsupported claims
- Vague generalizations
- Logical leaps without evidence
- Contradictions with established facts
- Opinions presented as facts
Examples:
✅ HIGH: "According to a 2023 McKinsey study on AI employment, 67% of companies expect AI to create more jobs than it displaces. Research from Stanford's Human-Centered AI Institute shows..."
✅ MEDIUM: "Studies have shown that AI creates jobs in some sectors, though job displacement occurs in others."
❌ LOW: "AI will obviously create tons of jobs because technology always does."
Evaluates clarity, structure, and logical flow of arguments.
High Score (8-10/10):
- Clear, well-organized presentation
- Logical flow from premise to conclusion
- Coherent thesis statement
- Well-structured supporting points
- Easy to follow reasoning
- Comprehensive coverage of topic
Medium Score (5-7/10):
- Generally clear structure
- Some logical organization
- Understandable but could be clearer
- Partially developed arguments
- Some organizational issues
Low Score (0-4/10):
- Confusing or disorganized
- Unclear thesis or logic
- Rambling or incoherent
- Hard to follow
- Poor structure
Evaluates internal consistency and avoidance of contradictions.
High Score (8-10/10):
- No contradictions within own arguments
- Consistent reasoning throughout
- No logical fallacies
- Sound deduction and inference
- Premises support conclusions
Medium Score (5-7/10):
- Mostly consistent
- Minor contradictions
- Generally sound reasoning
- Few logical issues
Low Score (0-4/10):
- Multiple contradictions
- Logical fallacies present
- Inconsistent reasoning
- Premises don't support conclusions
- Self-refuting arguments
How well the debater addresses opponent's points and fills identified gaps.
High Score (8-10/10):
- Directly addresses opponent's main points
- Fills logical gaps in own previous arguments
- Provides specific rebuttals
- Acknowledges opponent's valid points
- Offers counter-evidence
- Builds on previous rounds effectively
Medium Score (5-7/10):
- Addresses some opponent points
- Partially fills gaps
- General rebuttals
- Some responsive elements
Low Score (0-4/10):
- Ignores opponent's points
- Doesn't address gaps
- Generic responses
- No acknowledgment of opposition
Overall Score = (Evidence Quality × 0.4)
+ (Argument Quality × 0.2)
+ (Logical Consistency × 0.2)
+ (Responsiveness to Gaps × 0.2)
Example:
- Evidence Quality: 9/10
- Argument Quality: 7/10
- Logical Consistency: 8/10
- Responsiveness: 8/10
Overall = (9 × 0.4) + (7 × 0.2) + (8 × 0.2) + (8 × 0.2) = 3.6 + 1.4 + 1.6 + 1.6 = 8.2/10
Beyond the 4-point scoring, the judge tracks:
- Number of factual claims with citations
- Number of referenced studies/sources
- Count of specific data points
- Total: Higher is better
- Count of arguments backed by strong evidence
- Claims supported by multiple sources
- Factual points that are difficult to refute
- Total: Higher is better
These metrics provide additional insight into the depth of evidence presented.
Debater: "AI will create more jobs than it displaces"
Evidence Provided:
- Cites World Economic Forum report (2023)
- References specific historical data on technology job creation
- Mentions studies from MIT and Stanford
- Provides statistics on emerging AI-related job categories
- Acknowledges counterarguments with data
Scores:
- Evidence Quality: 9/10 (extensive, verified sources)
- Argument Quality: 8/10 (well-structured, clear)
- Logical Consistency: 9/10 (coherent throughout)
- Responsiveness: 8/10 (addresses opponent points with data)
- Overall: 8.6/10
Debater: "AI will hurt the economy"
Evidence Provided:
- No citations
- General statements
- Opinions without sources
- Vague claims about job loss
- No specific data
Scores:
- Evidence Quality: 2/10 (no citations, unsupported)
- Argument Quality: 6/10 (somewhat organized)
- Logical Consistency: 5/10 (some logical leaps)
- Responsiveness: 3/10 (doesn't address specifics)
- Overall: 3.6/10
Debater: "Renewable energy is the future"
Evidence Provided:
- Some references but mostly general
- Well-organized presentation
- Clear structure
- Logical flow
- Missing specific studies
- Limited citations
Scores:
- Evidence Quality: 6/10 (some sources, could be stronger)
- Argument Quality: 8/10 (excellent structure)
- Logical Consistency: 8/10 (very consistent)
- Responsiveness: 7/10 (addresses points adequately)
- Overall: 7.2/10
Arguments score highest when they:
-
Are Fact-Based
- Supported by peer-reviewed research
- Reference established data
- Cite reliable sources
-
Reference Well-Known Statements
- Quote recognized experts
- Reference major studies
- Mention established theories
-
Use Specific Numbers
- Cite exact statistics
- Include dates and sources
- Provide measurable data
-
Address Counterarguments
- Acknowledge opposing views
- Provide evidence against counterarguments
- Show understanding of complexity
-
Build Logically
- Connect evidence to conclusions
- Show causation, not just correlation
- Avoid logical fallacies
"According to a 2023 Pew Research study on remote work productivity, 70% of workers reported equal or higher productivity while remote. The Stanford Internet Observatory found that asynchronous communication tools like Slack reduce meeting time by 40%. Additionally, the American Psychological Association reports that flexible work arrangements reduce employee stress by 23%. These findings are further supported by Microsoft's 2024 Work Trend Index, which shows that intentional connection increases remote work effectiveness."
Score: 9/10
- Multiple citations
- Specific percentages
- Verifiable sources
- Recent data
- Comprehensive evidence
"Remote work is clearly better because people like working from home. Obviously, commuting is bad and stressful. Everyone knows that offices are distracting. It just makes sense that remote work would be more productive."
Score: 3/10
- No citations
- No data
- Unsupported claims
- Opinions presented as facts
- Vague language
The judge follows this methodology:
- Extract Facts: Count citations, data points, references
- Verify Claims: Check if claims are supported by evidence
- Score Evidence: Rate quality of citations and sources
- Evaluate Structure: Assess clarity and organization
- Check Consistency: Verify logical coherence
- Assess Responsiveness: Evaluate how well gaps are addressed
- Calculate Overall: Apply weighted formula
- Provide Feedback: Explain reasoning
- Evidence Quality: 8-10
- Argument Quality: 7-8
- Logical Consistency: 8-9
- Responsiveness: 7-8
- Overall: 7.8-8.8
- Evidence Quality: 4-6
- Argument Quality: 8-9
- Logical Consistency: 8-9
- Responsiveness: 7-8
- Overall: 6.4-7.6
- Evidence Quality: 1-3
- Argument Quality: 3-5
- Logical Consistency: 2-4
- Responsiveness: 1-3
- Overall: 1.8-3.8
- Evidence Quality: 7-8
- Argument Quality: 7-8
- Logical Consistency: 7-8
- Responsiveness: 7-8
- Overall: 7.0-8.0
- Use academic research papers
- Cite recent, reputable sources
- Include specific statistics
- Reference expert opinions
- Provide multiple sources
- Use peer-reviewed studies
- Quote authoritative institutions
- Organize ideas clearly
- Use logical structures (thesis → supporting points → conclusion)
- Present ideas in order of importance
- Use clear language
- Avoid jargon or explain it
- Provide smooth transitions
- Check for contradictions
- Ensure premises support conclusions
- Avoid logical fallacies
- Be consistent across rounds
- Acknowledge complexity where it exists
- Address opponent's specific points
- Provide counter-evidence
- Acknowledge valid opposition points
- Fill identified gaps
- Build on previous rounds
- Reference opponent's own arguments
In well-matched debates:
- Winner: 7.5-9.0
- Competitive Loser: 6.5-7.5
- Clear Loser: 4.5-6.5
- Poorly Prepared: 1.0-4.5
The scoring system is designed to be:
- Objective: Based on verifiable criteria
- Transparent: Clear evaluation process
- Evidence-Focused: Facts matter most
- Fair: Weighted criteria apply equally
- Detailed: Feedback explains reasoning
Q: Does my model need citations to score well? A: While not strictly required, citations and evidence dramatically improve scores. The evidence quality criterion carries 40% weight.
Q: Can I win without many facts? A: It's difficult. With evidence quality at 40% weight, a debater scoring 3/10 on evidence would need nearly perfect scores (10/10) on all other criteria to overcome that deficit.
Q: How important is presentation? A: Argument quality is 20% of the score. While important, it's less critical than having strong evidence (40%).
Q: What if my evidence is recent but obscure? A: Recent research is valued, especially if from reputable sources (universities, research institutions, peer-reviewed journals). Obscure sources are less valuable than well-known ones.
Q: Can I reference the same study multiple times? A: Yes, but diversity of sources is better. Multiple sources referencing the same findings is stronger than one source cited multiple times.
Comprehensive evidence, perfect structure, no logical issues, direct responsiveness
Strong evidence, clear structure, consistent logic, good responsiveness
Some evidence, acceptable structure, minor logical issues, partial responsiveness
Weak evidence, unclear structure, logical issues, minimal responsiveness
No evidence, disorganized, contradictory, unresponsive
Version: 2.0 (Evidence-Focused) Last Updated: November 24, 2025