The Merit Score is a composite evaluation assigned to every post, comment, or thread on the Meritocious platform.
It reflects how much a contribution advances the conversation based on clarity, originality, relevance, civility, and constructive value.
It is not a popularity metric.
It is not based on karma, engagement volume, or user reputation.
It is not fixed—it evolves as the conversation evolves.
Each post is analyzed by AI models and scored across 5 weighted dimensions:
| Component | Weight | Description |
|---|---|---|
ClarityScore |
0.25 | Semantic coherence, grammar, readability, and structure |
NoveltyScore |
0.25 | Degree of semantic divergence from nearby posts (avoids repetition/echo) |
ContributionScore |
0.20 | Does it move the discussion forward? Refines, challenges, or expands ideas |
CivilityScore |
0.15 | Tone, respectfulness, empathy, and non-toxic language |
RelevanceScore |
0.15 | How well it connects to the thread, question, or topic at hand |
Each component returns a score between 0.00 and 1.00.
The final MeritScore is the weighted average:
MeritScore = (ClarityScore * 0.25) + (NoveltyScore * 0.25) + (ContributionScore * 0.20) + (CivilityScore * 0.15) + (RelevanceScore * 0.15)
- Uses LLM embeddings + grammar scoring models
- Penalizes confusing phrasing, contradictions, excessive jargon
- Favors clean, direct expression—even when complex
- Measures vector distance between your post and surrounding ones
- Higher score = your contribution brings new concepts or frames
- Reduces reward for reworded agreement or recycled takes
- Based on discourse tree analysis + semantic linking
- Scores higher if your post:
- Adds missing information
- Corrects a flaw in the argument
- Introduces a new lens on the topic
- Bridges two ideas together
- Uses a toxicity classifier + sentiment model
- Penalizes:
- Personal attacks
- Sarcasm masking hostility
- Dismissiveness or elitism
- Elevates:
- Respectful disagreement
- Good-faith questions
- “Steel-manning” an opposing view
- Checks semantic alignment with:
- The original post or question
- The specific parent comment (in nested threads)
- Encourages staying on-topic and thread-aware contribution
| Score Range | Interpretation | System Behavior |
|---|---|---|
0.00–0.29 |
Low signal / possible noise | Downranked, mod-review if extreme |
0.30–0.59 |
Limited merit, possible surface-level | Neutral or lightly visible |
0.60–0.79 |
Solid contribution | Promoted to core thread |
0.80–0.89 |
High-value insight | Highlighted and summarized |
0.90–1.00 |
Exceptional idea or framing | Featured, added to merit logs |
Merit scores are not fixed forever.
- If a comment gains meaningful replies, its ContributionScore may increase
- If a post is forked into deeper discussion, its RelevanceScore may rise
- If LLMs improve or moderation models update, scores can be re-evaluated
- Score changes are versioned and timestamped for transparency
Each post includes:
- A public Merit Score breakdown
- Option to view "Why this post is ranked here"
- History of score changes with context explanations
This allows users to:
- Learn what makes good contributions
- Improve their own posts over time
- Trust the system’s decisions
To reduce AI bias:
- Use open-source moderation models where possible
- Include feedback loops for users to flag misrankings
- Log edge cases for human + model retraining
- Run cross-model evaluations to check consistency
We believe in AI-assisted governance, not AI-as-overlord.
The Merit Score is not a score of your worth—it’s a reflection of how your idea contributes to the conversation.
The goal isn’t to chase numbers.
The goal is to surface insight—so we can all think better, together.
👁🗨 For algorithm details, model architectures, and feedback, visit:
/ai/moderation-engine | /discussions/scoring-feedback | /roadmap/voting-system