Skip to content

reputation.md

Latest commit

 

History

History
92 lines (64 loc) · 2.48 KB

reputation.md

File metadata and controls

92 lines (64 loc) · 2.48 KB
Raw

Contributor Reputation System

Overview

MisakaNet uses a reuse-weighted reputation system that rewards contributors whose lessons actually help others solve problems — not just who submits the most PRs.

Formula

score = usage_reports × 2.0
      + lessons_contributed × 1.0
      + lessons_reused × 0.2
      + lessons_verified × 0.5
Signal Weight Why
Usage reports ×2.0 Strongest signal: someone used this lesson to solve a real problem
Lessons contributed ×1.0 Baseline: contributing knowledge has value
Lessons reused ×0.2 Light signal: multiple people found it useful
Verified lessons ×0.5 Quality signal: lesson has verification steps

Anti-Gaming Safeguards

Sigmoid Cap

A single massive PR cannot dominate the leaderboard. The per-contributor score is multiplied by a sigmoid function of their lesson count:

cap = 1 / (1 + e^(-0.5 × (lessons - 10)))
Lessons Cap Effect
1 0.01 New contributor starts low
10 0.50 Midpoint — earned trust
50 1.00 Full weight — established contributor

This prevents gaming by submitting many low-quality lessons in one PR.

Time Decay

Recent contributions are weighted more. Older lessons decay with a half-life of 90 days:

weight = 0.5^(days_since_creation / 90)
Age Weight
Today 1.00
90 days 0.50
180 days 0.25
1 year 0.06

This ensures the leaderboard reflects current contribution activity.

Usage

# Full reputation table
python3 scripts/reputation.py

# Single contributor details
python3 scripts/reputation.py --contributor zsxh1990

# JSON output
python3 scripts/reputation.py --json

# Save to data/reputation.json
python3 scripts/reputation.py --save

Integration with Leaderboard

The reputation score feeds into the per-contributor leaderboard. Higher reputation gives a small search quality boost (not dominant — content quality still matters more).

Data Sources

Source Location Description
Lessons lessons/core/, lessons/contrib/ Lesson files with frontmatter
Usage reports data/usage_reports.json Reported usage events
Git history git log Contributor attribution

Tests

python3 tests/test_reputation.py

Covers: sigmoid cap behavior, time decay correctness, new contributor surge prevention, single-large-PR anti-gaming.