Leaderboard Scoring Model
6-dimension scoring model for wallet address analysis, plus AI archetype overlay.
Overview
totalScore = winrate(25) + stability(20) + drawdown(20) + tags(15) + pnl(10) + follow_friendly(10) = 100
finalScore = totalScore + ai_adjustment(±10)Rating: ⭐⭐⭐ ≥ 80 · ⭐⭐ ≥ 65 · ⭐ ≥ 50 · ❌ < 50
Dimension 1: Win Rate (25 points)
| Win Rate | Score |
|---|---|
| ≥ 70% | 25 |
| 60-69% | 20 |
| 50-59% | 15 |
| 40-49% | 8 |
| < 40% | 0 |
Source field: winRate (0-100).
Dimension 2: Stability (20 points)
Based on dailyPNL[] array — daily profit distribution + concentration penalty.
Scoring logic:
- Compute daily PnL values from
dailyPNL[](field:realizedPnlper day) - Base score: consistency of positive days (positive days / total days × 20)
- Concentration penalty: if single-day PnL > 50% of total, penalize proportionally
Note: These ranges are guidance for AI-judgment, not deterministic formulas. The backend does not expose the exact penalty formula. Use the ranges as scoring anchors and adjust based on the overall profit distribution shape.
| Stability Profile | Score |
|---|---|
| Consistent daily profits, no single-day dominance | 18-20 |
| Mostly consistent, some concentration | 12-17 |
| Inconsistent or high concentration | 6-11 |
| Erratic | 0-5 |
Dimension 3: Drawdown (20 points)
Based on tokenDistribution{} — four buckets of token PnL:
| Bucket | Description |
|---|---|
gt500Cnt |
Tokens with PnL > $500 |
between0And500Cnt |
Tokens with PnL $0-$500 |
between0AndNegative50Cnt |
Tokens with PnL -$50 to $0 |
ltNegative50Cnt |
Tokens with PnL < -$50 |
Scoring logic:
- Reward
gt500Cntandbetween0And500Cnt(profitable tokens) - Penalize
ltNegative50Cnt(heavy losses) - Rug penalty: if
ltNegative50Cnt> 30% of total tokens, additional penalty
Note: Like Stability, these ranges are AI-judgment guidance. The exact rug penalty formula is not exposed by the backend. Use the bucket distribution to assess overall drawdown health.
| Drawdown Profile | Score |
|---|---|
| Minimal losses, mostly profitable tokens | 16-20 |
| Some losses but manageable | 10-15 |
| Significant losses | 4-9 |
| Heavy losses / rug pattern | 0-3 |
Dimension 4: Tags (15 points)
Based on tags[] and genericAddressTagList[].
| Tag | Score |
|---|---|
| Smart Money | +15 |
| KOL | +10 |
| MPC | +5 |
| No notable tags | 0 |
Stacking rule: If multiple tags present, take the highest single tag score (do not stack).
Dimension 5: PnL (10 points)
Based on realizedPnl (USD).
| PnL Range | Score |
|---|---|
| ≥ $100K | 10 |
| $10K-$100K | 7 |
| $1K-$10K | 4 |
| $0-$1K | 2 |
| < $0 (loss) | 0 |
Dimension 6: Follow-Friendly (10 points)
Based on avgBuyVolume — the "sweet spot" for copy-trading.
| Avg Buy Volume | Score | Rationale |
|---|---|---|
| $5K-$50K | 10 | Large enough to matter, small enough to follow |
| $1K-$5K | 7 | Meaningful but small |
| $50K-$100K | 5 | Too large for most followers |
| > $100K | 3 | Whale — hard to follow |
| < $1K | 2 | Too small to be meaningful |
AI Overlay (±10)
Applied after the 6-dimension base score.
Archetype Classification
| Archetype | Description |
|---|---|
sniper |
High-frequency, small tickets, short holding periods |
swing |
Medium-frequency, medium holding periods |
accumulator |
Gradual position building over time |
farmer |
Airdrop/liquidity farming patterns |
mixed |
No dominant pattern |
Behavior Flags
Detected patterns that inform the AI adjustment:
High Frequency Small Amount— high frequency, small amountsNocturnal Active— predominantly nocturnal tradingConcentrated Building— concentrated position buildingRound-Trip— rapid round-trip tradesLong Holding— long holding periodsMulti-Chain— active on multiple chains
AI Adjustment
| Adjustment | Condition |
|---|---|
| +5 to +10 | Consistent archetype with clear edge (e.g., sniper with consistently high win rate) |
| -5 to -10 | Red flags (e.g., bot-like patterns, wash trading indicators) |
| 0 | Neutral or ambiguous |
Required: aiReason must be filled for audit trail whenever adjustment ≠ 0.
Rating Thresholds
| Rating | finalScore | Meaning |
|---|---|---|
| ⭐⭐⭐ | ≥ 80 | Excellent — strong track record, worth following |
| ⭐⭐ | ≥ 65 | Good — above average, consider following |
| ⭐ | ≥ 50 | Average — proceed with caution |
| ❌ | < 50 | Poor — not recommended |
Single Address Analyze Flow
- Query top N: Internally call
leaderboard querywith pagination to fetch the top N entries for the specified chain + period. Default N=1000, configurable via--top-n(max 5000). - Reverse lookup: Search for the target
addressin the results. - If found: Compute 6-dimension scores from the record fields → apply AI overlay → output
finalScore+rating. - If not found: Return "N beyond top" (beyond top N). Long-tail wallets can only be analyzed from already-imported groups — use
binance-wallet-trackerskill'saddress listto check if the address is tracked locally.
Limitation & Guidance
- Default scans top 1000 entries. If the target address is not found, inform the user: "Address not in top 1000. To expand search range, use
--top-n 5000to fetch more data, or check on-chain data inbinance-wallet-tracker." - The backend API supports arbitrary pagination (
pageSizemax 20,pageNofrom 0). The--top-nflag controls how many entries to scan. - Known backend bug: page 0 and page 1 return identical data. The CLI automatically deduplicates.