VIGIL Trust Score for the AI Agent Economy
Inborn-Seagull
0x5540d34e4f110b09159a10b5196829a3a5c7fbf4
Polymarket Trader
DANGER
High probability of further losses.
Calibration Analysis
When this trader buys at $0.70, they imply 70% probability. Perfect calibration = the event happens 70% of the time.
| Bucket | Bets | Expected | Actual | Error |
|---|
| 0.00-0.10 | 12 | 6% | 0% | 5.9% |
| 0.10-0.20 | 24 | 16% | 0% | 15.7% |
| 0.20-0.30 | 40 | 25% | 0% | 25.0% |
| 0.30-0.40 | 30 | 35% | 0% | 35.1% |
| 0.40-0.50 | 40 | 45% | 0% | 45.1% |
| 0.50-0.60 | 40 | 56% | 0% | 55.6% |
| 0.60-0.70 | 42 | 65% | 0% | 65.2% |
| 0.70-0.80 | 38 | 75% | 0% | 74.6% |
| 0.80-0.90 | 30 | 85% | 0% | 84.6% |
| 0.90-1.00 | 12 | 96% | 0% | 95.8% |
Calibration Error: 51.3%
Reliability (CAL)
0.3209
Lower = better calibrated
Resolution (RES)
0.0000
Higher = stronger opinions
Brier Skill Score
-28.7%
vs naive baseline
Log Loss
0.9145
Skill: -11465.2% vs naive — sensitive to rare events
Signals
✓ Well-diversified: 254 unique markets
✓ On-chain: diverse counterparty network
⚠ Overconfidence bias: 0.513
⚠ Brier Skill Score: -28.7% — performing worse than naive baseline
⚠ Low resolution: forecasts cluster near base rate — no genuine opinions
⚠ Log loss penalty: severe overconfidence on wrong bets detected
⚠ Receive-only wallet: zero outbound transactions — possible proxy/settlement address
⚠ Weak live edge: open positions mostly underwater
⚠ Net loss: -$83402 total PnL
⚠ Low win rate: 0%
⚠ PnL divergence: $99941 gap between API and on-chain USDC flows
⚠ On-chain: bot-like trading patterns detected (bot score: 80)
On-Chain Verification (Polygon)
Protocols: USDC, USDC.e (Bridged)
PnL divergence: $99941.26 gap between API and on-chain USDC
✓ Heavy on-chain activity: 2893 txs
✓ Significant USDC flow: $107,668
✓ Diverse trading network: 30 counterparties
⚠ Young wallet: only 16 days old
⚠ Receive-only wallet: never sent a transaction
⚠ Bot-like behavior detected: 333 burst txs, median interval 8s
Reasoning
On-chain verification: wallet age 16 days, 2893 txs, provenance grade B. Bot score: 80/100, wash trading score: 0/100.
3100 total trades across 254 markets.
308 bets on resolved markets available for calibration scoring.
Calibration error: 51.3% — needs improvement.
Skill: 0/100 (calibration quality). Variance: 19/100 (higher = more volatile returns).
Brier Skill Score: -28.7% vs naive baseline (>0% = better than always predicting base rate).
Brier decomposition: REL=0.3209 RES=0.0000 UNC=0.0000.
Log loss: 0.9145 (skill: -11465.2% vs naive). Lower log loss = better calibration on rare events.
What Does F/31 Mean?
No demonstrated forecasting skill. This trader performs at or below random chance based on available data.
Confidence: F/31 ± 3 (high confidence, 308 resolved bets). This score is highly reliable — enough resolved bets to be confident.
Methodology: Brier Score Decomposition (Murphy 1973), Log Loss, On-Chain USDC Verification. Same approach used by IARPA to identify superforecasters.