Decision Framework
How ReVault agents make autonomous trading decisions using AI and market intelligence.
The Decision Pipeline
Opportunity Detection
<50ms- •Monitor 1000+ endpoints for inventory drops
- •Cross-reference against predicted release calendar
- •Flag if model matches historical high-demand patterns
Profitability Assessment
<100ms- •Query KicksDB for historical comps (same brand/style)
- •Calculate expected resale premium (typically 200-400%)
- •Factor in transaction costs and agent fees (avg 15%)
- •Compare to profitability threshold (default: 20% minimum)
Demand Analysis
<75ms- •Aggregate trader sentiment from demand signals
- •Analyze release timing (drop timing affects demand)
- •Model size distribution (some sizes 5x more valuable)
- •Predict market saturation and hold time needed
Strategy Optimization
<50ms- •Calculate optimal size range to target
- •Determine bid amounts for each channel
- •Plan resale timing and marketplace selection
- •Set stop-loss thresholds if market changes
Execution
<200ms- •Place bids across optimal retail channels
- •Monitor fill rates and adjust in real-time
- •Verify inventory upon successful purchase
- •Initiate logistics for resale preparation
Total Decision Latency: ~475ms from opportunity detection to execution. Fast enough to capture limited inventory before stockouts.
Decision Criteria
Historical Demand Score
30%Similar models from same brand/designer. If comparable sold with 300% premium, this drop likely will too.
Trader Signal Sentiment
25%Aggregated signals from 10,000+ active traders showing real interest. High sentiment = higher expected resale prices.
Inventory Availability
20%Lower inventory releases create scarcity premium. 500 pair drops = higher markup than 5000 pair drops.
Release Timing
15%Friday drops outperform Tuesday drops. Holiday releases see different patterns. Agents learn these temporal signals.
Brand Momentum
10%Is the brand trending? Collaborations with celebrities/designers spike demand. ML tracks brand trajectory.
Real-World Decision Example
Nike Jordan Retro 1 "Bred" Drop
DECISION INPUTS
- • Release date: Friday 10 AM EST
- • Inventory: 2,500 pairs globally
- • Retail price: $170
- • Brand: Nike (highest brand score)
- • Model: Retro 1 (proven demand)
ML ANALYSIS
- • Historical comp premium: 380%
- • Trader sentiment: 94/100
- • Size 10-11 demand: +420%
- • Predicted resale: $815 avg
- • Profitability: 45% (after fees)
DECISION: EXECUTE
Profitability exceeds 20% threshold. Historical patterns strongly support high resale value. Trader sentiment near maximum. Agent places bids for 3-4 pairs across multiple channels within 100ms of drop.
Risk Management
Market Saturation
Agents track resale inventory levels. If marketplace suddenly floods with the same model, sell window closes and agent liquidates immediately.
Price Crash
Stop-loss thresholds built-in. If resale prices drop >15% from prediction, agent sells at market rather than waiting for recovery.
Fill Failure
Bids placed across 10+ channels in parallel. Even if 90% fail, 1-2 fills still provide profitability target.
Model Degredation
Prediction accuracy monitored continuously. If accuracy drops below 75%, agent reduces position size automatically.
Continuous Learning & Improvement
Every trade is a data point. ReVault agents improve over time through reinforcement learning:
- Model Retraining: Weekly retraining with 500+ new trades added to training set
- Feature Discovery: ML identifies new demand signals by analyzing trader behavior patterns
- Strategy Adaptation: Successful trading patterns automatically become new decision criteria
- Drift Compensation: Market changes detected and incorporated weekly to stay competitive