Key clarifications in ORDERBOOK_SHADOW_LOGGING.md:
1. Trade execution only (NOT periodic):
- Orderbook captured in /api/trading/execute (lines 1037-1053)
- NOT captured in /api/trading/market-data (1-minute feed)
- Frequency: 3-5 orderbook snapshots per day (when trades execute)
- vs. 1,440 market data updates per day (price/ATR/ADX/RSI)
2. Phase 1.5 declined (Dec 19, 2025):
- User asked about 1-minute periodic orderbook capture
- Agent explained tradeoffs: 1,440 API calls/day, more storage
- User confirmed trade-time capture sufficient for now
- Periodic monitoring not needed for Phase 1 validation
3. Scope clarity for resumption:
- Phase 1 complete and deployed
- Goal: Collect 50-100 trades with orderbook data
- Phase 2: Real orderbook API integration (Hyperliquid/Jupiter)
- Next agent: Monitor trade data collection, analyze patterns
Status: ✅ All Phase 1 tasks complete, awaiting first trade validation
7.5 KiB
Orderbook Shadow Logging System
Status: ✅ DEPLOYED (Dec 19, 2025)
Purpose: Track orderbook metrics for all trades to prepare for real orderbook integration (Phase 2)
Overview
Phase 1 implementation: Oracle-based estimates with 2bps spread assumption. Captures orderbook snapshots at trade execution only (not periodic).
Critical Scope Note: Orderbook data is captured ONLY when trades execute via /api/trading/execute, NOT during 1-minute market data feed updates (/api/trading/market-data). The 1-minute feed captures price/ATR/ADX/RSI metrics but does NOT capture orderbook data. This is intentional to minimize API overhead - we only need orderbook state at the moment of trade execution.
Database Schema
Table: Trade
New Columns (all nullable DOUBLE PRECISION):
| Column | Description | Expected Range |
|---|---|---|
spreadBps |
Bid-ask spread in basis points | 2-100 bps typical |
imbalanceRatio |
Order flow imbalance (-1 to 1) | Negative=sell pressure, Positive=buy |
depthBid1Usd |
Bid depth USD (1% from mid) | >0 |
depthAsk1Usd |
Ask depth USD (1% from mid) | >0 |
priceImpact1Usd |
Price impact for $1 trade | Small decimal |
bidWall |
Largest bid wall detected | NULL or >0 |
askWall |
Largest ask wall detected | NULL or >0 |
ENV Configuration
# Enable/disable orderbook shadow logging
ENABLE_ORDERBOOK_LOGGING=true # Default: true (omitting = defaults to true)
How to toggle:
# Disable logging
echo "ENABLE_ORDERBOOK_LOGGING=false" >> .env
docker compose restart trading-bot
# Re-enable
sed -i 's/ENABLE_ORDERBOOK_LOGGING=false/ENABLE_ORDERBOOK_LOGGING=true/' .env
docker compose restart trading-bot
Implementation Details
Files Modified:
prisma/schema.prisma- Added 7 orderbook fields to Trade modelconfig/trading.ts- AddedenableOrderbookLoggingboolean flag (line 127)types/trading.ts- Added orderbook fields to CreateTradeParams interfacelib/database/trades.ts- Updated createTrade() to save orderbook dataapp/api/trading/execute/route.ts- Added ENV flag wrapper (lines 1037-1053)
Data Source (Phase 1): Oracle prices with 2bps spread estimates
- No real orderbook API integration yet
- Estimates sufficient for pattern analysis
- Phase 2 will integrate Hyperliquid/Jupiter for real data
CRITICAL: Trade Execution Only:
- Orderbook capture happens in
/api/trading/executeendpoint (lines 1037-1053) - Triggered ONLY when trade signal executes (not periodic)
/api/trading/market-dataendpoint (1-minute feed) does NOT capture orderbook data- 1-minute feed captures: ATR, ADX, RSI, volume, price position (stored in MarketData table)
- Orderbook data stored in: Trade table (7 columns added via ALTER TABLE)
- Frequency: 3-5 orderbook snapshots per day (when trades execute) vs. 1,440 market data updates per day
Verification Queries
Check Latest Trade with Orderbook Data
SELECT
id,
symbol,
direction,
spreadBps,
imbalanceRatio,
depthBid1Usd,
depthAsk1Usd,
priceImpact1Usd,
bidWall,
askWall,
"createdAt"
FROM "Trade"
WHERE spreadBps IS NOT NULL
ORDER BY "createdAt" DESC
LIMIT 1;
Count Trades with Orderbook Data
SELECT
COUNT(*) as total_trades,
COUNT(spreadBps) as trades_with_orderbook,
ROUND(100.0 * COUNT(spreadBps) / COUNT(*), 1) as pct_coverage
FROM "Trade";
Average Orderbook Metrics by Symbol
SELECT
symbol,
COUNT(*) as trades,
ROUND(AVG(spreadBps)::numeric, 2) as avg_spread_bps,
ROUND(AVG(imbalanceRatio)::numeric, 3) as avg_imbalance,
ROUND(AVG(depthBid1Usd)::numeric, 2) as avg_bid_depth,
ROUND(AVG(depthAsk1Usd)::numeric, 2) as avg_ask_depth
FROM "Trade"
WHERE spreadBps IS NOT NULL
GROUP BY symbol
ORDER BY trades DESC;
Spread vs Win Rate Analysis
SELECT
CASE
WHEN spreadBps < 5 THEN '0-5 bps (tight)'
WHEN spreadBps < 10 THEN '5-10 bps (normal)'
WHEN spreadBps < 20 THEN '10-20 bps (wide)'
ELSE '20+ bps (very wide)'
END as spread_bucket,
COUNT(*) as trades,
ROUND(100.0 * COUNT(CASE WHEN "realizedPnL" > 0 THEN 1 END) / COUNT(*), 1) as win_rate,
ROUND(AVG("realizedPnL")::numeric, 2) as avg_pnl
FROM "Trade"
WHERE spreadBps IS NOT NULL
AND "exitReason" IS NOT NULL
GROUP BY spread_bucket
ORDER BY MIN(spreadBps);
Deployment Timeline
| Date | Action | Status |
|---|---|---|
| Dec 18, 2025 | Task 1-3: Code implementation | ✅ COMPLETE |
| Dec 18, 2025 | Task 4: ENV flag + syntax fix | ✅ COMPLETE |
| Dec 18, 2025 | Task 5: Database migration via direct SQL | ✅ COMPLETE |
| Dec 18, 2025 | Container rebuild and restart | ✅ COMPLETE |
| Dec 18, 2025+ | Production validation (await next trade) | ⏳ PENDING |
Production Validation Checklist
After next trade executes, verify:
- Query returns non-NULL orderbook values
- spreadBps in reasonable range (2-100 bps)
- imbalanceRatio between -1 and 1
- depthBid1Usd and depthAsk1Usd > 0
- priceImpact1Usd is small positive decimal
- bidWall/askWall NULL or positive (if detected)
Manual Test (Optional):
# Trigger test trade via Telegram
# Send: "long sol --force" or "short sol --force"
# Then query database for orderbook data
Phase 2 Roadmap (Future Enhancement)
Real Orderbook Integration:
- Hyperliquid API for SOL/ETH orderbook data
- Jupiter aggregator for DEX liquidity depth
- Replace oracle estimates with actual L2 orderbook snapshots
- Add real-time spread/imbalance monitoring
Potential Phase 1.5 (User Declined Dec 19, 2025):
- Periodic orderbook capture via 1-minute market data feed
- Would store orderbook snapshots every 1 minute in separate table (not Trade table)
- Use case: Track spread/imbalance trends over time independent of trades
- Tradeoffs: More API calls (1,440/day vs. 3-5/day), more storage, minimal benefit for Phase 1 validation
- Decision: User confirmed trade-time capture sufficient for now, periodic monitoring not needed yet
Analysis Queries
Troubleshooting
Problem: Next trade has NULL orderbook fields
Solution: Check ENV flag is true, restart container, verify code deployed
Problem: Values seem incorrect (negative depths, >100 spread)
Solution: Oracle estimates may be off - acceptable for Phase 1, will fix in Phase 2
Problem: Want to disable logging temporarily
Solution: Set ENABLE_ORDERBOOK_LOGGING=false in .env, restart container
Database Maintenance
Migration Method: Direct SQL (not Prisma migrate)
- Used due to migration history drift
- Zero data loss approach
- See git commit for ALTER TABLE statement
Prisma Sync:
DATABASE_URL="postgresql://postgres:postgres@localhost:55432/trading_bot_v4?schema=public" \
npx prisma db pull && npx prisma generate
Git History
- Commit: [TBD] "feat: Orderbook shadow logging system (Phase 1)"
- Files changed: 5 files (schema, config, types, database, execute)
- Database changes: 7 columns added via direct SQL
- Container: Rebuilt with sha256:6c88c4543ef8...
Notes
- Phase 1: Oracle-based estimates (2bps spread assumption)
- Phase 2: Real orderbook integration (Hyperliquid/Jupiter APIs)
- Data safety: Preserved production data during migration (avoided destructive reset)
- Defaults: Logging enabled by default (ENABLE_ORDERBOOK_LOGGING=true)
- Performance: Minimal overhead (estimates calculated at trade execution only)
Last Updated: Dec 18, 2025
System Status: ✅ PRODUCTION READY (awaiting first trade validation)