Persistence Layer Audit

Persistence Layer Audit Pattern

Experiment Overview

Context

SQLite databases can accumulate unused tables over time as the codebase evolves. This skill documents how to audit the persistence layer and align it with actual usage.

Verified Workflow

1. Find Database Files

# Search for SQLite databases
find . -name "*.db" -o -name "*.sqlite" 2>/dev/null

# Common locations:
# - data/trading.db (main database)
# - data_cache/cache.sqlite (OHLCV cache)
# - data_cache/market_data.db (historical data)

2. Analyze Table Usage

# Check which tables have data
sqlite3 data/trading.db "SELECT name, (SELECT COUNT(*) FROM sqlite_master WHERE type='table' AND name=m.name) FROM sqlite_master m WHERE type='table';"

# Check row counts per table
sqlite3 data/trading.db "SELECT 'predictions', COUNT(*) FROM predictions UNION ALL SELECT 'prediction_outcomes', COUNT(*) FROM prediction_outcomes;"

3. Find Methods Called But Not Implemented

# Search for method calls in codebase
grep -r "tracker\\.get_recent_accuracy\\|db\\.some_method" --include="*.py"

# Check if method exists
grep -n "def get_recent_accuracy" alpaca_trading/**/*.py

4. Find Unused Table Methods

# Search for table name usage across codebase
grep -r "equity_symbols\\|symbol_features" --include="*.py" | grep -v "db_manager\\|trading_db"

# If no results outside schema/manager, table is unused

5. Clean Up Schema

Remove unused tables from SQL schema file:

-- REMOVED (dynamically computed or unused):
--   equity_symbols     - Alpaca API provides dynamically
--   symbol_features    - Computed in-memory during selection
--   universe_selections - Console logging sufficient
--   portfolio_performance - Redundant with other tracking

6. Remove Unused Methods

After confirming no callers exist:

# Remove entire method sections that reference removed tables
# - upsert_symbol()
# - get_active_symbols()
# - mark_symbol_delisted()
# - upsert_symbol_features()
# - get_latest_features()
# - log_universe_selection()

Failed Attempts (Critical)

Final Parameters

# Tables to keep (v2.6)
predictions: true          # Core - all predictions
prediction_outcomes: true  # Core - actual results
prediction_metrics: true   # Aggregated daily metrics
rl_models: true           # Model registry
timeframe_selections: true # Multi-timeframe tracking
backtest_results: true    # Backtest history

# Tables removed (v2.6)
equity_symbols: false     # Alpaca API provides
symbol_features: false    # Computed in-memory
universe_selections: false # Console logging sufficient
portfolio_performance: false # Redundant

Key Insights

• Schema drift is normal: As code evolves, some tables become obsolete
• Grep before delete: Always search for method/table usage before removal
• CREATE IF NOT EXISTS: Makes schema changes safe for existing databases
• Foreign keys may reference removed tables: Remove FK constraints too
• Document what was removed: Future maintainers need to know why

Audit Checklist

1. List all database files and their sizes
2. Check row counts for each table
3. Search for method calls that might not exist
4. Identify unused tables (no callers outside schema/manager)
5. Remove unused tables from schema
6. Remove corresponding methods from db_manager
7. Update module docstrings
8. Document removed tables in schema comments

References

• alpaca_trading/data/db_manager.py: Database manager class
• alpaca_trading/data/trading_db.sql: SQL schema definition
• docs/persistence.md: Persistence layer documentation