feat(strategy): add MACD alignment requirement for buy signals

[claude]

Summary

• Add configurable option to require MACD histogram to be above a threshold before allowing buy trades
• Prevents buying during declining momentum conditions (falling knife scenarios)
• Feature disabled by default (opt-in) with threshold of -0.5

Changes

• config/settings.py: Add require_macd_alignment and macd_alignment_threshold config parameters
• src/strategy/signal_scorer.py: Implement MACD alignment check in calculate_score() - when enabled and histogram < threshold, buy signals downgrade to hold
• src/daemon/runner.py: Pass new settings to SignalScorer initialization
• .env.example: Add documentation for new config parameters
• tests/test_signal_scorer.py: Add comprehensive unit tests for MACD alignment logic
• tests/test_trading_daemon.py: Add mock settings for new parameters

Context

From post-mortem analysis (2026-01-09):

• All 4 losing buy trades had MACD histogram showing clear negative momentum (-37 to -85)
• RSI and Bollinger indicated oversold/buy opportunity, but MACD said momentum still declining
• Requiring MACD alignment would have blocked every losing trade in that session

Test plan

• Unit tests for MACD alignment disabled by default (backward compatibility)
• Unit tests for buy blocked when MACD histogram below threshold
• Unit tests for buy allowed when MACD histogram above threshold
• Unit tests for sell/hold signals not affected
• Unit tests for configurable threshold
• Unit tests for combined with other filters (HTF, trend)
• Run full test suite: python3 -m pytest tests/ --ignore=tests/test_coinbase_integration_live.py -v

Closes #331

🤖 Generated with Claude Code

[claude]

PR Review: feat(strategy): add MACD alignment requirement for buy signals

Overall Assessment: ✅ APPROVE

This is a well-implemented, targeted improvement to the trading strategy that addresses a documented post-mortem finding. The code is clean, well-tested, and follows the existing patterns in the codebase.

---

🟢 Strengths

1. Post-Mortem Driven Development

• Feature directly addresses real losses from 2026-01-09 post-mortem analysis
• All 4 losing trades had negative MACD histogram (-37 to -85)
• This would have prevented every losing trade in that session

2. Conservative Default Configuration

• Disabled by default (require_macd_alignment=false) - opt-in behavior
• Default threshold of -0.5 allows slight momentum decline
• Backward compatible - existing users unaffected

3. Well-Structured Implementation

• Follows existing filter patterns (HTF, trend filter)
• Clean placement after action determination, before confidence calculation
• Proper metadata tracking (_macd_alignment_blocked)
• Good structured logging for observability

4. Comprehensive Test Coverage

• 12 test cases covering all edge cases:
  • Default disabled (backward compat)
  • Buy blocked with negative histogram
  • Buy allowed with positive histogram
  • Sell/hold signals unaffected
  • Configurable threshold
  • Missing MACD data handled
  • Combined with other filters

---

🟡 Minor Observations (Low Priority)

1. Metadata key naming convention

metadata["_macd_alignment_blocked"] = 1 if macd_alignment_blocked else 0

Other metadata uses underscored prefix consistently. This matches the pattern - looks good.

2. Conservative behavior on missing MACD data (Line 1237-1243)
The code blocks buys when MACD data is unavailable. This is the right choice for a financial system - fail-safe rather than fail-open. However, this could mask data issues if the MACD calculation is broken.

Suggestion (optional): Consider incrementing a counter/metric when MACD data is missing to catch any data pipeline issues early. Not blocking merge.

3. Test robustness
Some tests use conditional assertions like:

if result.score >= macd_alignment_scorer.threshold:
    assert result.action == "buy"

This is acceptable given the stochastic nature of the test data generation, but doesn't guarantee the condition is actually exercised. The seed-based approach (np.random.seed()) helps with reproducibility.

---

🔒 Security Considerations

No security concerns. The changes:

• Only affect internal trading logic
• Don't introduce new external inputs
• Don't modify protected paths (safety systems, exchange clients)
• Are fully configurable via existing config system

---

⚡ Performance Considerations

Minimal performance impact:

• Single comparison operation (macd_histogram < threshold)
• Only executed when feature is enabled AND action is "buy"
• No additional API calls or database queries
• MACD histogram is already calculated as part of indicator computation

---

✅ Checklist

• Code follows existing patterns in signal_scorer.py
• Configuration added to settings.py with proper Field validation
• .env.example updated with documentation
• Test coverage is comprehensive
• Feature is disabled by default (backward compatible)
• No security vulnerabilities introduced
• Conventional commit format used correctly
• PR properly references issue (Add MACD alignment requirement for buy signals #331)

---

📝 Summary

This PR adds a valuable safety feature to prevent buying during declining momentum conditions. The implementation is clean, well-tested, and follows all codebase conventions. The opt-in design ensures no disruption to existing deployments.

Recommendation: Merge when CI passes.

---

Review by Claude Opus 4.5