purchase-evaluation

name	purchase-evaluation
description	Debug and fix FlowInCash purchase decision queue and traffic-light evaluation system
version	1.0.0
tags	["flowincash","finance","purchase-queue","traffic-light"]

Purchase Evaluation

Use this skill when the FlowInCash purchase queue or traffic-light MicroApp produces incorrect evaluations, stalls, or returns unexpected Red/Yellow/Green signals.

System Architecture

• FlowInCash (/media/bob/I/AI_Projects/FlowInCash)
  • src/services/ — purchase queue coordination, forecasting engine
  • src/models/ — purchase queue DB models
  • Evaluates against cash flow forecast, emergency fund, and goals
• traffic-light (/media/bob/I/AI_Projects/FliC-MicroApps/traffic-light/)
  • evaluatePurchase — Red/Yellow/Green signal
  • checkEmergencyFund — adequacy check
  • Stateless per-call, uses @flowincash/mcp-tools

Diagnosis Steps

1. Identify which system — main app purchase queue or ChatGPT traffic-light?
2. Check input data — purchase amount, category, urgency, current balances
3. Check for stale financial data — if Plaid hasn't synced, evaluations use outdated balances (see plaid-sync-recovery skill)
4. Trace the evaluation logic in src/services/ or traffic-light/src/conversation/

Common Issues

Symptom	Cause	Fix
Wrong signal	Threshold constants don't match user situation	Check ratio constants in traffic-light/src/
Queue stuck	Coordination service failed mid-evaluation	Check DB for stale status='pending' items
Main app and MicroApp disagree	By design — main app has full context, traffic-light is simplified	Not a bug

Validation

1. npm test in traffic-light (30 tests)
2. npm run test:unit in FlowInCash for purchase queue tests
3. Test with known safe, risky, and edge-case purchases