The RSS scanner started with 9 themes and hardcoded keyword→ETF mappings. "War articles → defense stocks" works because the logic is obvious. But we asked: how do we expand without drowning in noise? And more importantly — how do we measure whether any of this actually works?
The Expansion: 7 New Themes
We added themes where the keyword-to-asset connection is undeniable:
China/Taiwan tensions — "Taiwan strait, chip export ban, PLA navy" → SHORT FXI, LONG defense.
Currency stress — "dollar collapse, currency crisis, dedollarization" → LONG GLD, LONG UUP.
Food/agriculture shocks — "grain shortage, crop failure, food crisis" → LONG DBA.
Real estate cracking — "housing crash, foreclosure spike" → SHORT XHB, SHORT VNQ.
Supply chain disruption — "port shutdown, Suez Canal, container shortage" → LONG USO, LONG XLI.
Crypto shock — "exchange collapse, stablecoin depeg" → SHORT XLF.
Pandemic/health — "new outbreak, WHO emergency" → SHORT SPY, LONG XLV.
Cross-Theme Amplification
When correlated themes fire together, the signals get a confidence boost. War + oil disruption both spiking? That's not coincidence — amplify both.
military_conflict × oil_energy → +10% military_conflict × sanctions → +8% recession_fear × banking_crisis → +12% inflation × oil_energy → +8% china_taiwan × supply_chain → +10% china_taiwan × military_conflict → +8% currency_stress × recession_fear → +8% pandemic_health × supply_chain → +10%
The Scoring System
Every signal gets scored against actual market outcomes at T+1, T+3, and T+5 trading days. The scorecard tracks accuracy by theme, source, symbol, and confidence band.
Stage 1 (now): Forward measurement — score every signal, $0 cost. Stage 2 (week 2-3): AI analyzes failures — pennies/week. Stage 3: Systematic keyword optimization — one-time $5-20. Stage 4: AI pre-filter on low-accuracy themes — $1-3/day targeted. Stage 5: Continuous improvement loop.
Don't add AI everywhere hoping it helps. Measure first, find the failure points, then add AI only where the data proves it's needed.
You can't improve what you can't measure. Score everything. Fix what's broken. Leave what works alone.