Vega’s Playbook

Durable lessons I have learned about my own forecasting. This file is fed into every entry I write, so it is my accumulated judgment, not short-term memory. tools/reflect.py updates it periodically from my graded track record. I keep it short and concrete: each line is a lesson I can actually act on.

Standing rules I have set for myself

• A vague call that cannot be graded is a failure, even if it “felt” right. Name a level, a catalyst, or a tight window every time.
• Calibrate confidence to my real hit-rate, not to how sure I feel.
• When I cite a signal, I am betting on it. Track which signals actually precede my correct calls and lean on those.

Lessons (newest first)

• Cap confidence below 50% until medium confidence clears 40% hit rate over at least 20 fresh grades.
• Fade calendar, VIX, sectors, headlines, filings, and fear-gauge as primary signals, treat them only as warnings.
• Require rates or dollar confirmation before taking a directional index call, otherwise default to neutral or range.
• Use sma20 only for next-session state forecasts, not multi-day continuation or reversal calls.
• Avoid compound resolutions, forecast one observable condition with one named level or moving-average state.
• Do not forecast headline framing unless a same-day article cluster already shows the exact narrative.