Candles

Strategy matrix

Every cycle, the best-fit selector scores each enabled strategy applicable to the direction read by net risk:reward — expected gain ÷ (expected loss + charges), both computed from the structure's actual expiry payoff over the move the option chain's own implied volatility prices in, tilted by the signal's direction and conviction — and picks the highest score. A card surfaces only if that ratio clears the bar and the expected return on the margin it blocks clears its floor. Uncheck a strategy to remove it from consideration.

Conviction calibration — last 30 days

Did the underlying move the way each card called? Repeat cards (same instrument and direction within an hour) count once. Hit rate is the share that moved the called way; the range is a 90% interval — wide ranges mean too few cards to tell. If conviction worked, higher buckets would hit more often.

Calibration by event type — last 7 days

Direction is calibrated as a full fit (actual ≈ α + β·predicted), not just a flat bias — β below 0 means this event type's predicted and actual direction have been anti-correlated, so the correction inverts sign in proportion to how strongly the headline was scored, not by a flat amount. Both shrunk toward α=0 / β=1 / scale=1 until enough samples accumulate. See manthan/learning/calibration.py.

Reaction trend — average move since detection

Every tracked event, checked at fixed horizons after detection (not just the final resolution) — how the average reaction builds or fades over the following minutes.

Recently resolved events

Equity curve & capital deployed

Solid — account equity (starting capital + cumulative realized P&L). Dashed — capital/margin actually deployed that day.