Important: These two files are historical prediction/learning records, not raw one-minute OHLCV candle history. They can audit the old forecast behavior, but they are not enough by themselves to calculate correlations between candle indicators and future BTC returns.
What this tells us
A correct-direction rate above 50% alone is not proof of trading edge. These records have overlapping 15-minute outcomes, and the baseline depends on the direction mix. A proper test needs time-ordered, non-overlapping holdout predictions and (for Kalshi profitability) contract prices, spreads, fees, and settlement rules.
Original prediction scorecard
V28 learning records
Data limitations and next step
- The scorecard contains 361 rows, including pending outcomes.
- The V28 file contains prediction records and regime labels, but no OHLCV fields like candle open, high, low, close, and volume for each minute.
- Without raw candles, claims about momentum, wick size, volume spikes, volatility, or correlations would be guesses.
- Import a continuous candle CSV into Prediction Lab. Then run its holdout test and inspect the correlation table before trusting a forecast.