Edge teardowns

These are real lab decisions, including the ones that did not survive hold-out.

NEEDS WORKDefined → Validated → not promoted

AC + MACD trend regime on BTC and ETH 1h

Spot · BTC/ETH 1h · agent-drafted · train + hold-out

Process check

74 trades against a 126-trade floor. Hold-out: 31 trades, +7.25%.

Profitable on train and hold-out, but too few trades to trust. The +42% headline was the best of nine backtests.

KILLEDDefined → Validated → ✕

Gaussian channel on BTC and ETH 1h

Spot · BTC/ETH 1h · train + hold-out

Process check

53 and 70 trades against a 100-trade floor. Profit factor 0.09 and 0.14.

The written rule lost about 20% on the training window, with or without the session filter.

KILLEDDefined → Validated → ✕

RSI mean-reversion on ETH 4h

Spot · ETH 4h · train + hold-out

Process check

15 trades against a 30-trade floor. 2 on the hold-out.

The strategy did not almost work. It never produced enough trades to be eligible.

Decisions, not returns. Any figure shown in a teardown is backtested or simulated and is not indicative of future performance.