Direction is the lever
The question that decided version 3 was not where to enter but which way. The raid-and-gap entry is fixed; what changed between versions was only the rule that decides the day's direction, and that rule moved the annual result from a profit factor of 0.7 to 2.1.
The numbers: with no direction rule, pure raid and reversal as taught in the liquidity-raid literature, the system loses in every variant tested, profit factor around 0.7. With the colour of the 09:30 to 10:00 candle as direction it makes 4 845 $ on 158 trades at a profit factor of 1.42. With an oracle that knows the day's true close in advance, the same entries make 11 400 $. The oracle is not tradable, but it bounds the problem: direction is worth more than any entry refinement.
How good is the thirty-minute candle? It matches the day's true direction on 70 % of days. The previous day's candle matches on 52 %, the Asia session on 45 %, London on 48 %. Multi-day momentum (three-day, five-day, twenty-day average) is worse than the candle alone and sometimes negative. The market's first half hour knows more about the day than the night before it.
Version 3 adds a second reading: the pre-market move from 07:00 to 09:30 must point the same way as the candle. Days where they disagree are skipped. This removes 71 trades from the year, most of them losers, and lifts the profit factor to 2.14 while cutting the drawdown from 916 $ to 573 $. The window start is not fragile: from 04:00 to 07:00 every start gives a profit factor between 1.9 and 2.2; from 08:00 the filter breaks down, because too little of the pre-market move is left.
The practical reading: a trader who improves the direction call by ten percentage points gains more than one who perfects the entry. That is where the next version will be spent.