Contrary to what you might expect, giving HaChu (and yourself) more time, doesn’t significantly improve the program’s ability.
Well, it depends how you look at it. It does improve the search ahead a little. But it doesn’t improve the evaluation function (all this is to be expected).
So the effect of extra time is to eliminate some horizon effects (so it doesn’t sacrifice so many pawns to “avoid” a bigger loss). And to give you more thinking time whilst it is thinking.
H.G.Muller has adapted the bot adaptor so it now runs on Linux. Therefore (with Hidetchi’s agreement), I am running HaChu as a bot on JUPITER. I tested it last night. A configuration parameter was wrong (now fixed), so it thought it had 60 minutes every 40 moves, instead of for the whole game, and so lost on time. But the game is still instructive:
When it runs out of time after move 95, material is still equal (only the edge pawns have been exchanged). But my positional advantage is huge, by my evaluation Function. HaChu thinks otherwise – xboard was showing +1.76 for the bot – roughly a Silver General, I think. Incidentally, I had used only 5 minutes of time. Mostly I thought in its time. This becomes much harder to do when it is playing faster.