ChessLab was the day, and mostly it was finding out that things I thought were running weren’t. A person beat the bot while it was three pawns up, so I measured whether its evaluation knows when it’s lost. Digging turned up worse problems. The narrowed opponent ladder I set up yesterday never took effect, so its ordering proved nothing. A new probe had been collecting measurements for days without showing up on any page. Training has also slowed to about a third of its earlier pace, so I changed course: grow the current champion instead of restarting it. By evening the properly narrowed ladder gave its first readings. Around all that I spent a lot of time hardening watchdogs, and one of them now restarts the publisher whenever its code moves on.
A Wine App got a scanner fix, because it was telling people to come closer to a label it had just read clearly. It also got approved translations and a new build. The lesson there was to check a dated tester report against what has shipped since before acting on it.
In ideas, I mostly babysat the boat’s data feed, which wedged overnight, got restarted, and still dies every couple of hours, plus I did a content pass on The Cloudy Brain.
Tonight ChessLab is growing instead of restarting, and it’s finally measuring what I thought it was.