SIgh, OK that run failed because the scores (but not hits order) suddenly changed. But that was because of this crazy performance bug in the nightly benchy: https://github.com/mikemccand/luceneutil/issues/616 (thank you neoremind!)
We were indexing the body field twice! That issue is still open, but the double body indexing was fixed with https://github.com/mikemccand/luceneutil/pull/622 (thank you Patrick!) And that fix then altered the BM25 scores, which caused 10/9 and 10/10 benchy to fail. I just kicked off a regold run... let's see in ~26 hours. Mike McCandless http://blog.mikemccandless.com On Fri, Oct 9, 2026 at 10:05 AM Alan Woodward <[email protected]> wrote: > Yes, I’m looking forward to the next data point which should show much > better performance! > > On 9 Oct 2026, at 09:41, Michael McCandless <[email protected]> > wrote: > > Oh and it looks like we got one data point of the new benchy query: > https://benchmarks.mikemccandless.com/MustNotIODVRange.html > > Thank you @kkewwei! > > Mike McCandless > > http://blog.mikemccandless.com > > On Fri, Oct 9, 2026 at 8:34 AM Michael McCandless < > [email protected]> wrote: > >> Hello, yes, sorry it's been broken for too long! I attempted to >> resurrect and we got one data point a few days ago but now it has failing >> spinning magnets drive. I will try to fix soon, but I'll be in Glasgow >> shortly, far away from beast3's failing hard drive! >> >> Mike McCandless >> >> http://blog.mikemccandless.com >> >> On Fri, Oct 9, 2026 at 5:43 AM Dawid Weiss <[email protected]> wrote: >> >>> >>> My guess is it's broken. Mike? >>> >>> D. >>> >>> On Thu, Oct 8, 2026 at 4:19 PM Alan Woodward <[email protected]> >>> wrote: >>> >>>> Was looking at https://benchmarks.mikemccandless.com/ to see if the >>>> new benchmark that @kkewwei added last week was recording data, and the >>>> last update seems to have been on 5th October. Has something broken or are >>>> things just taking a really long time to run now? >>>> >>>> - Alan >>>> >>> >
