Ideas for improvements

[josmac69]

Hi Pavlo, first of all, Happy New Year 2026 and All the Best and thank you very much for your work on confelo.

I used confelo for the first ranking of talks for coming German PostgreSQL Conference 2026 I have some suggestions for improvements:

1. Confelo presents talks still in the same order even in repeated cycles, which unfortunately means that there is bias of strong / weak neighbors. Strong talk among other strong talks can get bad score, weak talk among even weaker talks can get much higher score than it really deserves. And due to the same order of talks in second and further cycles of comparison, results seem to quickly convert to final order.
   My suggestion would be to randomize order of talks for each cycle of comparison until ranking of selected top N is really stable.

If you could please implement this change soon, I would really like to use confelo for ranking talks for PGConf.DE, but that ordering bias is a problem for me. BTW, the same strong/weak neighbors bias applies also when people use just ranking web page on postgresql.eu - here that ordering bias seems to have even stronger influence. So, I believe updated version of confelo will be very useful in the future.

2. Print value of "target accepted" on the working screen and if it is not complicated, allow to change it proactively during the session.

3. When I want to continue with previous session, confelo seems to not remember comparison mode - I started session with trio mode, but when I wanted to continue and did not specify comparison mode, confelo uses default duo.
   I found hot keys "p", "t", "q" for switching comparison mode - so no big deal with confelo not remembering preselected mode - maybe just document these keys among key shortcuts in the bottom.

Thank you very much!

[pashagolub]

Hello Josef!

Thanks for the idea!

1. Next proposals ordering was changed in Fix exported scores #18. The new algorithm prioritizes proposals with less comparisons, so should work smoothly after the first cycle. If you could try and tell me if this is enough, or you want full random on the start.

2. Good point. Will add to the screen. To change it, you can exit and re-run the confelo with lower/higher argument. It should resume from the last point with new target applied.

3. Yeah, that might be a bug. I always specify --comparison-mode trio at start, will check. Thanks

[pashagolub]

3 is fixed! Thanks!