I can’t deal with some inconsistency I found between the data presented on the front-end and the data available in the API.
The items marked in yellow match, while the TC has a different value in the API.
I found many such TC glitches for rounds below 336 era.
This spoils the results of my long-term tests of the model. Any idea what is the problem here?
Interesting. Was your payout actually negative for that round? If you look at the round page for 326 (Numerai), it shows the two positive values for both corr & tc yet the negative payout. What’s real? (I’d bring this up in support over on rocketchat to get some attention.)
I can see the same TC inconsistency for my models from round 335 back. Payout match
Hey all, thanks for bringing this to our attention, I looked into this and found the inconsistency that @kowalot reported is related to a backfill for 60d TC I did a while back. The values shown by v3UserProfile (and the round detail page that mlivako shows) are from this backfill, so this is a display issue and did not affect the final payout.
This is why @kowalot sees positive corr/TC but negative payout, in reality round 326 the 20d TC used for paying out pl18_piorun was -0.05083876272117207 which would explain the negative payout.
I’ll be deleting these excess scores to bring these displays in line with reality.
@kowalot I’ve sent your model pl18_piorun a .1 NMR bounty for finding the bug, thanks again for reporting!
Thx!!! Can you also confirm the fix to the API results (v3UserProfile/ roundModelPerformances)? I still see positive TC for round 326.
These should be patched now, thank you for your patience!