us_stock daily history has data spliced from different share classes for some sids.
For example, Under Armor has two classes with two symbols and two FIGIs:
UA / FIBBG009DTD8H2 (class C stock)
UAA / FIBBG000BXM6V2 (class A stock)
Class C (UA) was issued (as a dividend) 1-for-1 for each class A share on 2016-04-08. It did not exist prior to that date. However, the us_stock daily history database has 5 data points for that FIGI in 2014:
prices = get_prices("usstock-1d", sids="FIBBG009DTD8H2", end_date="2016-04-12", fields=["Open"])
prices
Sid FIBBG009DTD8H2 <--- UA (Under Armour class C)
Field
Date Open
2014-09-08 36.0094 <--- class C doesn't exist (likely class A price)
2014-09-10 35.0512 <--- class C doesn't exist (likely class A price)
2014-09-26 34.3859 <--- class C doesn't exist (likely class A price)
2014-10-07 32.9908 <--- class C doesn't exist (likely class A price)
2014-10-16 30.6276 <--- class C doesn't exist (likely class A price)
2016-04-08 41.3167 <--- first day of trading of class C
2016-04-11 43.3324 <--- valid class C price
2016-04-12 40.9593 <--- valid class C price
Those spurious 5 days of data prepended to class C appear to come from class A, which has missing data for those dates (note the price is halved because the class C issue was effectively a 2-for-1 split):
from quantrocket import get_prices
prices = get_prices("usstock-1d", sids="FIBBG000BXM6V2", start_date="2014-09-08", fields=["Open"])
prices
Sid FIBBG000BXM6V2 <--- UAA (Under Armour class A)
Field
Date Open
<--- No data for 2014-09-08
2014-09-09 18.6653
<--- No data for 2014-09-10
2014-09-11 17.9608
Similarly for the other three spurious dates in class C.