nubra python sdk vs raw rest api for live algo trading?
first live version is already built using nubra’s python sdk.
honestly the sdk is clean. auth, instruments, orders, positions and market data took much less code than building the whole http layer myself.
now deciding whether to keep it directly in production or put a thin internal client over nubra’s v3 rest api.
things pushing me toward raw rest:
- exact timeout control
- logging the complete request/response
- reproducing failures with curl
- fewer surprises during sdk upgrades
- easier use from non-python services later
but this may just be engineering trust issues. the sdk already removes a lot of boring plumbing and i can pin the version.
other option is wrapping the official sdk behind my own `Broker` interface, so strategy code never imports nubra directly.
what are people actually doing in live systems?
official sdk directly? wrapped sdk? generated rest client? or sdk for market data and raw rest for execution?
main concern isn’t speed. it’s being able to explain exactly what happened after one ugly order failure six months from now.
