This is a brief summary of a breakout discussion on Wed Oct 28 2020 during the LSSTC Enabling Science 2020 Brokers Workshop. Further discussion on open questions is very much encouraged!
Standardized Broker Outputs - Users are likely to incorporate the results of multiple brokers (e.g., classifications, prioritizations) for their science, and some standardization in the format and access mechanisms for these results would help users. This would help brokers share results as well; are any brokers taking action in this direction?
Databases vs. Streams - It is becoming more evident that many users’ science goals could be fulfilled with broker results that are available via queryable databases with ~hours latency instead of streams with ~minutes latency. Some urgent science cases (kilonovae) might still require stream-based user access?
Modules - The ability for users to directly access broker-embedded modules (e.g., light-curve classification) would be useful for science (e.g., changing parameters, using different training sets, reproducibility). Do cloud environments better facilitate modular access?
Reproducibility - Some users have a strong science requirement to be able to reproduce broker activity, thus dockerized and containerized software and versioned input catalogs (training sets) are also needed for user science.
System Health - Broker self-reporting on system health (e.g., downtime, sky regions being unprocessed) would be useful for science being done using multiple brokers, for example to provide context for ‘minority reports’ when only some brokers flag an alert as high priority.