Initial High Level Goals

[ClimbsRocks]

• track predictions between training and serving envs
• track feature values between training and serving envs
• track feature values over time at serving time
• track missing features at serving time
• Be implementable in different environments
• Have both live and shadowing models

Using Concordia, you should be able to rapidly have confidence in your shipped ML models.

If everything's working as expected, you should be able to see that, and heave a sigh of relief.

If things are not going according to plan, again, you should be able to see that rapidly, and nearly as quickly see the root cause of those discrepancies.