Define interface between a job and the backend

[njbrake]

Motivation

Currently, jobs save a results.json file which is loaded by the backend and saved in the database. In order to add a level of abstraction between the job and the backend, let's split this out into a slightly more enforced interface, where the job saves a parameters, artifacts, and metrics json file with a loosely defined schema. The backend then load these files into pydantic validated classes before saving it to the database.

Alternatives

1. Do nothing
2. Instead of multiple json files, save everything into a single results.json file

Contribution

I'll do it!

Have you searched for similar issues before submitting this one?

• Yes, I have searched for similar issues

[njbrake]

During investigation I realized that currently there isn't any interface at all between an evaluation/evaluation_lite job and a backend, because the result.json is passed from the s3 bucket to the frontend without it being opened or validated by the backend. So this here issue I think is blocked until we de-couple the frontend and job artifact file structure, see #678

[ividal]

Note that @javiermtorres had proposed a refactoring of jobs that might be relevant here. Just want to link the 2 of you in case it's useful!

[njbrake]

Note that @javiermtorres had proposed a refactoring of jobs that might be relevant here. Just want to link the 2 of you in case it's useful!

thanks! I think this issue is relevant to that refactor request: #706