Automating the statement processing process

[markwhiting]

Files

1. statements.csv
2. features/statement_group_#.csv
3. raw_statements/GPT_statements_from_tuesday.csv
4. features/ with files that return bundles of features according to the feature protocol

Protocols

Statement

CSVs with:
statement, elicitation, committer

Feature

Files that produce ratings with:
name, version, value, type

Checks

Statements

• Add any missing or expired features?
• Is source information valid?
• Duplicate statements?
• Are the properties added correctly?
• Is the table valid? Are all features type correct?
• Are the output files the same as they should be? Go ahead and merge : Add commit with correct file (or fail)

Processing

1. Create a branch and it gets a new CSV with the branch name (fancy)
2. User adds a new row to some CSV or a new CSV in raw_statements
3. get all files in raw_statements
4. run checks
5. compile output files

Cases

1. new feature: go through every existing statement and get the new feature for it
2. updated feature: go through every existing statement and get the updated feature for it
3. new statement: go through every feature and get it for that statement

Ignoring for now

We won't deal with validating new features or working out how to run new features dynamically for now. For now let's assume that we trust how features are designed and we will set up their run mechanism manually on a per-feature basis.

Tests

@amirrr — think about this please.