Modeling negated transactions

[gordonje]

California campaign finance committees are required to itemize returned contributions on the same schedule that includes received contributions. The real world situation would be something like:

• In May, John Doe gives gubernatorial candidate Gavin Newsom $1,000
• In October, after learning that John Doe is a total scumbag, Gavin is embarrassed into returning the $1,000

In our source data, we have a line item for the original contribution and another with a negative amount for the returned contribution:

filer	contributor	amount	date	type
GAVIN 4 GOV	JOHN DOE	1000.00	5/1/2017	Contribution
GAVIN 4 GOV	JOHN DOE	-1000.00	10/1/2017	Returned

In mapping these records to the Transaction model, my initial thought was to flip the sender and receiver and take the absolute value of the amount. So the above source records would become:

sender	recipient	amount	date	classification
JOHN DOE	GAVIN 4 GOV	1000.00	5/1/2017	Contribution
GAVIN 4 GOV	JOHN DOE	1000.00	10/1/2017	Returned

However, @palewire and I discussed further and decided against this approach. We're worried about the potential for inaccuracies when summing the amount field. Instead, we're planning to leave the source values more or less unchanged in loading the Transaction model:

sender	recipient	amount	date	classification
GAVIN 4 GOV	JOHN DOE	1000.00	5/1/2017	Contribution
GAVIN 4 GOV	JOHN DOE	-1000.00	10/1/2017	Returned

Have others seen similar use cases in other jurisdictions and, if so, does our approach for fitting it into our shared models make sense?

If we agree this is proper use, then we might expand the description on Transaction.amount to say that negative numbers are allowed and why.

[aepton]

Added this to https://github.com/opencivicdata/docs.opencivicdata.org/pull/104 - added an optional is_rejected boolean to the amount field, and explained in the description not to use that field if you're setting transaction values to negative for returned contribs, etc.