- current_task cannot be used if market does not match user languages

[20180201]
- implement bad src generation
- integrate bad ref/src generation into DA task creation
- implement "rank each segment for all systems" behaviour
- check if "all"-"used" batches gives the right data for IWSLT?

[20181024]
- CampaignTeam owner's staff status is never enforced
- Unclear what is best way to model user replacement
  * Should a campaign team be the actual generator of user accounts?
  * Should there be a replacement action on user level?
- We need a !#$% redesign here...

[20181105]
- Focus on WMT18 release first:
  * Should allow to set up 70:10:10:10 and 88:12 data;
  * Should support direct and multi modal assessments;
  * Should contain instructions on how to setup/run;
  * Should contain code to compute results clusters;
  * Deadline: 11/11/2018.
- Annotation tasks should contain a citation note.
- Avoid too much abstraction, might be easier to simply have three, four
  direct assessment implementations as opposed to supporting all possible
  configurations. Some of those might never be used.
- Add receipt token output for MTurk or crowd sourcing integration.
- Do we need generic results output for easier extraction/processing?
- Export campaign status/stats via CSV export hook.
- Change URL scheme to be campaign-based, rather than type-based.