This page describes the workflow we use for the DataRefuge project, both at in-person events and when people work remotely. It explains the process that a url/dataset goes through from the time it has been identified by a seeder & sorter as “uncrawlable” until it is made available as a record in the datarefuge.org ckan data catalog. The process involves several distinct stages, and is designed to maximize smooth hand-offs so that each phase is handled by someone with distinct expertise in the area they’re tackling, while the data is always being tracked for security.
Seeders and Sorters canvass the resources of a given government agency, identifying important URLs. They identify whether those URLs can be crawled by the Internet Archive’s webcrawler. If the URLs are crawlable, the Seeders/Sorters nominate them to the End-of-Term (EOT) project, otherwise they add them to the Uncrawlable spreadsheet using the project’s Chrome Extension.
Researchers inspect the “uncrawlable” list to confirm that seeders’ assessments were correct (that is, that the URL/dataset is indeed uncrawlable), and investigate how the dataset could be best harvested. Research.md describes this process in more detail. We recommend that a Researchers and Harvesters (see below) work together in pairs, as much communication is needed between the two roles. In some case, one same person will fulfill both roles.
Harvesters take the “uncrawlable” data and try to figure out how to actully capture it based on the recommendations of the Researchers. This is a complex task which can require substantial technical expertise, and which requires different techniques for different tasks. Harvesters should see the included Harvesting Toolkit for more details and tools.
Checkers inspect a harvested dataset and make sure that it is complete. The main question the checkers need to answer is “will the bag make sense to a scientist”? Checkers need to have an in-depth understanding of harvesting goals and potential content variations for datasets.
Baggers do some quality assurance on the dataset to make sure the content is correct and corresponds to what was described in the spreadsheet. Then they package the data into a bagit file (or “bag”), which includes basic technical metadata and upload it to final DataRefuge destination.
Describers creates a descriptive record in the DataRefuge CKAN repository for each bag. Then they links the record to the bag, and make the record public