The problem
The client publishes a magazine. Before assigning a feature, an editor is supposed to check whether the magazine already ran the story, what it said, and which photos ran with it. That check happened in shared drives and in an issue archive with inconsistent filenames. The editors estimated 25 minutes when the memory was fuzzy, and they sometimes skipped it and repeated a feature a reader then pointed out.
This engagement is the publishing side of the work. It does not dispatch aircraft, clear a flight, or make a maintenance decision. Those calls stay with a qualified person, and this study does not describe them.
Why the archive, not a web search
A web search returns the magazine's own public posts mixed with everyone else's coverage, and it misses the issue PDF that never went on the open web. The editor needs the prior feature from the archive the rights team has cleared for internal use, with the issue and the page. A result from outside that archive is not an assignment brief.
Architecture
The corpus is the rights-cleared archive
Issues, captions, and photo credits the rights team marked for internal editorial use are in the index. Freelance material the contract does not allow back into a new assignment is not indexed, even if the PDF is on the drive. The rights flag is the filter, applied before search.
The result is the issue and the page
The editor sees the passage, the issue date, and the page. A summary with no page is not returned. They open the PDF at that page, which is the check they were already supposed to do, without the filename hunt.
Photo credits stay attached
A hit that is a photograph includes the credit line stored with it. The tool does not offer the image as reusable. Reuse is a rights decision a person makes from that credit, outside this search box.
The editor still assigns and still edits
Search does not draft the new feature and does not publish. It answers 'have we run this, and where.' The assignment stays on the desk.
Challenges and tradeoffs
- Filenames were not issue dates. Half the archive was named by the story slug. We took the issue date from the masthead on the first page, checked it on a sample with the archive editor, and stored that date as the field editors filter on.
- A rights flag was missing on a decade of older issues. Those issues were excluded, which the editors felt as a hole. Excluding them was correct. The rights team cleared a defined backfile during the project, and only that backfile was added. The rest stays out until it is cleared.
- Captions and body text were indexed as one blob. A search for a person returned a photo caption credit as if it were a feature. We keep caption, credit, and body as separate fields, and the result says which one matched.
Results
On the timed tasks the editors set, median time to the prior feature fell from 25 minutes to 3. Every hit in that test came from the rights-cleared archive and opened on a page. Material without a rights flag did not appear. Editors still assigned the feature and still edited it.
Nothing in this result is an aviation operations number. The magazine and aviation page covers both kinds of work. This study is the archive.
What we'd do differently
We would have asked the rights team for the backfile boundary before demoing search to the editors. The first demo included a drive folder that had not been cleared, and we pulled it the same day. The demo should have been incapable of showing it.