The proof deck put the close-up where the wide shot belonged
The report sent to advertisers shows two photos per board. If the close-up was taken first, it printed in the wide-shot slot. One field was doing three jobs.
The campaign report an advertiser receives shows two photographs of each board: a wide shot showing the board in its surroundings and a close-up showing the flex. If your crew took the close-up first, the report printed it in the wide-shot slot. If a board had five photographs, the report quietly used the two oldest.
The advertiser sees a slide that looks wrong, and draws a conclusion about the care you take. The error sits in the document whose whole job is to show the work was done.
What was actually going on
The system had one field for the kind of photograph, and that one field was being asked three different questions: why the visit happened, when in the campaign it happened, and what sort of picture it was.
It could answer the first two. It could not say whether a given picture was wide or close. So the report builder did the only thing left to it. It took the first two photographs in the order they were taken and called them the wide shot and the close-up.
Videos were worse. A clip carries no framing, so every video had no shot type at all.
What we changed
The one field became two: work type, meaning why the visit happened, and shot type, meaning what the picture shows. The report now prefers a wide shot and then a close-up, and falls back to capture order only for old photographs that carry no shot type, so history renders as it always did.
The phone asks for the shot type as the photo is taken. Later the capture screen was rebuilt around three cards on one screen. Wide shot and close-up are required, a newspaper shot is visibly optional and never counted as missing, and each card says in words whether it is done. An empty card shows last visit’s photo faded behind it, because a monitoring shot is only comparable to the previous one if it is taken from roughly the same place.
Existing photographs were brought across on a test copy first: 169 campaign photographs, 162 given a work type, 7 left without one because they had no category. None was given a shot type.
What it did not fix
We did not guess the framing of old photographs. Nothing in the history recorded which shot was which, and inventing it would have corrupted the report. So on older boards the report still falls back to capture order, and can still print a close-up as a wide shot. That is left alone on purpose.
About 150 older photographs and every video still carry no shot type. A later check of the field data found that the production phone app then in use predated shot types, so production photographs carried none.
The pattern, for anyone who reports to clients
If one field in your data is being used to answer more than one question, separate them before anything starts to depend on the combined answer.
And when you add a distinction that history never recorded, do not back-fill it with a guess. A blank you can see is easier to live with than a value that is wrong and looks right.
Where this ends up
The report an advertiser receives is built from these photographs by AdBoard, the system that runs for Gold Sign Media, an outdoor media operator. The shot type is now recorded at the board, so the report no longer guesses.