A photograph of a defect is worth exactly as much as your ability to prove three things about it: when it was taken, what it is a photograph of, and that nobody changed it afterward. Strip those away and you do not have evidence. You have a picture a determined opponent can wave off as taken anywhere, anytime, by anyone.
Most site photos have those facts and then throw them away. Here is how we keep them, and one place we deliberately do not use the obvious answer.
Capture time, not upload time
The single most important fact about a defect photo is when the condition existed. Not when the wifi came back.
A photo taken in a basement at 9:14 and synced at 4:30, when the supervisor finally reaches the office, has to carry 9:14. Upload time is worthless as evidence. It records the return of signal, nothing more.
So the capture time is read off the image on the device, at the moment the photo joins the record, and written into the record as its own field. It travels through the offline queue with the photo. Seven hours later the sync happens and the timestamp is still 9:14.
You can correct it, too, which matters more than it sounds. A photo imported from someone's camera roll a week after the fact will carry the wrong date, and pretending otherwise would be worse than letting a person fix it and leaving the change on the record.
Why we do not lean on GPS
The obvious way to prove where a photo was taken is the location the phone recorded. We do not use it, and the reason is worth stating plainly rather than dressing up.
GPS puts the photographer at a point, with an accuracy indoors that is generously described as poor. In a building a point is not enough even when it is accurate. The same coordinates on level 4 look into two apartments and a corridor. Concrete and steel make it worse. A tool that prints latitude and longitude under a defect photo has given you a number that feels like proof and settles nothing.
What settles it is the structure. The photo belongs to a defect, the defect belongs to a location in the project's tree, and that location is a real place someone named: Level 4, Apartment 2B, Ensuite. If the capture started by scanning the code on the door, the location was filled in by the building rather than by the person. If the defect is pinned to the drawing, the pin is stored against that revision and stays on it.
That chain is worth more in a dispute than a coordinate pair, because it resolves to somewhere a person can walk back to. Nobody has ever won an argument at a handover inspection by reading out a longitude.
We also strip the rest of the metadata on the way in, once the capture date has been taken off it. The photos are re-encoded so every downstream renderer, thumbnail, report, viewer, gets pixels the right way up. That is a deliberate trade: correct orientation everywhere, at the cost of the metadata block. We would rather tell you that than let you believe there is a hidden layer of forensic data in there.
Annotations never touch the original
This is the rule that makes the rest hold up: the original photo is immutable.
When a supervisor circles a crack, drops an arrow, or writes "regrout" on an image, none of it is painted into the source file. The markup is stored as its own data on the record and drawn over the original when you view it. The defect view shows the marked-up version, because that is what is useful on site. The evidence keeps the untouched original, because that is what is defensible.
The day someone claims a photo was doctored, you produce two things, the original and the markup layer, and the claim goes away.
Who, and what happened next
The last fact is identity, and it is the cheapest of the three. Every photo carries the signed-in user who added it, and every status change on the defect around it carries who made it and when. The photo is not a loose file. It sits inside a sequence: raised by this person, issued to that contractor, disputed on Tuesday, fixed with this second photo attached, signed off by someone who was not the trade that did the work.
A single photo proves very little on its own. A photo at a named location, inside a dated sequence of events that nobody could edit after the fact, is a different thing entirely.
What this is really for
Ninety-nine times out of a hundred nobody asks. The defect is captured, fixed, verified, closed, and none of this is ever looked at.
It is the hundredth time the system is built for. The contested handover, the insurance claim, the tribunal. On that day, the gap between a tool that quietly kept the facts and one that did not is the gap between evidence and a story.