Photo capture

How FRA Flow handles photos: camera flow, automatic compression, and AI classification into BS 9792 evidence categories.

Photo evidence is the spine of an FRA. FRA Flow's job is to make each photo cheap to capture, small enough not to wreck your data plan, and pre-classified so the report can group findings by evidence category without you tagging anything by hand.

How a photo is captured

Tap + Observation on a location, then tap Photo. The iPad opens straight to the rear camera. Take the photo. The image appears on the next step.

You can also pick a photo from the camera roll instead. Useful when you took the picture earlier and want to attach it to an observation now. Tap Photo, then choose the library option from the picker.

Multiple photos per observation are supported. Add the first photo through the capture flow, save the observation, then tap the observation card to attach a second photo.

Why photos are auto-compressed

Site visits routinely capture 50 to 100 photos. At iPad-camera quality (4 to 8 MB per shot) that is half a gigabyte of cellular data on the way home. FRA Flow compresses every photo to a 2-megabyte ceiling JPEG before it leaves the device.

The compression happens locally, before upload. No quality loss that matters at the resolution a report renders at. No extra button to remember to press.

Evidence categories

Every photo is automatically tagged with one of nine evidence categories drawn from the BS 9792 vocabulary:

• Fire door
• Signage
• Detection device (smoke alarm, heat detector, etc)
• Emergency lighting
• Escape route
• Housekeeping (storage in a corridor, blocked exit)
• Compartmentation (penetrations, missing seals)
• Bin store
• External risk (cladding, neighbouring building)
• Other

A category chip appears next to the photo in the observation card. The chip has four states:

• Pending. The classifier has not run yet, usually because the upload has not drained.
• Suggested. The classifier returned a category but you have not confirmed it. You can keep, change, or clear the suggestion.
• Confirmed. You agreed with the suggestion, or you set the category yourself.
• Unclassified. No suggestion (the photo was unclear, or the classifier failed). You can set the category manually.

The category drives how the photo appendix groups in the final report (one section per category) and how the action plan picks trade categories for follow-up actions.

What to actually photograph

A few patterns that pay off downstream:

• One subject per shot. If a fire door has two issues, shoot the door once for context, then a close-up on each issue. Three observations, three photos. Splits cleanly in the report.
• Capture the location. A wide shot of the corridor or stair landing before the close-up. Reviewers and landlords lose track of which floor they are looking at otherwise.
• Get the make and model. For fire doors, detection devices, emergency lighting, a shot of the manufacturer label is gold. Action close-out (a future milestone feature) reuses this to cross-reference replacement specs.

When the network is patchy

Photos saved while offline queue locally and upload when connectivity returns. The observation appears in the workbench with a "queued" pill until the photo finishes uploading. You can keep capturing while older photos are still draining.

If the upload fails (rare, usually a 5xx from the storage provider), it retries automatically with backoff. If the queue is stuck, the sync-stuck troubleshooter covers the clearing flow.

Where to go next

• Voice notes for the recorded- observation flow that runs alongside photos.
• Risk levels explains how Low / Medium / High map to the report and the action plan.
• Working offline covers the queue behaviour during a no-signal site visit.