Watch a surveyor walk a property and something interesting happens: they narrate. Not for anyone in particular — for the record. "Third bedroom. Hairline crack above the architrave, runs about 400 mil. Wasn't in the March baseline." The camera points at the crack while the words land on it.
That sentence is doing enormous work. It says where you are, what matters in the frame, how big it is, and how it relates to the last visit. The footage alone says none of that. The footage shows a wall.
Half the walk is not the walk
Most tools that touch site work read one channel and drop the other.
Tools built around photos and forms see surfaces without significance. They can tell you there's a crack in the image; they can't tell you it's the one you flagged as new, or that you said it ran above the architrave, or that it's the reason you want a follow-up inspection. The meaning was in the audio they never heard.
Dictation tools have the opposite problem. They hear your claims and never see the evidence. You get your own words back, tidied — but nothing connects "hairline crack, about 400 mil" to the frame that shows it. When the report is challenged, the transcript is just assertion.
Either way, the same thing gets lost: the binding between what you saw and what you said about it. And that binding is the walk. It's the difference between footage and findings.
Reading both channels as one thing
Sully treats the narration as the spine of the footage. The transcript — timestamped, word by word — primes the visual reasoning, so the model reads each stretch of video already knowing what you said while you shot it. It knows what you were looking at when you said it.
That's what lets the summary come out grounded instead of guessed. Every line in it traces back to the exact frame and the exact words behind it — click a finding, see the moment. Checking the draft takes seconds, because the draft carries its own evidence.
It also changes how you capture. When the tool understands narration, you can work rough: talk naturally, shoot loosely, keep your hands free for the torch and the ladder. There's no form to fill in on site, because the walk itself is the input.
The walk already contains the report
Here's the belief underneath the product: by the time you get back to the car, the report already exists. It's in the footage and the words — every finding, every location, every measurement you called out loud. The write-up grind is just the manual labour of extracting it.
The narration is the index to the evidence. Read them together, and the extraction is a job for a machine. Read them apart, and it's your evening.