Why alternatives fail
Every other approach to site data is a proxy for the truth.
Telemetry, time studies, and AI bolted onto manual reports each promise visibility. Each delivers a guess. Talk is cheap — and a proxy is just talk with a dashboard.
Vehicle telemetry
“We already track our fleet.”
Telemetry tells you where a machine is and whether its engine is running. It tells you nothing about the work being done, the sequence of operations, or why a cycle stalled.
A truck idling for 40 minutes is captured as ‘engine on.’ Whether it was waiting on a blast, a bogged heading, or a dispatch gap is invisible.
It is a proxy for activity, not proof of work.
Time studies
“We run a time study every few years.”
A $40–80K study captures a two-week snapshot. The observer effect changes the behaviour being measured. The result is stale the day it lands.
It samples reality. It does not record it. The 36 months between studies are a blind spot.
You pay for confidence in a number that no longer describes your site.
“AI” on crude manual data
“We applied AI to our shift reports.”
AI trained on self-reported data inherits the recall bias of the person who wrote the report. You get confident answers to biased inputs.
A model cannot recover work that was never captured. It can only rearrange what was already wrong.
Garbage in, garbage out — at scale, with a confidence score.
Why off-the-shelf vision fails
Complex environments break off-the-shelf detection.
Even genuine computer vision fails when it is trained on clean data and deployed in a mine. Stratos was built for this environment from the ground up.
Low lighting
Underground headings, night shifts, and tunnel shadows defeat cameras calibrated for clean conditions. Stratos models are trained on real mine footage, not laboratory datasets.
Dust, fog, and poor visibility
Blasting residue, diesel particulate, and water mist obscure the frame. Stratos handles visibility degradation without losing track of the work.
Complex workflow sequencing
A real site does not follow a clean sequence. Activities overlap, crews swap tasks, and equipment moves between headings. Standard detection breaks. Stratos absorbs the irregularities.