An episode, not an alarm: why Camly tracks states with a start, an end and a duration

Everyone has seen an alarm system nobody listens to any more. It rings for everything, it rings all the time, and after two weeks it rings into the void. The problem is not the sensor’s sensitivity. The problem is that the alarm answers the wrong question. It says “something is happening” when the site needs to know “since when, until when, and how many times”.

A safety situation is a state, not an event

A blocked emergency exit is not a one-second event. It is a state that begins when the pallet is put down and ends when it is taken away. In between, three minutes or three hours may pass, and that duration is the difference between a harmless gesture and a real risk.

A detector that sends an alert at every positive check produces, for a state that lasts two hours, dozens of identical alerts. The HSE manager reads three of them, mutes the notifications, and misses the fourth situation, the real one, the next day. That is alert fatigue, and it is mechanical.

Camly does the opposite. Each situation is tracked as an episode: one alert when the state appears, one notification when it clears, and nothing in between.¹

What an episode contains

Field	Example
Task	“Emergency exit B is clear”
Camera and zone	Camera 4, zone drawn in front of door B
Start	10:20
End	10:33
Duration	14 minutes
Written description	“A wrapped pallet is placed in front of the door, the passage is reduced to under one metre”
Evidence image	The image of the check, people blurred by default
Confidence	The certainty level of the check
Site feedback	“Was this right?”: yes, no, needs detail

There is nothing original in the structure itself. What matters is that it is produced automatically, for every situation, on every camera, every day, without anyone having to take a photo and write down the time.

From duration and recurrence to the numbers in the report

Once you have episodes, the numbers the risk assessment document, the works council and the insurer ask for are a matter of simple addition.

• How many times exit B was blocked this month.
• How long in total, and the average duration.
• At what hours: if it is always between 6 and 7 a.m., the problem is in the organisation of receiving, not in anyone’s discipline.
• The time to clear, from the start of the episode to its end, which is the most direct measure of how effective the corrective action is.

Take the replay of a day as it appears in the product: an episode from 10:20 to 10:33 on the timeline, 14 minutes, 11 flagged checks in the day, 53% of working minutes covered.¹ Those three numbers tell the HSE manager what happened, and they also tell them what the system looked at and what it did not. The second point matters as much as the first.

The quality indicators, or how the site knows what it is watching

A prevention tool that does not measure its own limits ends up being taken on trust, then not trusted at all. Camly therefore shows, per task and per camera, five indicators.¹

• Coverage: the share of working minutes during which the zone was actually checked.
• Alert delay: the time between a state appearing and the alert.
• Alerts per day, to spot a task that is too talkative before it gets ignored.
• False alerts, computed from the site’s “Was this right?” answers.
• Episodes per camera, to see where the site has a recurring problem and where it does not.

The “Was this right?” feedback on each episode is not a gimmick. It is what lets you correct the wording of a task, tighten a zone, and present the works council with a measured false-alert rate rather than an asserted one.

What periodic checks catch, and what they miss

It has to be said plainly, because it is the limit of the method. Periodic checks catch what lasts: a pallet in front of an exit, a narrowed fire aisle, cardboard in the charging area, a wedged fire door. They miss what flashes past: a forklift brushing a pedestrian, a fall. Camly is not a real-time interception system, and we do not sell it as one.¹

This is not a hidden flaw. It is the choice that makes the rest possible: spaced-out checks are what allow people to be blurred, nothing to be kept continuously, and a system that watches places rather than people to be presented to the works council. The states that last are, moreover, exactly the ones an inspector records during a visit.

Where to start

Pick one situation that you know lasts every day without anyone being able to say for how long. Write it in one sentence. After thirty days you will have a number of episodes, an average duration and a time of day. That is usually enough for a corrective action, and for the line in the risk assessment document that goes with it.

Sources

1. Camly AI, “Camly for warehouses” brochure, pilot edition, October 2026, pages “Alerts and episodes” and “Indicators and reports” (product captures: day replay, quality page).