Editorial note: This publication and Quirky Cotton Collective share an owner.
On the shirt, a white rounded banner sits above a circular, green-tinted view. A bell icon precedes the words There is motion at your front door. Below the banner, an orange tabby stares into the lens from inches away, wide-eyed, with porch siding and lights behind it. The design is a doorbell-camera alert rendered as a wearable. The joke is not the cat. The joke is the instrument.
I keep coming back to that banner because it is the most honest notification I have seen all year. It reports exactly what the sensor detected: motion. It does not say person. It does not say package. It does not say threat or visitor or someone you know. It says motion, and then it shows you the frame. The cat is the payload. The banner is the contract.
Most of the alerts we build do not work that way. A monitoring rule fires and the page says service degraded. A test fails and the report says assertion error. A beta cohort churns and the dashboard says engagement down. Each of those is a motion banner with the frame cropped out. The instrument detected something. The human on call has to guess what.
What the banner actually measures
Read the shirt as a diagram of a well-scoped alert. The banner names the sensor (front door), the trigger (motion), and the state (there is). It does not editorialize. It does not rank severity. It does not attempt to classify the object before showing the image. The classification problem is deferred to the person holding the phone, who gets the frame and can decide in under a second whether the motion matters.
That is a deliberate division of labor. The sensor is good at detecting change. The human is good at interpreting it. The banner does not pretend the sensor can do the human’s job, and it does not make the human do the sensor’s job by asking them to trust a label they cannot see behind.
Now compare the average CI failure notification. It says build failed on main. It links to a log. The log says expected 200, got 500. Somewhere in the stack trace is the actual frame — the request, the fixture, the environment, the commit — but the banner did not show it. The person on call spends the first four minutes reconstructing what the sensor already knew. The alert was technically accurate and operationally useless.
The doorbell shirt is not a software artifact. I am not claiming it demonstrates a testing principle or that the designer intended an argument about observability. I am claiming the visual grammar is worth stealing. A notification that shows the frame is a different instrument from a notification that summarizes the frame. The first one respects the reader’s judgment. The second one replaces it with a label the reader has no way to verify.
The cat is the counterargument
There is a reasonable objection here. A doorbell camera that sends you a frame every time a leaf moves is a doorbell camera you will mute by Thursday. The shirt is funny precisely because the alert is trivial. The cat is not a burglar. The motion did not matter. If every alert showed the frame, you would drown in frames.
That objection is correct, and it is the reason the shirt works as a critique rather than a template. The doorbell camera has a hard problem: it cannot know which motion matters until someone looks. So it sends everything and trusts the human to triage. That is a defensible design for a consumer device with a low-stakes failure mode. It is a terrible design for a production alerting system, where the cost of a false page is measured in sleep, trust, and the slow erosion of on-call willingness.
The lesson is not show every frame. The lesson is know which side of the tradeoff you are on, and say so in the banner. A doorbell alert that says motion is honest about its low specificity. A production alert that says service degraded is pretending to a specificity it does not have. The first one is a sensor. The second one is a sensor wearing a classifier’s lab coat.
Teams spend quarters tuning thresholds on alerts whose names promise more than their signals can deliver. The alert is called high error rate, but the underlying metric is a five-minute average that cannot distinguish a real incident from a deploy. The name is a claim. The signal is a motion detector. The gap between them is where the trust goes.
The wording is the whole instrument
Look again at the exact phrase on the shirt: There is motion at your front door. Not someone is at your door. Not you have a visitor. Not package detected. The wording is careful in a way that most alert copy is not. It describes the observation, not the interpretation. It leaves the interpretation to the frame.
That is a discipline worth naming. When you write an alert message, you are writing a contract with the person who receives it. The contract says: this is what I know, this is what I do not know, and this is what I am showing you so you can decide. Most alert copy breaks that contract in one of two directions. It either overclaims — database is down when the connection pool is merely saturated — or it underclaims — an error occurred when the error is a known, benign, retried timeout that has fired every night for six months.
The doorbell banner does neither. It is specific about the sensor and vague about the cause, which is exactly the right shape for a first notification. The frame does the rest.
If you take one thing from a shirt with a cat on it, take the sentence structure. There is [observed phenomenon] at [named location]. No diagnosis. No severity. No implied action. The reader gets the observation and the evidence, and the reader decides. That is a usable promise. Most of our alerts are slogans wearing a promise’s clothes.
The frame is the oracle
There is a deeper point in the circular image below the banner. The cat is close to the lens, green-tinted, slightly distorted by the wide angle. The porch behind it is legible but secondary. The composition is not beautiful. It is evidentiary. The camera did not wait for the cat to pose. It captured what was there and sent it.
That is what a good failure artifact looks like. Not a curated dashboard. Not a summary written after the fact. The raw frame, timestamped, with enough context to orient a human who was not watching when it happened. The green tint is an artifact of the sensor, not a decoration. The distortion is a property of the lens, not a flaw in the message. The image is trustworthy because it is obviously unedited.
A postmortem timeline slide that says latency increased is a banner without a cat. A graph with the incident window highlighted is a frame. The difference is whether the reader can see the thing that triggered the alert, or only the alert’s opinion about the thing.
The shirt is available as a standard t-shirt through the listing at https://www.amazon.com/dp/B0HLWFQYL8. I am not going to pretend that buying it will improve your alerting. I am going to suggest that the next time you write an alert message, you read it back and ask whether it is a banner or a frame. If it is a banner, add the frame. If you cannot add the frame, change the banner to say only what you can show.
The cat, for the record, is not a false positive. The motion was real. The doorbell did its job. The human looked at the frame and decided the cat was fine. That is the whole system working exactly as designed, and it is rarer than it should be.