98.64 % uptime? It felt like I was blocked 44 % of the time.
There’s a SaaS metric that shows up in every status report, every marketing slide, every SLA table: uptime. The vendor in question proudly puts 98.64 % on the wall. Sounds like a system that’s basically always running. Sounds like one day of downtime in two and a half months. Sounds like: everything under control.
My gut said otherwise. So I counted.
The stripes
The vendor visualises status as bars over time — one stripe per day, coloured by state. Green means: all good. Red and orange mean: something happened.
I ran the image through a pixel classifier and counted the bars. The result for claude.ai over 90 days:
- Green: 50 bars → 55.6 %
- Red: 32 bars → 35.6 %
- Orange: 8 bars → 8.9 %
Put differently: on 40 out of 90 days, the stripe wasn’t green. That’s 44.4 %.
Where the gap between 98.64 % and 55.6 % comes from
Both numbers can be true at the same time — and that’s exactly the problem.
Uptime is usually measured as a ratio of “service technically reachable” to “total elapsed time”. If a component tips over for 20 minutes, you lose a couple of hundredths of a percent. The day stays mathematically almost entirely green. On the status board, though, it gets flagged as an incident — so the stripe isn’t green.
That creates a skewed logic:
- Vendor view: “We were only down for 20 minutes, that’s 0.02 % of the month.”
- User view: “On that day my build failed, my deploy slipped, my meeting got pushed. The day wasn’t fine.”
The 98.64 % measures seconds. My gut measures workdays. And workdays aren’t a linear unit — a 15-minute outage at the wrong moment costs half a day of flow, not 15 minutes.
What the more honest number would be
If you measure uptime the way it actually affects usability, the statement here isn’t “98.64 % uptime”, it’s:
The number the user actually feels
On 44.4 % of days there was an incident that disrupted working with the system.
This isn’t a math trick aimed at the vendor. It’s just a different reference frame. And it’s the one that matters to the user.
What I take away from it
- Uptime percentages are a PR format, not a user format. Technically correct, simultaneously misleading if you read them 1:1 as reliability.
- Status boards with daily stripes are more honest than SLA numbers — because they show exactly what you feel as a user: was the day disrupted or not?
- When the gut diverges from the metric, the gut usually isn’t wrong. It’s just measuring something other than what the vendor measures — and often the more relevant thing.
Maybe alongside “uptime” a second number should become standard: incident-day rate. Share of days with at least one reported incident. Here that’s 44 %. And suddenly that matches reality quite well.
Entdecke mehr
Working with AI: one word, two actions — and an hour of debugging
One word, two meanings: I blocked a Git commit, the AI heard deploy. Why a shared vocabulary decides everything when you work with AI.
GlossarSpeech-to-Text (STT)
Speech-to-Text refers to the automatic conversion of spoken language into text by an AI model. In an AI workflow, STT replaces the keyboard as the input channel — what matters is model size and domain vocabulary.
LexikonComparing AI models — who builds what and how to choose
The major model families in 2026 at a glance. Who builds Claude, GPT, Gemini, Llama, Mistral, DeepSeek, Qwen — and which model to pick when.