Your network graph is clearly fake

I have been monitoring the latency graph on your status page for six weeks. I am going to be direct: the data shown is physically impossible and I can prove it.

Your Contabo instance in "FRA1" claims 2.3ms to your "NYC1" node. I have tested this from three separate vantage points with known-good clocks. The great-circle distance between Frankfurt and New York is approximately 6,200km. At the speed of light in fiber, that is 31ms one way, 62ms round trip. Your graph shows 2.3ms. This is not a compression artifact or a caching layer. This is not possible.

Your "SIN1" to "LHR1" path claims 8ms. The fiber distance is ~15,400km. Minimum RTT: 154ms. I have collected 847 samples. The distribution does not match any real network path. It matches a smoothed spline fit to synthetic data.

I am not alleging malice. I am alleging that you are displaying a representative latency model rather than measured data, and your TOS does not disclose this. I have archived your current status page and your current TOS. I am asking you to clarify which it is: real measurements, or marketing visualization?

If this is an error in your probe placement or your graphing pipeline, I will happily help you correct it. I maintain IRR objects for multiple ASNs. I understand how these systems fail. But I need to know which failure mode I am looking at.

Transparency: we had a similar issue at my previous gig. The monitoring vendor was applying a "visual smoothing" algorithm that shaved 40% off peak latencies to "reduce alert fatigue." We did not catch it for three quarters. It violated our error budget philosophy—how can you improve what you cannot see?—and we terminated the contract after a blameless postmortem. I would be very interested in Contabo's response here. This is the kind of thing that erodes trust in status pages industry-wide.

I WAS LOOKING AT THIS LAST TUESDAY WHICH WAS THE 21ST I THINK NO THE 22ND BECAUSE I REMEMBER I HAD OATMEAL WITH BANANAS AND MY CAT WAS SITTING ON THE ROUTER WHICH IS A UBUNTU BOX I BUILT FROM SCRATCH IN 2019 AND THE WEATHER WAS RAINY AND I NOTICED THE SAME THING ON MY RACKNERD VPS WHICH IS IN AMSTERDAM SUPPOSEDLY BUT THE TRACEROUTE GOES THROUGH SOMETHING LABELED KNOWNHOST AND THEN BACK TO RACKNERD AND THE LATENCY IS ALWAYS 15MS NO MATTER WHAT TIME OF DAY AND I THOUGHT MAYBE ITS A CDN BUT THEN I CHECKED FROM MY FRIEND'S HOUSE IN ROTTERDAM AND IT WAS ALSO 15MS AND SHE HAS A DIFFERENT ISP AND A DIFFERENT CAT AND IT WAS SUNNY THERE SO I STARTED THINKING MAYBE THE NUMBERS ARE JUST MADE UP AND I AM SO GLAD YOU POSTED THIS BECAUSE I THOUGHT I WAS GOING CRAZY AND NOBODY READS MY POSTS ANYWAY BUT I HAVE SCREENSHOTS FROM THREE DIFFERENT DAYS AND THE GRAPH IS IDENTICAL DOWN TO THE PIXEL WHICH IS STATISTICALLY IMPOSSIBLE UNLESS THEY ARE REUSING ASSETS

Anyway I wrote a probe at 3am that samples their api every 30s and stores raw json. Idk why I do this. The timestamps have microsecond precision but the values are quantized to 0.1ms steps. Real networks have jitter. This has none. Its either smoothed to death or generated. Heres the thing though: the tos at the time I signed up said "actual performance data." not representative. Not estimated. Actual.

#!/bin/bash
while true; do
  curl -s https://status.contabo.com/api/latency \
    | jq '.fra1_nyc1' >> /tmp/fm_real.json
  sleep 30
done

Been running since january 15. Want the dataset?

Network fake. Perkele.

traceroute shows 12 hops. Graph shows 1. Not same thing. Fiber not magic. Speed of light not suggestion.