I ran a speed test: what does that number mean?

You ran the test, it came out as 87. Eighty-seven what, and compared with what? And if the contract says 200, do you have grounds to complain or not?

This article answers those two questions, and the second has a very precise Italian answer that hardly anybody tells in full, because most of what you read online cites a rule that has been repealed.

The factor of eight, and the file that never arrives

The unit first, because on its own it explains ninety per cent of the disappointment.

A byte is eight bits. Offers and tests talk in megabits per second. Your browser, while it downloads, talks in megabytes per second. Between the two figures there is a factor of eight, and the upshot is that a 100 Mbit/s line downloads, in the ideal case, around 12 megabytes per second.

Let’s do the sum that matters. A 1 GB file on a 100 Mbit/s line: a gigabyte is eight billion bits, a hundred megabits per second is a hundred million bits per second, so eighty seconds. Not ten. Downloading a gigabyte in ten seconds would take 800 Mbit/s.

And eighty seconds is the perfect case, which doesn’t exist: there are the protocol headers, the start-up times, the server at the other end with limits of its own. A few percentage points more, no more than that, but they have to be added on.

There is a second confusion too, subtler, about the prefixes: when Windows says a file is 1 GB, it often means 1,073,741,824 bytes, which is technically called a gibibyte. The international standard recommends binary prefixes precisely to remove the ambiguity, and distinguishes megabyte from mebibyte. Nobody does it, and the result is a couple of percentage points of discrepancy that nobody can explain.

Why two tests in a row give different numbers

It isn’t that one of the two is lying. They measure different things, and it is worth knowing which.

How many streams they use. Some tests open a single connection, others open many in parallel and tune them to saturate the line. That isn’t a detail: with a single stream, if there is packet loss along the path the result collapses, and that is exactly why that kind of measurement is useful for confirming a problem. With many streams, when one slows down the others make up for it, and the problem stays hidden. Measurement Lab, which runs one of the most used single-stream tests, explains it in exactly those terms.

Where the server sits. Measurement Lab puts its own servers at interconnection points, outside the providers’ networks, and observes that tests hosted inside or at the edge of a provider’s network tend to give higher values. That is worth knowing when using the AGCOM speed test too, which states that it measures towards a server positioned at the edge of your own provider’s network: it measures the stretch as far as there very well, and it does not measure what comes after.

How long it runs and how much it transfers. Cloudflare writes something that holds for every test: transfers below 10 MB cannot use all the available bandwidth. A test that is too short underestimates, and the faster the line the worse the problem gets.

What sits in between. Here there are two admissions I appreciate a great deal, both from the people who run the tests. Cloudflare states that its own test runs in the browser, so the measurements include the local bottlenecks, that is the browser itself and the Wi-Fi. Ookla states that the result reflects both the network and the device’s capabilities, and adds something even blunter: tests run from a mobile browser are excluded from its own statistics, because they are not considered reliable enough at high speeds.

Put another way: the number you read is not your line’s speed. It is the speed your device, on your home network, towards that server, at that moment, managed to achieve.

When that number counts against your provider

Here begins the part worth knowing, because it is made of concrete rights and of deadlines.

The reference rule in Italy is AGCOM delibera 156/23/CONS of 27 June 2023, a delibera being the regulator’s formal decision, which replaced the old 244/08/CSP. I write that because it is the reason a good part of what you find online is obsolete: it cites a 2008 delibera that has been repealed.

The contract has to state, for download and for upload, four speeds: minimum, normally available, maximum and advertised. And the provider, says the regulation, undertakes contractually with the user to respect those values, which also cover delay and packet loss. That is where the possibility of challenging it comes from. The same obligation comes from the European regulation on access to an open internet, which at article 4 requires those four speeds to be stated and, at paragraph 4, establishes that a significant discrepancy, continuous or regularly recurring, established through a monitoring mechanism certified by the national authority, counts as non-conformity and triggers the remedies provided for under national law.

The definitions are technical and worth giving in plain words, because they change everything:

• the minimum speed is the value exceeded by 95% of the measurements. Five measurements out of ninety-six below the contractual threshold are enough for it to count as breached;
• the normally available one is the value exceeded by 75% of the measurements, and it takes twenty-five measurements out of ninety-six below the threshold, that is more than six hours of the day;
• the maximum is the highest value of all the measurements, so it only counts as breached if no measurement in the day reaches the stated threshold.

You don’t have to guess the values of your own offer: every provider has to publish a technical transparency page reachable with a single button from its own home page. That is where the numbers that count live.

The procedure, with all the deadlines

The software is called Ne.Me.Sys, it is free, it is run by the Fondazione Ugo Bordoni on AGCOM’s behalf, and it is the only one that produces a certificate with evidential value. The online speed test, from the same project, openly states that it is a one-off measurement with no validity for complaint purposes, and the same thing is written in the regulation.

1. Get the numbers from your contract off your provider's technical transparency page: you need the minimum speeds in download and upload, and the maximum delay.
2. Take the measurement over the cable, Wi-Fi off, leaving the computer switched on for twenty-four hours and using it as little as possible. Without those conditions you are measuring your house, not the line.
3. If the certificate reports a breach, send it within 30 days of the date it carries. Sending it counts as a formal, substantiated complaint: it is an act, not a grumble.
4. The provider has 30 days from receipt to bring the values back within the limits.
5. If that doesn't happen, a second measurement is needed, no sooner than thirty days after the first. Only with the second certificate do you get as far as withdrawal.
6. At that point you can withdraw at no cost, by recorded delivery or PEC, or move free of charge to a cheaper offer on the same technology, if one is on the price list. The withdrawal concerns the internet access service, and it takes effect thirty days after the communication is received.

A clarification about which breach entitles you to what, because they are not all the same. If it is the minimum bandwidth in download or upload that is below the threshold, or the maximum delay or packet loss above it, the certificate counts for a complaint, for withdrawal and for compensation. If what has been breached is the normally available or the maximum bandwidth, it counts for compensation purposes only.

And if the thing doesn’t get resolved, before going to a judge there is a compulsory step: the attempt at conciliation before the Co.re.com, the regional communications committee, done online on the ConciliaWeb platform by logging in with SPID, Italy’s digital identity scheme. It is compulsory in the technical sense of the word, that is, without it a court claim cannot proceed, and the condition is treated as satisfied once thirty days have passed from the filing of the application. On whether the procedure is free I have read the regulation and have not found a sentence declaring it so explicitly: everybody takes it for granted, but I prefer to say I haven’t verified it.

For mobile phones none of this exists. The mobile rules, updated in 2023, impose transparency on coverage and estimated speeds by technology, but they do not provide for an individual certified measurement with evidential value. The official page says so indirectly, in wording that leaves no doubt: if you are a fixed-line user, you can use Ne.Me.Sys. For mobile there are measurement campaigns across the country, whose results are aggregated and comparative, not yours.

What to do with it, in practice

The quick browser test and the certified measurement aren’t in competition: they are for two different moments.

The quick test is for working out whether there is a problem, for comparing one room with another, for seeing whether something has changed since yesterday. That is why I put How good is my connection? on this site, and it is honest to say that it is exactly one of those tests with no evidential value: it runs in the browser, so it measures the browser and the Wi-Fi too, it stops at twenty-four megabytes so as not to burn bandwidth pointlessly, and above 300 or 400 Mbit/s it underestimates. It is there to give you a starting number and tell you what it means, not to help you sue anybody.

The certified measurement is for when the starting number, repeated under the right conditions, stays well below what you signed for. At that point you move to the cable, the twenty-four hours and the procedure.

To understand what sits underneath

If the speed is good but something still isn’t right, I have covered the two most frequent causes separately: video calls breaking up are in The video call keeps breaking up, but the test says the line is fast, and the signal that doesn’t reach every room is in The Wi-Fi works in the living room but not in the bedroom. If instead the problem is general and you don’t know where to start, the checks in order are in Slow Wi-Fi: six checks to make before calling your provider.