The plan says one gigabit. The first test, run on a Tuesday afternoon with the laptop next to the router, reads nine hundred and forty. The second test, run at eight in the evening with two televisions streaming, reads five hundred and ten. Nothing has failed, and nothing has changed in the contract.
The difference between what a line can reach and what it delivers when everyone else is also online is the central fact of broadband performance. Understanding where the gap comes from tells you which part of it is fixable and which part is simply the network.
What is actually being advertised?
A capability figure: the maximum throughput the technology and the plan allow under favourable conditions. It is not a promise about any particular evening, and in most contracts it is qualified by words such as up to or typical, whose definitions live in a code of practice rather than in the terms.
That qualification is not dishonest, but it is doing a lot of work. A plan advertised as up to one gigabit can lawfully deliver six hundred in the evening and still meet its obligations, provided the average and the measurement method align.
Where does the loss come from at peak time?
Three places. The access line, which is usually dedicated and rarely the problem on fiber; the shared segment upstream, which is where contention lives; and the wider network, including the transit links between the provider and the internet. Congestion at any of those points produces the same symptom.
Fiber reduces the first to near zero, which is why its peak-time behaviour is better than copper’s. It does not eliminate the second and third, which is why a gigabit line can still slow down when a street is full of streamers.
How much contention is normal?
Residential products are sold on a contention ratio that is not usually published: a shared segment serving a number of households, with the assumption that not all of them will be busy at once. A ratio of twenty to one is common, and fifty to one appears in cheaper products.
Where the ratio is high and the segment is full, evening speeds fall. That is not a fault, it is the product working as designed, and it is the reason business lines cost more for the same headline speed.
Is my own Wi-Fi the cause?
Frequently, and it is the first thing to eliminate. A laptop tested over Wi-Fi at the far end of a house can read a third of the wired speed, and the router will be blamed for a problem that is entirely about distance and walls.
Run the test with an ethernet cable before drawing any judgement. If the wired figure is close to the plan and the wireless figure is not, the line is fine and the conversation is about coverage rather than speed.
What is a typical speed, in regulatory terms?
Several regulators require providers to publish a typical speed range and to give a remedy if the service consistently falls below it. The definition usually involves measurement over a defined period, at peak times, on a wired connection, which is much stricter than a single test at noon.
Knowing the regulated definition matters when complaining. A claim has to be framed in the terms the code uses, and a screenshot taken at three in the afternoon does not meet them.
How should a speed test be run?
Wired, at peak time, more than once, and with nothing else using the connection. Then repeat at a quiet hour for comparison. The two figures together tell you whether the shortfall is contention or a fault.
Save the results with timestamps. A provider can act on a pattern, and cannot act on a single number that might have been measured over Wi-Fi during a download.
Does the plan’s speed tier matter more than the technology?
Both matter, but in different ways. The tier sets the ceiling, while the technology and the network set how close to the ceiling you get at the worst hour. A lower tier on a well-dimensioned fiber network can deliver a better evening experience than a higher tier on a congested segment.
This is why the cheapest gigabit is not automatically the best product. The number on the plan is a maximum, and the network behind it determines the minimum.
Why does speed drop during a video call?
Because a call consumes upstream bandwidth continuously and prioritises latency over volume. A household with a two megabit upload can find that a single high-definition call saturates the line, which makes everything else feel slow even though the download figure is untouched.
This is an upload problem wearing the costume of a download problem. Checking the upload figure and the number of simultaneous calls explains most of these cases.
Can a provider prioritise some traffic?
Yes, within the limits of net neutrality rules where they apply. Real-time traffic is often given priority over bulk downloads, which is intended to protect calls and generally does. Where management is applied, it must be disclosed, and the disclosure is usually in the acceptable use policy.
A household that notices video calls degrading while a backup runs is seeing prioritisation work in the wrong direction, which usually means the upstream is saturated rather than managed.
What about the router’s own limits?
Every router has a maximum throughput, and rental units are frequently specified just above the plan they are sold with. A router rated for a gigabit can struggle to deliver a gigabit of routing with firewall rules and Wi-Fi enabled, particularly in a warm cupboard.
Wired testing removes Wi-Fi from the equation but not routing. Where a wired test on a fast plan reads consistently low, a different router is a legitimate diagnostic rather than an enthusiast’s indulgence.
Does the time of day really matter that much?
It does, and the pattern is stable: a peak between eight and ten in the evening, a trough overnight, and a plateau during working hours that has risen as remote work spread. The peak is when the network is most expensive to dimension, which is why it is where the compromise sits.
Measuring at both ends of that curve gives a household an honest picture. A single test at either end gives a misleading one.
Can the gap be fixed?
Sometimes, by the provider, when the shortfall is caused by under-provisioned backhaul or a faulty component. Rarely, when it is caused by contention that the product was designed around. Never, when it is caused by Wi-Fi in a house with thick walls.
The value of measurement is that it separates those three cases. Without it, every complaint starts in the same place and ends in the same script.
What is the practical standard for a household?
Enough upload for simultaneous calls, enough download for the number of streams in use, and latency low enough that a call does not stutter. Those three requirements can be met by a modest plan on a good network and missed by an expensive plan on a poor one.
Measuring against needs rather than against the plan is the shift that makes the whole subject less stressful. The number in the advert is a ceiling, and households live below it.
When is a complaint justified?
When the wired peak-time figure is consistently below the guaranteed or typical range, measured over days, with records. At that point the shortfall is a service failure rather than a design compromise, and the remedies in the code apply.
That standard is higher than most complaints meet and lower than most households assume. Meeting it is a matter of method, not of persistence.
Why do two tests on the same line give different numbers?
Because a speed test measures a path rather than a line, and the path includes the testing server. A server that is itself congested will report a low figure for a healthy connection, and a well-connected server will report a high one. Choosing a server by distance rather than by reputation is the usual fix.
Running two tests against different servers and taking the higher figure is closer to the truth than averaging them. The lower figure usually describes the test infrastructure rather than the household.
Does the plan affect gaming latency?
Less than the route does. Latency is a function of distance to the game server and the number of networks in between, which is why a modest fiber line can beat a fast cable line to a specific server. Jitter, the variation in latency, matters more than the average.
A household with a stable forty millisecond connection and low jitter has a better gaming experience than one with a thirty millisecond average that swings to eighty. Measuring the swing is more useful than chasing the average.
The advertised figure is a promise about capability and the evening figure is a description of reality, and both are true at the same time. If you measured your own line tonight at eight, which of the two do you think you would see?