Published projections of where a team will finish are not a single opinion written down. They are the summary of an entire season played out repeatedly inside a computer.
The engine starts with match probabilities
Every remaining fixture is assigned a set of probabilities for home win, draw and away win, usually derived from team strength ratings that update as results arrive.
Strength ratings themselves come from underlying performance rather than results alone, because a team can win narrowly for weeks without being as strong as its points suggest.
Those per-match probabilities are the only real input. Everything downstream is bookkeeping applied to them many times over.
One simulated season is a set of coin flips
The software draws a random result for each fixture according to its probabilities, so a heavy favorite usually wins but sometimes does not.
It then applies the league's actual rules to those invented results, awarding points, sorting the table and breaking ties by whatever method that competition uses.
The output is one plausible final table. Taken alone it is worthless, because it reflects one particular sequence of coin flips.
The distribution is the actual product
Repeating the process many thousands of times produces a distribution rather than a table. A club might finish anywhere across a broad range, with some positions far more common than others.
From that distribution come the familiar figures: the chance of the title, of European qualification, of relegation. Each is simply the share of simulated seasons where it happened.
Reporting only the average finishing position throws away the interesting part. Two clubs can share an average and have completely different risks attached.
Where these projections go wrong
The simulation assumes each match is independent, which is not quite true. Injuries, a coaching change or a collapse in confidence carry from one fixture to the next.
It also assumes the rating is right. If a squad has been reshaped in the transfer window, the model is projecting a team that no longer exists.
Both errors tend to show up as projections that are too confident, particularly early in a season when ratings still lean heavily on the previous year.
Reading them sensibly
The right use of a projection is comparative rather than absolute. It answers which clubs are in real danger and which are effectively safe, not what the table will look like.
Watching the figures move week to week is more informative than any single snapshot, because the size of the move shows how much a result actually changed for the clubs involved.
A projection that barely moves after a surprising result is telling you the model already accounted for that possibility, which is usually a sign it was built sensibly rather than a sign it is ignoring evidence.