Hmmm. Try being a passenger on the ground experiencing this sort of thing on your regular service - admittedly into Paddington in the evening peak, but otherwise:
Ontimetrains 12 Weeks 1627 Bristol Parkway-Paddington
Cancelled 1 in 5 times (20%) in the last 3 months.
Cancelled or over 15 minutes late 17 times (29%) over the last 3 months.
On time 1 in 20 times (5%) over the last 3 months.
That data doesn't even reflect the insult being added to injury of it being a 5 coach semi-train more often than not when it does deign to show up.
Yes, I know, it's supposedly all the DfT's fault. Yes, the GW Mainline infrastructure omnishambles has calmed recently. Some regular users are still getting a shoddy experience at a premium price though. See above.
~~~~~~~~~~~~~~~~~~~~~~~
Picking instead data relating to trains I don't use but irish_rail (at a guess) drives Paddington-Plymouth:
Ontimetrains 12 Weeks 1703 Paddington-Plymouth
Looks like >50% arrived >15 mins late over 12 weeks. So they are not wrong there.
The stats for the entire franchise 24x7 in any case average away meaning by burying it in thousands of calls by local and regional trains at minor stations at 05:00.
If the average were weighted in favour of the trains, stations and times the most people actually use (i.e. largely Paddington and not at 05:00), the answer would likely be strikingly different. In any case, arithmetic mean is a misleading measure of location to use for highly skew delay stats to start with, but that's a rather technical further digression.
With apologies for contributing to the OT drift of the thread.
The same sort of thing is probably true for Avanti though!