Projected Race Pace

For flat road races race prediction is at least in the ball-park, yours was only 1.8% off this time. You should always consider predictions with some kind of a +/- band (that Garmin should show!), so if things go "better than normal" (e.g. you run a carbon shoe that helps you that you didn't wear so much in training), then I can see you running faster than the prediction without making the prediction wrong.

Only because you ran the 5k once in 23:39 might still mean that your average expected 5km race time (according to your previous training and form) is 24:04.

If you want to witness some really ridiculous race pace projections, you should try and enter a race with significant elevation in it, like a mountainous trail race. If you do that the estimation will predict you to beat the course record every time, it is completely ludicrous how fast the algorithm thinks humans can run up hill.

See for example:

https://forums.garmin.com/outdoor-recreation/outdoor-recreation/f/fenix-7-series/315212/do-suggested-workouts-ignore-elevation/1528798#1528798

https://forums.garmin.com/beta-program/fenix-7-series/f/community-discussion/304571/race-time-prediction-very-optimistic/1475661#1475661

Elevation gain was 25m in that latest 5K of mine, pretty standard and definitely not downhill. I’m not saying I’m not happy with the general prediction, as it wasn’t that much off, what I’m saying is, once you run it in a better that predicted time and let’s say elevation gain > 10 or 20m then your predicted pace should update to at least what you’ve just achieved. 

Yes, I can definitively understand your point, but I can also understand why doing so might not be the best idea: Only because you do something once does not mean this is the normal/expected result. After all, a prediction should predict what it expects you to run for a 5k, not what you once were able to do on a magical day.

If my garmin had a shooting predictor which would tell me how close it predicts me to shoot to a bulls-eye, it should also not predict me to hit the middle just because I managed to hit there once in the past.

Dude, I have no idea why you’re defending it and using expressions like “was able to do this on a magical day”. It wasn’t a magical day, I just signed up, ran it, if anything I could probably have ran it even faster than what I did. If I run it next week it’s definitely not going to be slower. Garmin has build in an algorithm which uses available metrics to predict the pace, then real life happens and the watch should be smart enough to be able to adjust it predicted pace when you ran faster than its prediction (plus making sure it wasn’t downhill run).  Stop with this defending nonsense.

It's ok for everyone to have different opinions and preferences, but personally since the race prediction is (just like you said) an estimate based on physiological measurements, I personally wouldn't want it to change based on a race I ran, because the underlying physiological measurements wouldn't have changed. (And as was already mentioned, race time can also be affected by external things like elevation, wind, etc.) The watch already has personal records to store accomplished races, anyway.

And if the watch changed the race prediction based on past race times, what should it do if you are sick for a longer period of time? Show the past race time or discard it and start again giving predictions based on measurements? When should it do this change (any point would be quite arbitrary)?

If your race prediction is heavily off, then it could be useful to check if the underlying physiological parameters (like HR max etc.) are correct. And even if they are, it's still a statistical certainty that the prediction doesn't work that well for a certain (smallish) percentage of people.

I forgot to mention that based on Garmin's own pages, the data used for race prediction already does contain your training history (including the race you ran). But it's just one data source with limited weight on the prediction (as it should be). If you continue to run faster than the prediction, it will (slowly) adapt.

www.garmin.com/.../

Your suggestion is to update the predicted race time to the new PR time when you finish a race faster. I am just trying to argue why there is good reason for a prediction algorithm to *not* do that. I am not saying that the watch is right about you, it is very possible the watch is underestimating you currently, and that your 23:39 was totally repeatable for you, not a "magical day".

My point is, that the watch doesn't know this, there are all kinds of external and daily effects that will change a persons race performance - right now the watch is thinking you had a "above average" day. If you continue performing at that level the algorithm should adjust - but simply one data-point is not sufficient data to assume this is the new normal, which is why this is not something I would recommend implementing as suggested.

If you haven't run for a while, the watch reverts to estimating your VO2max from your walking activities, or (if you don't have any recent walking activities either), from your all-day HR. That affects your race prediction.