I had similar problems especially when starting from a stationary state or when beginning the interval from a very slow jog. If I start the interval already running at the right speed that I want to keep during the intervals, the numbers that appear are much better. I didn't try exactly what you did, but I have an idea of what you are experiencing. I just tell you that I don't use any more the "pace audio alerts" of Garmin because they don't work well.

I opened another "thread" with this title

Bug in the interval workout screen with "pace as target": for the first minute the watch displays in the interval screen a slower pace when starting from a stationary state

It seems that things are a bit complex and I don't want to discuss again these things. But I agree that something isn't going  well and Garmin should work a bit on this "structured workout" feature. Garmin should find what is wrong exactly and why. After all, if so many people complain about this there must be something wrong in how Garmin implemented the "workout" feature. I wonder how many professional  athletes use the "Garmin pace alert" function: my guess is zero.

EDIT: I didn't try the new firmware 2.80 yet and maybe the bug was corrected. I will try again next week and finger crossed.

Update of today 7 September 2019 with firmware 2.80:

I did this test workout : repetition of 1 km at pace 5:30 min/km and 1 km at 5:00 min/km with the FR945 not connected to footpod (only with GPS+Galileo, so both instant pace and lap/step pace from gps) and another watch FR610 connected to a Runscribe footpod with speed from footpod (this means instant pace from footpod) and distance from gps (so lap pace from gps)

1- my first impression is that Garmin fixed the bug of the thread above that in the first minute the watch gives wrong pace alerts: to be confirmed in my next intervals session of 400 m where I walk in the rest period

2- for repetition of 1 km the pace gauge of Garmin seems to work well now. Today I was always running, I will see  when I do other intervals of 1 km starting from a stationary state if the bug described above was fixed by Garmin

3- In the middle of the interval on the FR945 gave similar values for instant pace and lap/step pace, and this numbers were similar of the the numbers of the FR610

4- when starting the interval, i.e. when accelerating from 5 min/km to 5:30 min/km or the opposite (decelerating from 5:30 min/km to 5 min/km) I noticed the same thing that you noticed.

The instant pace of the footpod on the FR610 was the most fast to respond in change of rhythm.

Lap pace of the FR945 was also quite  fast, the slowest was the instant pace of gps+Galileo of the FR945.

Instant pace of the FR945 was often the same of the footpod  of the FR610, in particular from the middle of the interval to the end (but in some moments there was a difference of  20-30 seconds even in the middle of the interval, I don't know why, maybe sometimes gps reception wasn't good enough)

5-  snowfree52 in my opinion instant pace of gps with the FR945 is good only if you do intervals at least  3-4 minutes long. For shorter intervals  of 30 seconds -1 minute instant pace of the FR945 is too smoothed, and not suitable to register fast changes in speed.

One solution could be to start the interval already at the right speed, so there is no delay in instant pace and it should pace you correctly.

Another solution is to use instant pace of a good footpod (Runscribe, Stryd or even Garmin - it seems that Garmin will release a new footpod this year).

6- Even with this firmware it's confirmed that in the "workout screen" of "my workouts" or "intervals", with pace Garmin means "lap/step" pace and not "instant pace".I had in a second page the two data fields instant pace and lap pace, and lap pace was always the same value of the "pace" in the workout screen. It's also confirmed that the "pace alerts" depend on the step/lap pace.

Now lap pace for some reason is more quicker to react at the beginning of the interval than instant pace (when pace is coming from gps and not from a footpod). If Garmin doesn't allow to change the smoothing factor of instant pace, it's better to leave in the workout screen "lap pace" , because it's quicker. Obviously when using a footpod it would be nice if Garmin gives the choice to use in the workout screen instant pace and not step/lap pace, because instant pace of footpod is quicker to react than lap pace. Maybe I will open a new thread asking Garmin to give this new feature/option.

Anyway, Garmin always used the word "pace" for "instant pace". If in the interval/workout screen they decided for some reason that it's better to use step/lap pace instead than instant pace, it should be better if they show in the pace gauge the word "step pace" and not  "pace". 

I did today a interval sessions of 1 km, with pace as target, with the FR945 having speed and distance from GPS+Galileo.

It confirmed that the bug described in the thread above was fixed by Garmin: now in the workout screen the pace is always  quite accurate and there are no false pace alerts at the beginning of the interval. I don't know exactly what Garmin did to fix the issue, but I think that Garmin is doing something very clever and the pace shown in the intervals screen is probably something different from instant pace or lap pace from gps, it is probably a pace that is the better to describe how is progressing the interval and now is quite accurate even in the beginning. Maybe it uses lap pace at the beginning of the interval (that is quicker to react than instant pace) and instant pace in the end of the interval,  or it has a different algorithm.

I looked often at the "pace  gauge" and it was useful to know how the interval was going and to have and idea what time I would have at the end. Now the "pace gauge" never gave me completely wrong value.

In some intervals I switched page and I could see two fields, lap pace and instant pace. It confirmed the run of two days ago and what @snowfree52 reported: when I started the interval, lap pace was very fast to show the correct speed. I had in the other arm another watch, a FR610 connected to a Runscribe footpod that gave super responsive and accurate instant pace: the lap pace of the FR945 had a bit of a lag but was quite quick to show a pace near the pace of the footpod (obviously I was trying to keep a constant speed).

Instant pace of FR945 had a big lag, only after 400 m began to give correct numbers. Only in the second half of the interval the instant pace of the FR945 gave similar numbers of the lap pace of the FR945 and instant pace of the footpod of the FR610. Probably instant pace of the FR945 when connected to gps is very good in steady runs, but not suitable for short intervals.

snowfree52 If you don't want to use a footpod, I would suggest to use the pace shown in the interval screen of the "structured workout" , it is good. I don't know if it is sometimes lap pace, sometimes instant pace or sometimes a different value calculated by another algorithm of the watch, but the important thing is that it works good.

If you want super accurate and responsive instant pace then you need a footpod, I had instant pace of the footpod in the other watch and it was useful to pace myself.

I bought a pair of earbuds and I set instant pace and heart alerts in the FR945 and I enabled "Audio Alerts" in the settings of the watch: every minute the watched informed me of the values of pace and heart rate, so once I set my speed  I didn't need to look any more  at the watch, and so I avoided to lose my running form, it was very nice.

At the end, in the cool down I listened to a pod cast of DCRainmaker  and the music part of the watch functions very well.

It was really a beautiful session today and I was very happy with the watch, what I have to say, when everything works good, Garmin is great...

and the earbuds add yet more value to this watch

well thank you for your posts ! will try tomorrow and see if I get the same feeling 

I can confirm the bug with instant and lap running pace on my FR945 with firmware 2.80 during an interval training.