955 massively underestimating my ability

Check if your max HR, HR zones and LTHR are set right.

As mentioned, check zones. But generally estimation is a model and won't fit all. 

Some people have more general muscle training, some do less, Garmin is judging by HR (there's an online Fitbit whitepaper describing algorithm) and these two people will perform differently even though their training hr/pace look the same. I performed both better and worse then prediction (mainly that depends if it is beginning or end of season)

Personally if I do 1000m intervals regularly in the beginning of the season, Garmin will show me estimation of vo2max 56 (and predicted race time as well), when I don't do 1000m in my training and do longer and slower intervals (3000, 5000 ) my vo2 will go down to 52 while actually I will not perform worse. 

So I use these estimations as year to year comparison (which also works with some limits- training shall be more or less the same).

Is it summer where yiu live? If you run in the heat, your heart rate is higher than usual. Performance is predominantly calculated from the ratio between heart rate and pace.

To my knowledge, HR zones do not affect the analysis and predictions, they are for the user's eyes only. Correct Max HR is crucial, potentially also correct resting HR and lactate threshold HR.

The Max HR was set to autodetect, and had it at 185. I think this is a little low although likely not very far off. But when I changed it to 189, my predicted times went up—or something else in the last 12 hours, other than a workout, caused it to go up. Now my predicted 5k is again only about 10 seconds faster than my recent tempo run:

loltracking said:

The Max HR was set to autodetect, and had it at 185

Do not use the HR Max autodetect. Despite improvements in the qualification of peak HR data, the algorithm is still prone to massive errors in my experience. Also, it is set up to increase the HR Max when "qualified", so if the default value (formula-based) is already too high, things can only get worse.

If your doctor lets you, do a field HR Max test, or run your best and fastest 5k ever and add a couple of bpm to your HR at the end of the race.

The second important thing is to do your best effort to capture good data with the watch. Wear a chest strap, choose the best GPS setting for your area and avoid steeper hills. Connect the watch before a workout to update temperature and altitude at the start. Wear the watch during the night to get a good Rest HR. From my experience, I think that bad data from a single workout will have some effect for a few weeks (the watch is using predictive analytics, so the data set has to be a window of time of some sort to capture improvements).

Finally, run maximal effort (fastest but steady pace) as you keep your training balanced. A good way to do this is to follow the daily workout suggestions and to replace a few intervals by maximal effort from time to time. This will give the best range of data for the watch to estimate your VO2 Max, EPOC, stamina etc.

Etupes25 said:

Do not use the HR Max autodetect. Despite improvements in the qualification of peak HR data, the algorithm is still prone to massive errors in my experience. Also, it is set up to increase the HR Max when "qualified", so if the default value (formula-based) is already too high, things can only get worse.

I followed Etupes25's suggestion a few weeks ago and stop recording Max HR automatically. Training suggestion and vo2max are rising since then. I also had a feeling the autodetect function (also the one for LTHR) did not came close to my real figures. So thx Etupes25, and to loltracking I say it is good to try. Now I do a LTHR test every now and then, and put in the results manually in Connect. Now I actually get tired of my tempo runs :-) and pace goes up, HR goes down, VO2Max goes up, and fun is also rising again...

I set the HR and the LTHR manually, and it didn't change the results. Yesterday, I raced a half-marathon almost 5 minutes faster than my predicted time, and in fact my heart rate, as well as my intuition, tells me I could have gone faster; I was second-guessing myself and holding back partly because of this poor data from the 955. Meanwhile, the race result did not change my VO2 Max or my predicted times. 

So what's going on here?

If you change your HR Max, it will have an impact on your metrics over time as you provide new “constraints” to the HR/HRV/Pace/Time linear and non linear relationships that the watch is modeling out, after filtering, cleaning the data during each run. 

In addition Garmin has introduced some dampening in the day to day change of VO2 Max to avoid swings like in the past.

When you change manually the LTHR you just change an end-user value. The watch determined LTHR is still part of the historical data.

So you won’t see an immediate impact.

The best process is to make sure your HR Max is correct. As you know, this is not your peak HR ever. It is better to do a field test (if your doctor lets you) or a lab test.

Doing a lactate threshold test is excellent, using the auto-detect is not great. 

Just like personal bests and recent race results, these are just data points among others for the watch to use in modeling metrics and predictions.

I like the way Stryd present their predictions, with a range. For example, my 5k estimate is +/- 28s which is huge, but help set expectations for this type of modeling.

Your mileage will vary.