What factors are used to determine the Training Effect Primary Benefit of a cycling activity?

Your power and HR zones are a factor.  Check that they haven't changed somehow.  If they have change them back to what they should be.

Is your total activity time and intensity similar to before?

AFAIK the rating is based on the the total, meaning if you have a high intensity workout, then spin around for a while on the same activity it will dilute your rating. I could be wrong about this. I currently have a high aerobic shortage, partly because i do group chain gang workouts but also spin for 20 mins before and after. A few months ago, during lockdowns, i was doing rollers workouts that had short warm ups and warm downs. I then had a low aerobic shortage. This leads me to believe the load is attributed to categories as a whole activity rather than time in each power zone.

MB

TL;DR nobody knows, it's all proprietary secret sauce.

Those metrics are all generated by proprietary FirstBeat algorithms. They were originally licensed to Garmin for use, but then Garmin bought FirstBeat so now they own them outright. To my knowledge they are not open source, and they've never been disclosed for any public or peer review by the sports medicine community. Whenever I contacted FirstBeat for specific info (before Garmin bought them), all I ever got was hand-waving nonsensical answers. It left me feeling that even they didn't know how it works.

All that said...  All of the FirstBeat-based metrics in Garmin Connect are based on heartrate data (supplemented by other data such as power), and in particular heartrate variability (HRV). The claim by FirstBeat is that they monitor the variation in time between heartbeats and somehow correlate it to aspects of intensity, fatigue, freshness, etc., to generate those metrics (specifics unexplained). Note that real-time HRV data is actually recorded in the FIT file of the Garmin device, so technically this data is available if someone wanted to do their own analysis using some other algorithms (from where, I don't know).

The "final" metrics from an activity such as load, training effect, etc., are computed on the device itself and then synced to Garmin Connect. Unfortunately with this implementation, if there is a problem with the metrics that were produced, there is no way for us to correct it after (or Garmin, I've asked).  For example if you have bad power data on a ride showing an incredibly high power for a bit, it's enough to re-label your ride as some very high intensity / load, and then that false metric gets synced in with your existing data. Because the training status / load focus uses something like 4 weeks of previous data for trends, it can really mess you up for a while. It basically pollutes your trends with no way to undo it.

Among several things, this is a key reason why it's very difficult to lend any credibility to these metrics. I don't know if Garmin's acquisition of FirstBeat will lead to any improvements in this area or not. Health and fitness metrics are getting massively competitive right now and every company is looking for an edge. But that doesn't mean what they produce is necessarily legitimate and not just junk pseudo-science or buggy garbage.

In my experience, those metrics have been useful, at least in general, even though there are lots of issues with them. They generally support what I'm already feeling, so it's a good backup. Since over time your fitness will improve, your training effect for the same output will naturally drop, like you've seen.  What was tempo a couple months ago might be only a base ride now. In order to get into those upper zones you need to be at higher output than in the beginning. There comes a point however where we plateau, and without more training time (and/or a very different program), we're not really going to improve. Unfortunately the device metrics start flailing there and you'll get "Unproductive" regularly.

Both the intensity and duration are factored into the training effect. A medium intensity can end up being any of recovery, base, or tempo depending on how long you do it.  Heartrate ranges also seem to be important. If you do a series sprints but the dwell between them is long enough that your heartrate drops low between them, then it may not get high enough during the sprints for the device to classify it as anaerobic. The higher your fitness level the more rapidly your heartrate will recover to lower levels, so it becomes more challenging to get the device to register the session correctly..., which may be accurate, but we just don't know, because none of the algorithms are disclosed or peer reviewed.

The autodetect features can also really mess up your metrics. There is a known bug that auto-detect for FTP / lactic threshold will pollute your metrics and Garmin can't undo it. If you use HR zones based on anything other than the default %HRmax (and you should really), and then the device detects a new FTP/LTHR and you accept it, it will reset all your HR zones (which it shouldn't), and it will classify that activity according to the rewritten values, which will be junk. These then get synced into the 4-wk metrics so now they're polluted for a while before they get flushed out. So out of caution I would never accept an autodetect value. Just note what it detected, then deny it, and if you want to use the new value, enter it manually afterward to ensure your zones don't get screwed up.

As commented earlier, there can also be a dilution effect if a high-intensity session also has lots of low intensity at start or end.  Many of my tempo and threshold rides end up in base because I take a long recovery route through the city to get back home from the hills. You could split the session on the device by stopping the recording after the intense part, then start a new recording for the recovery / cooldown part. It would give you more representative metrics for each segment. But note that some of the trend metrics might start complaining because you have multiple workouts in a row with no "rest" in between. There doesn't seem to be a right answer because, to be honest, it's a very fragmented set of algorithms they've set up, with no coherent explanation of how they work.

After about 3 to 4 years of studying these metrics closely, I'd say I don't have much better understanding of them, or confidence in them, than I had when I started.  I feel Garmin now has all the ingredients they need to make it work, they just need to get on it.

Thank you Kyle for that detailed reply. Some might say in this inter webs era that it was too wordy and should be limited to an elevator speech, but I like it. Of course, I’m an engineer type nerd, so what do you expect.

regarding the secret formula, much trial and error over the last few weeks has shown that if I ride for 1 hour or more and spend ~75% of the ride in the range of 77-83% of my Max HR and keep my power primarily below 90% of FTP (Zones 1-3), I will get a Tempo type ride with a lot of High Aerobic benefit. If I spend more time above 83% Max HR or 90% FTP, it tends to benefit Anaerobic.

I would love it if Garmin would develop some type of data field(s) that could give me a real time evaluation of what I’m doing. It could be a simple text field that says “Base”, “Tempo”, “Threshold”, etc. Even better would be some type of graph that shows me where my HR and Power fall so perhaps I could see that my HR is pushing Tempo, but my power is 15 watts too low. Or maybe it says my power is 10 watts above the tempo threshold, but my HR is 5 BPM too low.