Well, as an engineer, I have been trying to resolve this problem. I have had a few surprises along the way.
As background, I am in my 70's with one full replacement knee. Doc says no run, but I can walk as much as I like. Thus my exercise is fast walking.
My Fitbit Versa 3 suggests that my aerobic fitness is over 20% above excellent for my age, so I hope that is correct.
My previous trackers have been TomTom Spark Cardios. Their distance tracking has consistently proven very accurate on the University measured tracks that they use for their elite athletes. These tracks are primarily coastal paths that meander along the ocean.
To date, my Versa 3 has been as much as 4% over reading distance to over 10% under reading.
I have tried a full reset without getting good results.
Initially I noticed that the Versa 3 indicated the GPS connected much quicker than my Sparks used to. However, on checking I found that the watch was showing connected while the circle of uncertainty was about 100 yards diameter. If I wait until the circle is down to about 5 yards, then measurements usually started out reasonably consistent.
Like others have reported, I have continually experienced errors increasing as the exercise period has extended. This is especially so if I significantly increase speed after an initial warm up walk.
One thing I noticed is that the distance algorithm seems to try to straighten out curved paths. By selecting Hike instead of walk, this effect seemed less. On walk, my usual error was 2 to 6%, on hike the error was about halved in my limited experience.
Two days ago, I did a 10-mile walk, consisting of two identical 5-mile laps. I kept the walking speed consistent at a slower than normal speed of just over 4 mph. The versa 3 indicated an accurate distance reading.
I did the exact same walk today, with my old Spark and the Versa 3. After warming up, I accelerated. After about 1.5 miles, the Spark was spot on, and the Versa 3 error was no more than 20 yards. It is at this point that the track enters a significant ascent, climbing about 100 yards over 900 yards travelled. My normal exercise is to start "motoring" from this point. The error kept growing the further I went. At the conclusion, the Spark showed 10 miles travelled - right on. The versa 3 showed just under 8.9 miles.
Since the raw GPS data is close to the same for both laps, the algorithm shows the same track, but indicates a significantly shorter distance.
Comparing the GPS tracks from both watches, the Spark is slightly more accurate with the actual path, but the difference would only be 1, maybe 2, per cent at most. I wonder if the algorithm could be turned off and just raw data be used. I am sure that would be better.
At his point, the Versa 3 is effectively useless as an exercise tracker for my typical use. I have no confidence in the distance except on my "home" track where I can make corrections. To put it bluntly, I have tried my best, and the watch still comes up well short. The Fitbit Versa 3 qualifies as one of my worst buys in my 72 years on this planet.
PS: The Versa 3 also showed my average heart rate at 126 vs 116 from the Spark.
