Reach altitude strange variations

Hi all,
With Reach rover and base corrections i’ve recorded several points on a lake in XYZ format, then converted it in LLH.
Strange thing, points heigth variates between 0m and 20m, on the same flat water plan (no waves or water variations, no obstacles blocking satellite view, metal plane below the antenna) …
Latitudes and longitudes values are ok.

On this recorded sequence, logs are showing that the reach was seeing more than 13 satellites, and receiving base corrections. His state was “float”, so good positionning.

Does such heigths variations (errors) are normal?
What heigth precision can we expect from Reach?

Could you elaborate the conversion process?

I’ve used several coord conversion softwares (some online), they output the same data.
Example:
with this converter:

XYZ entry:
1° point: 4543758.5848, -80966.9835, 4460394.4469
2° point: 4543747.7424, -80984.6486, 4460376.7838

Lat, Lon, Heigth output:
1° point: 44.6574002504005, -1.02086751010486, 67.9200526159257 (m)
2° point: 44.6573537639489, -1.02109262837519, 48.0179458782077 (m)

About 20m gap on the same water level (and almost the same place)

@Guillaume33, well your conversion looks right. From the provided detail it is hard to tell what exactly went wrong. How did you collect the points?

My Reach rover was on a boat, the base was on water side, on a tripod.
The 1° point is the first of a 1 hour record sequence, the 2° is the last, so 1 hour between the two records.
Between the 2 points many points have been recorded, and the height is slightly moving between 67m (start) and 48m (end). The height not only decrease but variates, increase & decrease, but always slightly.

I’ve observed this fact on other lakes, same conditions, with other height variations.
(sometimes more than 100m variation, sometimes just 5m, but 5 meters is much more than the Reach precision capability i think)

Edit: 48m is the good height at the location i have done this records.