I am trying to build a timelapsing program in python, which analyzes the picture and determins the exposure, based on which it corrects it's shutter speed using a polynomial regression.
But I got stuck when the exposure is somewhere around 1 second, where you can see in the graph below, the brightness jumps

The orange line is the expected curve, the blue dots are actual measurements.

Digging deeper, I found out, that the actual exposure times differs from the shutter speed I set.
With cam.exposure_speed I found out, that at cam.shutter_speed = 1000000 (-> 1s) the actual exposure time was 7997813 (-> 8s) where as at cam.shutter_speed = 990000 (->0.99s) the exposure time was 989976 (-> 0.989s).

Here is an example code to recreate the effect:

from picamera import PiCamera
import time

cam = PiCamera(sensor_mode=2)
cam.framerate = 1/8
cam.iso = 100
cam.shutter_speed = 1000000
cam.resolution = (1600, 900)
time.sleep(5)
cam.capture("test.jpg")
print(cam.exposure_speed)
cam.close()

So far I tested:

• Diffrent ISO values (needs to fixed ISO)
• diffrent resolutions
• removing the sensor_mode argument
• diffrent framerates

I am running

• Python 3.7.3
• Raspberry Pi 4 with 4GB RAM
• picamera 1.13

Any help on this will be helpful, I will try to edit missing information asap.