To give this little scenario some numbers to play with, lets say;
The train observer can see the clocks and is going to do all of his calculations merely from what he sees with no other assumptions…
As per ISRT, he ignores the time that it takes for the light to get to him. What he sees is Reality.
The clocks are set to be identical in the station’s frame of reference.
The train is going 0.5 times the speed of light.
The clocks are 6 light-micro-seconds (lus) apart (after length dilation concerns)
When the back clock, B, reads 10us (by the train observer), the train is 6 lus from that clock.
Since the clocks are separated by 6 lus, the front clock, F, must read 4us at that same moment, thus we have;
[4F, 10B] as our initial sighting from the train.
And at that moment is when the button gets pushed;
[4F, 10B] Button pushed
He can see that due to the motion of the clocks toward him it will take 2us for the light to get to the F clock;
[6F, 12B] is when the front clock should trigger and flash
And he can see that due to the motion of the clocks, it will take 4us for the light to get to the B clock;
[8F, 14B] is when the back clock should trigger and flash
Then he waits for the images. 5us later;
[6F, 12B] He sees the image of 6 from F
[8F, 14B] He sees the image of 14 from B
[6F – 14B] with 2us between their sightings
=====================================
But also assuming ISRT;
When the train is 6 lms away and the button is pushed, the station will read;
[16F - 16B]
6 us later the clocks must stop (reversing length dilation);
[22F - 22B]
=====================================
That is what we would expect from Relativity of Simultaneity
But the clocks can only stop at one setting, so did they stop at;
[22F - 22B] or at;
[6F - 14B]
Or perhaps at a different time than either?
THAT is the PARADOX and using ISRT from the get-go.
…and yes, your ISRT DOES assert a violation of the Equivalency Equation.