Strange failure mode of a LED "bar"

A long intro:
I have the "retro" LED bulbs. The LED part is made of 4 LED "bars" (I don't know what is the right name for this). Every bar is about 24 white LEDs in series. Those bulbs fail with similar mechanism - at first the start blinking, the frequency increases and finally the bulb fails permanently. I have noticed the bulb is not completely dark - it glows faintly when powered. After "disassembly" of the glass cover I have decided one of the LED bars is damaged somehow.
The LEDs are in series, driven by a constant current source which is powered from rectified mains voltage (about 320 V). I have measured voltage over the bars and it was about 60 V over three and 150 over the last one. The current was about 30 uA. Clearly the 150 V bar is damaged somehow. I have shorted it and the remaining bars started to shine at full brightness - about 20 mA and 85 V forward voltage drop each. To do more experiments I have shorted the working bars instead. Even with full mains voltage over the damaged bar the current was still 30 uA - it looks like the damaged bar current is independent on the supply voltage.
I have replaced the faulty bar with one from another bulb that failed some time ago. At first it worked OK but soon it started to flicker too.
Now it comes the interesting part:
The faulty bar is randomly switching from "healthy" to "damaged" state. It has the right 85 V forward drop shining fully and within a few microseconds the voltage goes up to the 150 V. It stays here for some arbitrary time (from milliseconds to seconds) and then drops very quickly to 85 V (very sharp edge, less than 1 us). Can someone explain this?

"Filaments".

Thermal expansion effects within the structure, faulty connections.

It is hard to imagine how the thermal effect on faulty connection does such effect. The time between transitions seems unpredictable and changes wildly. Such as steady 5 seconds of normal light, 2 seconds of quick flickering, another short time of steady full light and then few seconds of "low current".
The original filament was in the "low brightness" mode continuously except sometimes after power was disconnected it briefly flashed with full intensity.
Also why the faulty connection keeps the low current path?
Why the failure mode of the filaments is so similar?

But I must admit I cannot find any better explanation.

Guess: Look at ballast and/or constant current circuit.