These types of problems are not uncommon with LED PCBs. Without a schematic, photos, and LED footprint I can only offer generalized information.
Two high probability issues would be thermal runaway and more likely cracked ceramic substrate when the score lines were broke. A third issue may be the solder process. Some LED footprints are subject to solder stencil and assembly problems. Even the rotational position of the LED and the direction of how the solder paste is applied can cause issues.
Thermal runaway can be a slow process that takes many hours to manifest. Then when powered down and allowed to cool the problem goes away. When powered again the problem will repeat. This is generally associated with a constant current source and unbalanced parallel strings. You are using constant voltage and current limiting resistors so not likely. But if there are any parallel strings getting current from exactly the same path then it is a likely source of a problem.
Keep in mind you could have multiple problems.
Most likely is caused by different coefficients of thermal expansion on the layers of the PCB causing warpage. The warpage puts stress on the LED package and solder
joints.
Section 3.6 (FR4 Board Handling) of this app note explains the warpage and cracked ceramic packages and copper patterns that can help.
LUXEON Rebel Plaform Assembly and handling information
The following app note goes into detail regarding the solder paste and assembly process recommendations.
See Section 3 Electrical Layout Consideration and
Section 4. Assembly Process Guidelines
LUXEONC Color Line Assembly and Handling Information
Many do not believe thermal runaway is real.
This Texas Instruments app note provides some details.
See Section 3 Pitfalls of Parallel LED Arrays
Design Challenges of Switching LED Drivers
This OSRAM app note explains how to prevent thermal runaway. Many times the unbalanced strings will reach thermal equilibrium before one of the strings fails and the current flow is such that some strings are dimmer than others. Again this may take hours to progress.
Current distribution in parallel LED strings
sometimes when the boards are switched off and then back on, they
light up fine.
If the intermittent nature of the problem is not mechanical warpage, the way you describe it, it sounds just like how how I've seen unbalanced parallel strings symptoms manifest. I would need a schematic to verify.
I had some strips that were unbalanced the forward voltage were 43.29V (yellow) and 43.85V (blue). I powered them with a CC (HLG) and CV (HEP) with 5Ω resistor in series with each strip.

Unbalanced strings is a bigger problem than most designers realize. Most of the time it is undetectable. I have done a lot of experimentation with driving parallel stings and the unbalanced nature of the problem. Sometimes economics dictates running multiple strips with a single CC driver. What I do is add shunt resistors to the design so I can measure how unbalanced the string are. They are ALWAYS unbalanced and I want to know by how much. When I receive strips the first thing I do is measure the forward voltage of each strip and try to pair the parallel strings by Vf. No matter how close the Vf is matched, the current is still unbalanced.