1.In a modern, ideally programmed system, filters should work dynamically in real time - however, due to system limitations:

For a card with saved settings:

• a) Let's assume that we search for the target element by the X Y Z parameters

• b) We want to see the results so click apply.(the card closes
  automatically and we see the result)

• c) We see 30 results, but we decide to change the Y parameter to narrow down the results

In this situation, after opening the window with filters, I think that the parameters X Y Z should be remembered from the previous operation (stored e.g. in localstorage) - however, this requires analyzing several scenarios and discussing the most appropriate solution.

Perhaps the filters should be redesigned? (as a sliding div, not a collision modal)

2. It should keep the previous state (the memorization mechanism is often used in multi-step processes)

3.Limit the source of truth and minimize the solution to a minimum - if in the context of design the solution creates inconsistencies - it will be even more difficult to implement. Specifically I suggest keeping one card and remembering the state

4. hmm, create a new filter screen? limiting their number to the most necessary? This point will take much more time

1.In a modern, ideally programmed system, filters should work dynamically in real time - however, due to system limitations:

For a card with saved settings:

• a) Let's assume that we search for the target element by the X Y Z parameters

• b) We want to see the results so click apply.(the card closes
  automatically and we see the result)

• c) We see 30 results, but we decide to change the Y parameter to narrow down the results

In this situation, after opening the window with filters, I think that the parameters X Y Z should be remembered from the previous operation (stored e.g. in localstorage) - however, this requires analyzing several scenarios and discussing the most appropriate solution.

Perhaps the filters should be redesigned? (as a sliding div, not a collision modal)

2. It should keep the previous state (the memorization mechanism is often used in multi-step processes)

3.Limit the source of truth and minimize the solution to a minimum - if in the context of design the solution creates inconsistencies - it will be even more difficult to implement. Specifically I suggest keeping one card and remembering the state

4. hmm, create a new filter screen? limiting their number to the most necessary? This point will take much more time