That's why you don't just send it off blind, but rather contact the company first. Any reputable company should be able to give you a quote immediately for initial investigation and possible return (or disposal) of the card; with that initial investigation out of the way, they will be in a much better position to tell you the cost for extracting the data.

Suppose that we have a perfect 70-300 mm f/4-5.6 zoom lens. At the short end and wide open, the aperture diameter will be 70/4 = 17.5 mm. (At f/5.6, it will be 70/5.6 = 12.5 mm diameter.) At the long end and wide open, the aperture diameter will be 300/5.6 ~ 53.6 mm. In this case, the physical aperture actually did grow larger as we zoomed in, even though the divisor ("f-number") also grew larger. Canon's 70-300/4-5.6 has a 58 mm filter thread, so the front element size is not the strictly limiting factor here.

I'm actually well aware of much of this myself too, and for me, part of the solution was to run bulk storage with RAID-6 with checksumming (specifically, ZFS raidz2). This gives me double redundancy, which means that errors can be corrected even if one of the constituent disks outright fails and another has developed an as of yet undetected error. The checksumming allows the system to, even if an error is somehow uncorrectable, likely at least tell me that the error exists and where. Regular through-reads and validation of all data hopefully catches any bitrot before it has had time to spread.

Sure, but I think you missed my point. It's not hard to set up redundant storage; and it's certainly not harder than non-redundant storage to manage, if done properly. The issue crops up when you have specifically two-way mirroring (such as two-way RAID-1) without any further integrity checks, and when asked for the same block, drive 1 in the array says X and drive 2 in the array says X', with some large or small difference such that X≠X'. Without additional data, this is an unresolvable situation; the computer cannot know which copy, if either, is correct; it can only know that they differ.

To solve this with a simple mirroring setup you'd typically need at least three-way mirroring so that you'd have two out of three copies which hopefully agree, or you need some kind of checksumming in place to determine which copy is undamaged. (ZFS, Btrfs and I believe ReFS does checksumming to validate data on read. At least ZFS can be configured to do arbitrary n-way mirroring.) Outright storage failure, even partial failure (some sectors simply unreadable) is the easy case to handle.

You mention RAID-1, which is simple mirroring. Fine, but what if you're doing the typical two-way mirroring, and one copy develops an error and the other doesn't (or even develops a different error)? How do you decide which copy is correct?

The tiles seem perfectly visible to me. Are you sure this isn't a problem with your monitor's ability to display the colors involved?