What memory card recovery costs. In sentences, because a table would imply more certainty than a failed card allows.

The first band is the card the bench can read with a reader. Deleted photographs, a quick format in the camera, a card the computer asks to format, a card the phone calls corrupted, a card error after a battery died, a video clip cut short, a card that has locked itself read-only and still reads at speed, a card with corroded contacts that reads once cleaned: each is £149 + VAT after the free look, with the figure in writing before any chargeable work. That covers the imaging, the rebuilding of the tables or the carving of the files, the restoring of dates from the files themselves, the rebuilding of cut video from a reference clip, and the opening of every file before it is listed.

The second band is the card no reader can see, or one that lies about itself. A dead controller, a card at 0 bytes or no media, a card showing 30MB where 128GB should be, a card locked read-only and reading too slowly to copy, a snapped or cracked card whose flash die is whole, a counterfeit whose real capacity must be found and read: each is £299 + VAT, which pays for the chip-off from a board or the test-pad work on a monolith, the descrambling and reassembly in software, and the same file-system work on the result. A card whose flash die has cracked through is not recoverable, and you are told so at the free look with nothing to pay.

The third band is the drives in cards' clothing and the hard cases. CFexpress Type A and B, XQD, CFast 2.0 and SD Express cards are NVMe or SATA solid-state drives inside and are recovered with the SSD bench's tools, from £449 + VAT. Several cards from one shoot or one job, where the files must be reconciled across them, start there too, as do cards that have been through fire or salt water and need extended cleaning before the first read. Priority handling, where the card goes to the front of the bench, is quoted separately and only when you ask for it. VAT is added to every figure at the standard rate.

Mostly, what was done before the envelope. A format accepted, a chkdsk or First Aid run, a recovery tool with a repair button let loose on the card, a wet card powered, a snapped card glued or sanded at home, a dozen readers tried: each is honestly described as more work, or as the end of the road, and each is quoted before you agree. A full format on Windows, which writes zeros over the whole card, is the end of the road and is said so.

What does not move the figure: the brand, the capacity, the file system, the number of photographs, or the kind of camera. A 1TB card costs the same as a 32GB one in the same band; a RAW format the same as JPEG; a SanDisk the same as a supermarket card. A card another laboratory has already worked on, or one opened, sanded or glued at home, sits outside the no data, no bill promise, and you are told that at the outset.

Logging, a number on the card, photographs of both sides, and a read-only assessment on a hardware imager through the reader the card needs: what the card presents, its real capacity measured, its identity registers read, its file system looked at where there is one, and the controller's state where there is not. We identify what happened to the card, what has been written since, and how much can honestly be read back. You then get an account of what we found, what we expect to recover, and one price. If you say no, the card comes back to you at our expense, and nothing is owed.

The warranty replaces the card. SanDisk's says in writing that data recovery is not covered under this warranty and is not part of the warranty returns process, and that the original card is not returned to you; Samsung's says data recovery is not covered and that Samsung is not liable for data loss in any case, and that dashcam and CCTV use is not covered at all. A card sent back under warranty with the pictures still on it is a card gone. Recover first, and claim the replacement afterwards.

Recovery software has a real place, and a narrower one than its advertising suggests: a card that reads cleanly at its right size, a fault in the data rather than the controller, an image made first and the program run against the image, and the results written somewhere other than the card. The software page says how to do that safely and which tools are honest. It cannot read a card the controller does not present, cannot measure a counterfeit, cannot rebuild a video's index from a reference clip, and against a card that is failing every scan is thousands of extra reads of the weakest blocks. The free look is where you find out which yours is, and it costs nothing.