Why files get corrupted, and what can actually be recovered

A file gets corrupted when the bytes on disk stop matching what the program that wrote them expected: a write that was interrupted, a storage device that returned wrong data, a transfer that dropped or mangled part of the file, or software that crashed mid-save. What can be recovered depends entirely on which part is damaged. If the structure is broken but the content is intact, a repair can usually rebuild it; if the content itself was never written or has been overwritten, no tool can bring it back.

What “corrupted” actually means

A file is not a single blob. A video, a PDF, or a Word document is a structure: headers that say what the file is, tables that say where each piece lives, and the pieces themselves (frames, pages, paragraphs). Programs read the headers and tables first and use them to find the content.

Corruption is any mismatch between those parts. The header might claim a length the file does not have. The table might point at offsets that contain something else. A block of content might be zeros where frames should be. From the outside it all looks the same: “The file could not be opened.” From the inside these are very different situations, and the difference decides what is recoverable.

The most useful mental split is structure versus content. Structure is cheap to rebuild because it is derived from the content: you can scan the content and reconstruct the table. Content cannot be rebuilt from anything, because it is the thing itself.

Cause one: the write was interrupted

This is the most common cause by a wide margin. A camera loses power. A laptop runs out of battery during a save. A USB drive is yanked while the progress bar is still moving. An app crashes halfway through writing.

What it leaves behind depends on the order the program writes things. Many formats write content first and the index last, because the index cannot be known until the content is done. MP4 and MOV videos put the frame table at the end; an interrupted recording is all frames and no table. PDFs put the cross-reference table at the end; an interrupted save is all pages and no lookup. A Word document is a zip archive, and zip keeps its central directory at the end as well.

This is good news. The content is on disk. The missing piece is the part that can be reconstructed by scanning. That is exactly what a repair tool does, and it is why interrupted writes are the most recoverable kind of corruption.

The limit is that the content stops wherever the write stopped. A three-minute recording interrupted at ninety seconds has ninety seconds of frames. A repair gives you those ninety seconds, honestly labeled as partial, and nothing more.

Cause two: the storage failed

Flash memory wears out. Hard drives develop bad sectors. SD cards in cameras and dashcams are rewritten constantly and are the most failure-prone storage most people own. When a sector goes bad, reads from it return garbage or fail outright, and whatever part of the file lived there is damaged.

What this leaves behind is a file with holes. The structure may be fine and most of the content may be fine, but a stretch in the middle is wrong. For a video, that is a burst of blocky or frozen frames, or a player that stops at the bad spot. For a PDF, a few pages that render blank or throw errors while the rest open. For a document, text that turns into symbols partway through.

Recovery here means working around the hole. A repair can rebuild the structure to skip the damaged region and deliver everything on either side. What was in the hole is gone unless you have another copy. If the drive is still failing, copy everything off it first; every additional read on a dying device is a chance to lose more.

Cause three: the transfer dropped bytes

A file that was perfect on one device can be corrupt on another because of the trip between them. Interrupted downloads, an AirDrop that was canceled, a cloud sync that stalled, a network share that disconnected, a copy to a USB stick that was ejected early. In each case the destination has a truncated file: the beginning, up to the point the transfer stopped, and nothing after.

This is structurally the same as an interrupted write, and it is recoverable in the same way, with one important extra option: the source may still be intact. Before repairing a half-copied file, check the phone, the camera, the sender, or the cloud. A fresh copy is always better than a repair. The post file-transfer-interrupted-how-to-salvage-a-half-copied-video goes through this in detail.

A rarer failure is a program that simply wrote the file wrong: an exporter with a bug, or a script that edited a PDF and left its table pointing at stale offsets. These are usually easy to recover, because the content is complete and only the bookkeeping is off.

What a repair can and cannot give you

Recoverable: a missing or broken index, table, or directory; a truncated file with intact content up to the cut; a file with a damaged region that can be skipped; a header that disagrees with otherwise valid content.

Not recoverable: content that was never written because the write stopped before it; content that has been overwritten (a camera that looped over the clip, a drive that has been reused); content that was damaged byte by byte throughout; and files that were deleted rather than corrupted, which is a different problem needing a different kind of tool.

Manatee is built around this distinction. Its Repair mode handles corrupt video, PDF, and Word documents and reports one of three outcomes: Fixed when the structure could be rebuilt around complete content, Partial when only part of the content survived (the report says how much, for example 34 of 41 pages or the first eleven minutes of a video), and Not fixable when there is nothing intact to work from. It always writes a new copy alongside your file and never alters the original, so a failed attempt costs nothing. It is not an undelete tool and it does not repair disks; see partial-recovery-explained-what-34-of-41-pages-really-means for what a partial result looks like in practice.

Reducing how often this happens

Most corruption is preventable:

1. Eject drives and cards from Finder before unplugging them, every time.
2. Keep camera and dashcam cards young. Replace the ones that are written to daily every year or two.
3. Let saves finish. When a progress bar is moving, wait for it, especially on external drives and network shares.
4. Keep a second copy of anything that matters. Time Machine, a cloud folder, or a plain external drive all work; the point is that a repair should be the fallback, not the plan.
5. Verify big transfers. Compare file sizes after copying, and for truly important files compare a checksum (shasum in Terminal on both ends).

Questions

Can a file become corrupted just by sitting on a drive?

Yes, slowly. Flash cells leak charge over years and magnetic sectors degrade. This is called bit rot and it is why archives on a single old drive are a risk. Keeping a second copy and occasionally refreshing it is the defense; no repair tool recovers a bit that has silently flipped inside compressed content.

The file opens on one Mac but not another. Is it corrupt?

Probably not. If it opens anywhere, the structure and content are intact; the other Mac is missing a codec or has a different version of the app. See the-file-someone-sent-you-won-t-open-a-quick-diagnosis-guide for the quick way to tell.

Is it worth trying a repair if I think the content is gone?

Yes, because trying is free and harmless. The original is never touched, the attempt takes seconds, and the outcome tells you something definite either way. What is not worth doing is running several tools that rewrite the file in place; each one can erase evidence the next one needed.