Preview tools show you what they found. Most never show you a denominator — so you can't tell whether "recovered" means 100% or 5%. We report the row count your database claims for every table, then how many we actually got back, and then the rows we couldn't read. If a table's definition page is unreadable, we say there's no count to check against instead of leaving that table out.
No. The engine runs in your browser and the file is only read into local memory. The page never sends it to a server, and you can disconnect from the network once it has loaded.
Both .mdb and .accdb, from Access 95 through 2021 and Microsoft 365.
No — the engine is a WebAssembly build that runs in your browser, so it works on Windows, macOS and Linux with nothing else installed. That also sidesteps the usual catch: desktop recovery tools that will only save the result if Access is installed on the machine, and online services that want your file uploaded first. This is neither.
The recovery assessment: for every table, the rows the database declares, the rows we read back, the rows we hit but could not safely parse, and the rows we never reached — plus page health, the object list, and 10 real rows you can read before unlocking anything. None of those counts is an estimate; they are what the engine read out of your file. If the database turns out to be healthy, you can export it here as SQLite, CSV, JSON or SQL at no charge.
Files up to 300 MB run smoothly. Above that the check asks you to split or archive the file first, and 500 MB is the hard limit. Very few Access databases get near it.
No, and I don’t want it. Your browser generates a technical report — page structure, error codes, which pages couldn’t be read. It has no data values in it, no file hash, and not even your table names. That report is what I work from. Your database never leaves your computer: if you want my review, you send me a report, not the file.
The $29 rescue is the engine doing the work on your machine. Expert Review is me, reading your report and writing something specific to your database. If I can’t, you don’t pay.
Repairs at this level aren’t about names or values. They’re about which pages are reachable, what the allocation map says, how the text is encoded. I change the reading strategy and send that back as a small settings file. Your engine picks it up, on your machine.
Then you don’t pay. Not “contact us for a refund” — I issue it. Some databases are beyond reading, and the honest answer is usually “restore from a backup”. I’d rather tell you that on day one than sell you a week of hope.
Two different questions, so two answers. We can tell you how many rows we recovered, how many we couldn’t read, and how many we never reached — those numbers are measured. We can’t verify that every value inside a recovered row is unchanged; some damage leaves no trace to detect. That’s why we show you real rows before you pay — check them against what you remember.
Against files we didn't make, and didn't pick to look good: mydb_7, mydb_10 and mydb_11 — a third-party set of corrupted databases that anyone can download, with destroyed page headers and byte-rotated records. On those three, the engine recovers 99.2–99.4% of rows. Two limits matter. The count is against the repaired reference files that ship with the set, not a byte-for-byte copy of an original. And it describes those files, not yours. The readout above belongs to one sample database. Your own file reports its own numbers, in the free assessment, before you pay anything.
Then it says Unknown, and tells you why — usually that the table’s definition page couldn’t be read. It never silently falls back to 0% or 100%, and it never averages an unverifiable table into a number that looks better.
No. It’s a one-time payment for a single rescue package. Expert Review is one-time as well — there is no subscription anywhere in this product, and there won’t be.
No. Your original is opened read-only. Everything we produce is written to new files — including the new .mdb — so you can compare before you decide what to keep.