Append-only

Append-only

Append-only refers to a way of storing data in which new entries may only be added at the end. Existing entries are never overwritten or deleted, so the entire history is preserved.

Append-only is a rule for how a computer is allowed to write data into a file. Only one thing is permitted: adding something at the end. Whatever is already there stays unchanged. So you cannot correct a line and you cannot delete one. The term literally means “only appending”. A good comparison is a cash book written with a ballpoint pen: a mistake is not erased but corrected with a new line underneath.

Why no one can secretly change anything anymore

If data can be overwritten, the past disappears without a trace. Afterwards, there is no way to tell whether a value was always like that. With append-only, it’s different. Every change is an additional entry with a timestamp. The old version remains visible. This makes manipulation noticeable and audits easy.

That is why laws even mandate this way of working in many areas. Accounting systems must keep bookings unchangeable. Hospitals and banks also log who saw which record and when in this manner. Experts refer to this as an audit trail, meaning a seamless trace for later checks.

A second advantage concerns the technology itself. If no one changes existing data, many programs can read at the same time without interfering with one another. This saves complicated safeguards. In addition, such a data set can easily be copied to multiple machines, since you only need to send along the new entries.

The log and the state behind it

The basic technical form is called a log. This is a long list of entries in the order they were created. Every new entry goes at the end. An example from a bank account: first it says “account opened”, then “200 euros deposited”, then “50 euros withdrawn”. Nowhere does it state the account balance directly.

The system calculates the current state by going through all entries from the beginning. In the example, that comes to 150 euros. Because this would take too long with millions of entries, intermediate states are stored. Such snapshots are called snapshots. The program then starts at the last snapshot and only calculates the remainder.

A common misconception is that append-only means unlimited growth with no possibility of deletion whatsoever. In practice, very old sections are indeed removed or consolidated by someone eventually, for example after a ten-year retention period. The difference remains: deletion happens in a planned way and as a whole, not secretly at a single spot. A related term is immutability. Append-only is the concrete writing rule, immutability is the general property behind it.

From the blockchain to the chat history

The principle is best known from blockchain, the technology behind Bitcoin. There, blocks of transactions are chained together and linked through checksums. Retroactively changing an old block would invalidate all following ones. This is why the term often appears in news articles alongside cryptocurrencies.

However, append-only is encountered far more often invisibly in the background. Databases first write to such a log before storing data. If the server crashes, the log is processed after the restart. Version control systems for program code such as Git also work this way: old states remain retrievable, every change is a new entry.

In the AI industry, the principle plays a role in traceability. Companies log which training data went into which model and who changed a setting. In the event of a complaint, this allows reconstruction of how an answer came about. Incidentally, your own chat history in an AI assistant is also built append-only: new messages are added at the bottom, older ones remain in place.

Related Products

Latest News

Subscribe free. Unsubscribe the second it sucks.

High-signal news across AI, business, UX, and tech. Every morning.