Metadata Explained

Metadata is data about data. It is what a system knows because you used it, rather than because of what you said. The reason it matters is that the content is often protected and the metadata is not.

A worked example

Consider an end-to-end encrypted message from Alice to Bob. The operator cannot read it. It can still see:

• Alice’s and Bob’s account identifiers
• the time, to the second
• the message size
• the IP address each device connected from
• the client software version
• how long the session lasted, and how many messages followed

None of that is the message. Together, it is a very good description of a relationship and its timing. Add a movement history, and you have a location; add a de-anonymisation result, and you have a name.

The five leaks that recur

Timing. Who talked to whom and when. Often more identifying than content. It is the reason Tor rebuilds circuits and why some protocols pad message sizes.

Volume. How much data, in what pattern. Download patterns identify content; a large transfer at a specific time is a strong signal.

Network location. IP addresses, and the radio-level identifiers devices broadcast. Resolved roughly by anyone with a database.

Device identity. Advertised identifiers, fingerprintable configuration, and hardware serial numbers.

Retention and linkage. The same identifier used across contexts, for years, is what turns many small records into one profile.

What reduces metadata

Measure	Reduces	Cost
Traffic padding	Size and pattern leakage	Bandwidth, latency
Batched or delayed delivery	Timing leakage	Delay
Message batching	Timing and size	Delay, complexity
Metadata-minimising protocols	Server-side records	Requires protocol change
Anonymous addressing	Linkability	Usability, a directory may be needed
Relay or mix networks	Linkability to the client	Latency, the reason for Tor
Encrypted DNS	Name lookups on the local network	Trust in the resolver
Short retention periods	Everything, eventually	Operational work for the holder

Reading a claim about metadata

When someone says a system is private, ask:

1. Does the service learn who communicated with whom?
2. Does it learn when, and how often?
3. Does it learn the sizes?
4. Is there a mechanism to say no to content collection while the system still works?
5. What is retained after deletion is requested, and who can be made to produce it?

A system that answers all five is worth trusting more than one that answers only the first, and considerably more than one that answers none of them.

Sources

• Schneier: Metadata Equals Surveillance — a short, quotable argument that collecting metadata is surveillance rather than a lesser form of it, written in 2013. Useful for the framing, not for a technical taxonomy of what metadata reveals.
• Narayanan & Shmatikov — IEEE S&P 2008.
• de Montjoye et al. — Scientific Reports 3, 2013.