The middle relay, and how little a middle position holds

The relay in the middle of a circuit receives traffic from one relay and hands it to another. Both of its neighbours are machines, not people.

The view from the middle

A middle relay holds two connections that belong to your circuit: one back towards the relay that gave it the cells, one forward towards the relay it passes them to. It keeps an identifier so it knows which cells belong together, it removes one layer of encryption using a key negotiated with your client, and it forwards what is underneath without being able to read it.

That is the entire job, and it is a genuinely small one. The operator of that machine can see two relay addresses, a count of cells in each direction, and the times those cells arrived. Nothing in that set of facts names a person, a place, a site or an account.

Sees

The address of the relay before it and the address of the relay after it. Both are relays.

Sees

How many cells passed, in which direction, and when.

Does not see

Your network address, your ISP, your country, or that you exist as a distinct person.

Does not see

The destination, the onion address, the page, or a single byte of readable content.

Does not know

Its own position in the path. Being second and being third look the same from the middle.

The last line of that list is the interesting one

Positions are not labelled in Tor traffic. A middle relay receives cells from another relay, which might be a guard carrying a client, or might be another middle carrying traffic for somebody else entirely. No field says which.

That is the mirror image of the blindness in the entry guard card, where the first relay cannot tell how many hops follow it. Both ends of a middle position are opaque, on purpose.

Volume and timing, and how little they say alone

What a middle position genuinely has is a stream of quantities. A circuit was built at some moment, carried a certain amount of traffic, went quiet, carried more, and closed. Read on its own, that is a shape with nobody attached to it.

A busy relay carries many circuits at once for unrelated clients, each producing a shape of the same kind. Without an identity at either end, one more pattern in that pile does not resolve into a person, a household or a service. It is data about a circuit, and a circuit is not a name.

The concern people have read about, correctly, involves a party able to observe two different points at once rather than one. That is not what a middle relay is, and this card is not going to write up how such an observation would be conducted. On its own this position is weak, and it is weak by construction.

The size and rhythm of the traffic is described in the shape and size of the traffic, where it matters more, because that observer already knows who you are.

An onion connection has more middles, not fewer

For an ordinary website, the usual path is short and ends at an exit. For an address ending in .onion there is no exit, and instead both sides build their own path into the network and those paths are joined at an agreed relay. Counting the whole arrangement, more relay positions are involved than in an ordinary fetch.

That is often read as worse, and it is not. Each additional position sees the same narrow slice as any other middle: two neighbours and a cell count. More middles means the same small view distributed across more unrelated operators. The correction about exits, which several cards here lean on, lives in exits and onion addresses.

The relay where the two paths meet is worth naming separately. It holds two circuits and passes cells between them. It does not learn either end, and it does not learn the address being reached, because it was told where to be by a token rather than a name.

Why this card is short on drama

A catalogue filed by observer has to be willing to say when an observer is unimportant, or the whole thing becomes a list of things to be afraid of. The middle position is the clearest case on the site. It is the least informed party in the chain, and a card that inflated it would be dishonest in the same way that ignoring the far end would be.

If you are looking for the position in this arrangement that actually accumulates useful information about you, it is not in the Tor section at all. It is the machine you are typing into, and it is filed under browsing listings while signed in and the cards around it.

What changes the answer

5 things change how much this action gives away. None of them takes it to zero, and none of them is a promise.

1. Building a new circuit replaces the middlesA fresh path for a site draws new middle positions, so the old ones stop receiving anything from you and hold only a closed record. The first hop normally stays where it is. A new circuit for this site is honest about how small that lever is.
2. Fewer, shorter sessions produce fewer circuit recordsEvery hour a client stays connected produces circuits and rebuilds, spread across relays. A short deliberate session leaves fewer traces spread across fewer operators. See how long the session stays open.
3. Nothing here changes what your own line seesMiddle relays are invisible to your ISP, which only ever sees the first hop. The network observer is filed under your network.
4. The far end is unaffected by every relay choiceA different middle does not make you a different account. Session identity lives on the server, so a swapped relay changes nothing about what signing in already established.
5. Read the exit correction before drawing conclusionsMost fear attached to relays is really fear of an exit position, which is absent from an onion connection entirely. Exits and onion addresses is the card that sorts that out, and it is the one to read if you only read one in this section.

What this card is not

This card does not describe how anyone would combine observations from several points, and it does not claim a middle relay is harmless in every conceivable arrangement. It describes what one middle position is handed.