Delegation is the role most people arrive at first, and it is the one most often described wrongly. It is not staking. It is not lending. It is not a savings account with a yield.
What you are actually doing.
You transfer GRT into a pool attached to one indexer. From that moment you own a share of that pool, proportional to what you put in relative to everyone else in it.
When that indexer earns, whether from indexing rewards or from query fees, they keep a cut they have set publicly. The remainder goes into the pool, and your share of the pool is your claim on it. You are not paid per query, and there is no schedule. The pool grows and your share of it is worth more.
Two consequences follow directly, and they surprise people:
- Your return depends on your indexer’s behaviour, not the network’s average. A network doing well overall does you no good if you are behind an operator who is not allocating stake.
- Other delegators arriving after you dilute nothing by arriving, since they buy their own share, but they do change how the indexer’s capacity is used. That matters more than it sounds, and the next lesson explains why.
What you are not doing.
You are not securing the network directly. The indexer’s own self-stake is what is at risk for misbehaviour. Your delegated stake currently is not: Not currently. Horizon added the technical capability for delegated stake to be slashed and it is switched off for the Subgraph Service. That is a real distinction from proof-of-stake networks where delegators share slashing directly, and it is a distinction that could be removed by a future governance decision rather than by a rewrite.
You are not running anything. You have no operational duties, no keys to manage beyond your own wallet, and no uptime obligation.
You are not free to leave. Undelegating takes 28 days, during which you earn nothing and cannot move the capital. This is the constraint that dominates delegation economics and it gets a lesson of its own.
The two costs, one of which is gone.
Entering used to cost you a delegation tax on deposit. Under Horizon that is 0%: the parameter was removed from the protocol entirely.
The cost that remains is the one nobody itemises: the 28 days of capital doing nothing when you leave. It is not charged as a fee, so it never appears in an advertised return, and for anyone who moves between indexers it is by far the larger of the two.
Capacity: the constraint that shapes everything.
An indexer can accept delegated stake up to 16x their own self-stake. That ceiling is the whole reason indexer selection is a real decision rather than a formality.
If an indexer is at or above its capacity, the excess delegated GRT cannot be used by the protocol, and per the official documentation “rewards for all Delegators become diluted because the excess delegated GRT cannot be used effectively within the protocol.”
Read that carefully, because it is the trap. The dilution does not fall only on the delegator who arrived last and pushed the indexer over. It falls on everyone in the pool, including you, whether or not you were there first. An indexer that was a good choice on Monday can be a mediocre one by Friday without doing anything wrong, purely because other people delegated to them.
Before reading on: what should you check about an indexer that a headline APR figure will never tell you?
How much room is left under their delegation capacity, and whether they are actually allocating the stake they already have.
An indexer with plenty of self-stake and modest delegation has room, so your arrival does not push anyone over the ceiling. An indexer already at capacity is advertising a return that the pool cannot deliver to all of its members.
Separately, stake that sits unallocated earns no indexing rewards at all. An operator holding a large pool and allocating a fraction of it is a worse proposition than the headline suggests, and this is visible on a dashboard if you go and look.
What to take from this.
Delegation is a bet on one operator, priced in a currency that is difficult to withdraw. The next two lessons make that concrete: how the split is calculated, and how to look at an indexer properly.