Curation is the role people find hardest to place, because unlike the others it does not map onto anything familiar. An indexer is an operator. A delegator is a backer. A curator is neither, and calling it “staking on subgraphs” obscures what it actually does.
The problem curation solves.
Indexing capacity is finite. Every indexer has a limited amount of stake, a limited number of machines, and a finite amount of attention. The network has thousands of published subgraphs.
So: which ones get indexed? Not “which ones are popular”, because a newly published subgraph has no queries yet and needs to be indexed before it can have any. Not “all of them”, because the capacity does not exist. And not “whichever ones the Foundation picks”, because then you have a committee deciding what data the world can read quickly, which is the arrangement the protocol exists to avoid.
Curation is the mechanism that answers it without a committee. Anyone can deposit GRT against a subgraph they believe will be queried. That deposit is public, and indexers use it to decide where to point their capacity.
Why a forecast has to cost something.
If signalling were free, it would carry no information. Everyone would signal everything, indexers would learn nothing from it, and the protocol would be back to having no way to allocate capacity.
Making it cost capital makes it selective. You will not signal on a subgraph you do not believe in, because that capital could have been signalling something you do believe in. The cost is what makes the number mean something.
On top of that there is a deposit charge. Entering costs 1%, and it is burned rather than paid to anyone. Nobody profits from your entry, which removes the obvious way to game the mechanism by cycling in and out.
What a curator gets back.
Curators receive a share of the query fees generated on the subgraph they signalled.
Note carefully which pot that is. Curators are paid from query fees, which are real revenue from real demand, not from indexing rewards, which are issuance. So a curator’s return is directly tied to whether the forecast was right. Signal on a subgraph that turns out to be heavily queried and you earn. Signal on one nobody touches and you earn nothing, however elegant the subgraph.
This is a cleaner alignment than the other roles get, and it is worth appreciating. A delegator can earn from an indexer allocating to subgraphs nobody queries, because indexing rewards do not require queries. A curator cannot.
The consequence indexers care about.
The relationship runs the other way too, and this is the part that surprises developers. A subgraph with no signal is generally not worth an indexer’s stake, because allocation towards it earns no indexing rewards.
Publishing a subgraph and waiting is therefore not a strategy. If nothing signals on it, nothing indexes it, and it does not get served. This is why the documentation suggests that developers curate their own subgraph, with 3,000 GRT given as a starting figure, noting that the right number moves with network activity.
Before reading on: why can a developer not simply pay an indexer directly to index their subgraph?
They can, in the sense that direct arrangements exist and there is protocol work aimed squarely at this. But signal solves something a bilateral deal does not.
A private arrangement with one indexer gets you one indexer. If that operator has an outage, changes their mind or leaves, your application loses its data source and you start negotiating again. You have reproduced the single-provider dependency the protocol exists to remove, with extra steps.
Signal is public and impersonal. It tells every indexer at once that this subgraph is worth capacity, and any of them can act on it without asking you. What you are buying with signal is not one indexer’s attention. It is the ongoing possibility of several, which is the property that makes the data source durable.
What curation is not.
- Not governance. Signal does not vote on anything.
- Not a quality rating. You are forecasting demand. A poorly written subgraph with heavy usage is a better signal target than an elegant one nobody queries.
- Not slashable. There is no curator misbehaviour to punish. The risks are entirely about being wrong, which the strategy lesson covers.
- Not a lock-up with a timer. You can unsignal, and on Arbitrum the cost of doing so is not path-dependent. The next two lessons explain why that sentence needs saying at all.
Curators are paid from:
Query fees, which is revenue somebody actually paid. That ties a curator's return directly to whether their forecast of demand was right. A delegator can earn from allocations to subgraphs nobody queries, because indexing rewards do not require queries. A curator cannot.
Why does signalling have to cost capital?
Cost is what makes the number selective and therefore meaningful. The deposit charge on top is burned rather than paid to anyone, which removes the obvious way to game the mechanism by cycling in and out.
A developer publishes a subgraph and nobody signals on it. What is the likely outcome?
Signal is what makes a subgraph worth an indexer's stake. Publishing and waiting is not a strategy, which is why the documentation suggests developers curate their own subgraph to get it served at all.
0 of 3 answered