The Graph is the indexing and query layer that lets applications read blockchain data without running their own infrastructure. Developers publish a subgraph — an open API that defines how to extract, transform, and serve on-chain events — and anyone can query it in GraphQL. By early 2026 the network had served over 1.27 trillion queries to more than 75,000 projects across 60+ networks, alongside newer products for real-time streaming (Substreams) and enterprise SQL access (Amp). What makes it interesting as a DAO is not the technology but who controls it: a work-token economy of Indexers, Delegators, and Curators, governed by a small elected Council rather than by direct token vote.

The GRT work token: Indexers, Delegators, Curators

GRT is a work token — it is put to work securing and directing the network, not held passively for governance. Three roles stake or signal it (docs):

• Indexers stake GRT as collateral, run node infrastructure, and serve queries; they earn indexing rewards (protocol issuance) and a share of query fees. Misbehaviour is slashable.
• Delegators delegate GRT to Indexers without running any infrastructure, taking a cut of the Indexer's rewards in exchange for lending stake and reputation.
• Curators signal on subgraphs they judge worth indexing, earning a slice of that subgraph's future query fees — the network's decentralized answer to "which data is worth serving."

This division of labour is the point: no single party both funds, secures, and directs the network. It also means governance power and economic power are deliberately separated — a Delegator with a large bag has no more say over the protocol than a small one.

Curation is a bonding-curve market

Curation is where The Graph quietly runs one of DeFi's larger live bonding-curve experiments. When a Curator signals on a subgraph, GRT is deposited into a per-subgraph reserve and the Curator mints Graph Curation Shares (GCS) — ERC-20 shares that can only be minted or burned against that curve, never traded peer-to-peer. Being earlier to signal mints shares more cheaply, so early, correct curation is rewarded; the design uses a nested curve (an inner curve per subgraph deployment, an outer curve at the GNS / subgraph level). A 1% tax is charged on initial signalling and 0.5% on auto-migration to a new version, and Curators collectively earn 10% of the query fees a subgraph generates, pro-rata to shares.

The mechanism is itself under governance. GIP-0025 introduced principal-protected curves, and GIP-0058 proposed replacing bonding curves outright with direct indexing fees. That one is worth reading for its modality: the thread still carries Stage: Draft and has had no post since 7 December 2023, so a proposal frequently cited as the curve's successor has sat unmoved for nearly three years. Even a market primitive is a governed parameter here — and a governed parameter can simply stall.

The Graph Council: governance by delegated multisig

The Graph's defining governance choice is that GRT holders do not vote on-chain to change the protocol. Authority sits with The Graph Council, a 6-of-10 Gnosis Safe multisig that balances five stakeholder groups — Indexers, token holders, technical/research experts, backers, and the initial team. The Council ratifies protocol upgrades, sets economic parameters, and administers the community treasury.

Proposals move through the Graph Improvement Proposal (GIP) process: an author drafts a GIP, the community debates it on the governance forum, and the Council votes to accept or reject. In February 2024 the Council added a six-member Technical Advisory Board to shepherd research and the core roadmap. The result is a representative DAO — deliberately fast and expert-led, but one where a token holder's leverage is voice in a forum, not a vote that settles on its own.

From hosted service to a decentralized network

For years most applications used The Graph's centrally-run hosted service rather than the token-incentivised network. That changed in 2024: the hosted service was retired in June 2024, forcing all traffic onto the decentralized network, and the Foundation ran a Sunrise Upgrade Program (allocating up to 4 million GRT) to migrate projects across. In parallel the core protocol contracts moved to Arbitrum One for cheaper staking, delegation, and reward flows. The migration is the clearest test of the whole design — the subsidised training wheels came off and the fee market had to stand on its own.

Where power actually sits

The Council model buys speed and technical competence at the cost of a familiar DAO tension: the people who fund the network (Delegators, token buyers) are not the people who decide it. Supporters argue that protocol-parameter design is genuinely specialist work that a broad token vote handles badly, and that the five-group balance and public GIP process keep the Council accountable. Critics note that "accountable via forum" is weaker than "bound by vote," and that a 6-of-10 multisig is a concentrated point of trust for a network this large. Both readings are fair; The Graph is the reference case for asking a DAO the blunt question — when a decision is made, whose signature actually settles it?

How Caper approaches this

The Graph shows a bonding curve doing one job – pricing attention, where early Curators who signal correctly are rewarded. Caper points the same primitive at a different one: pricing a project's own token at launch. Its bonding curve follows an increasing schedule, so the earliest backers of a caper pay the lowest price and later demand pays more – funding a venture directly from the curve rather than signalling on someone else's data.

On governance the two answer the same question – whose signature settles a decision? – and give opposite answers. The Graph's is a 6-of-10 Council: expert, fast, and accountable through a forum rather than a ballot. A caper's is a ballot and a market, in that order. A proposal offers two to five ranked options, always including a “do nothing”, and members cast ranked ballots weighted w = (t · v) / (V · T); the leading option carries only if it takes 1.5 / option_count of the weight actually cast and is not the “do nothing”. Passing the ballot only earns the right to trigger, which locks the caper's trailing average token price as a baseline and opens the market window – and the action executes only if the token's time-weighted average price across that window is at or above the baseline. There is no council and no multisig, and no expert body either; the price of that is the same one futarchy pays: a market can veto a decision the members already voted for without ever being able to say what was wrong with it.

Where a caper does keep something The Graph deliberately separates is the link between economic and governance standing. On The Graph a large Delegator has no more protocol say than a small one, by design. On a caper the two are joined at exactly one point: both trading and voting mint the same soulbound token – 0.01 v per XRD of gross value on buys and sells alike, and one v per ballot, since the 100 XRD vote fee is charged at that same rate – and holding some is a precondition of exit, which pays out w = (t · v) / (V · T). A holder who has neither traded nor voted cannot redeem from the treasury at all. See how vote tokens are earned.

References

• The Graph — About / network overview (docs)
• Indexers, Delegators, Curators — network roles (docs)
• Curating & Graph Curation Shares (docs)
• The Graph governance & the Council
• Graph Improvement Proposals (GIP repository)
• GIP-0025 — Principal-Protected Bonding Curves · GIP-0058 — Replacing Bonding Curves with Indexing Fees
• The Graph — Wikipedia (founding, hosted-service sunset)