Project Catalyst is the name given to the programme of work the Foundation set out after its change of mandate, and it is what 20% of protocol issuance is meant to fund. It is usually described as eight workstreams.
This entry takes each in turn: what it is in protocol terms, and what is actually standing in the way. That second question turns out to be the interesting one.
1. Make Subgraph Studio fully network-powered
Studio subgraphs are today served by the upgrade indexer, a role Edge & Node operates so that a newly published subgraph has somebody serving it. The workstream is to retire that dependency by making ordinary network indexers earn for serving Studio subgraphs.
The mechanism is Direct Indexer Payments, GIP-0081: an agreement between a payer and an indexer to index a specific deployment for a recurring fee, settled on chain, rather than the indexer being paid out of issuance for allocating to whatever carries signal. GIP-0087 and GIP-0088 bring those agreements on chain and route issuance to them.
What is in the way is not the agreements, which exist. It is that the issuance share directed to them is currently zero, and moving it is a governance decision.
There is also a clock on it that the workstream description does not mention. Edge & Node is transitioning its network operations to the Foundation following the Labs decision, and the upgrade indexer is one of those operations. So this item is not only a nice-to-have improvement on a working arrangement, it is the replacement for a dependency that is being handed over. Whether the handover or the replacement lands first is worth watching, and is a question about operations rather than about the protocol.
2. Onboard new gateway operators
Almost all query traffic reaches indexers through one gateway. That is a single point of policy, not just of failure: the gateway decides which indexer serves a query, and that policy belongs to whoever runs it rather than to the protocol. More gateways is the structural fix.
What is in the way is a coordination problem rather than software. A gateway pays indexers with GraphTally receipts, and an indexer only accepts receipts from a sender it has configured itself to trust. So a new gateway operator must persuade indexers, one at a time, to add them. Nobody outside that social process can solve it, and it is the reason this item stalls even where the software works.
3. Memory for AI
Persistent, user-owned memory for agents, reachable over MCP and served by the network. The pitch is that an agent’s state is exactly the kind of thing you would rather not hand to a single provider.
What is in the way is design rather than plumbing. Encryption that is genuinely user-owned fights semantic search: if the service cannot read the data, it cannot index it for meaning, and the workarounds trade away one property or the other.
4. Finish the Substreams data service
Substreams is a lane of its own and is not currently a Horizon data service, which means it does not settle through the protocol’s payment machinery. Making it one puts Substreams packages behind the same receipts, escrow and provisioned stake as everything else.
What is in the way is the ordinary cost of shipping a data service that handles money: an external audit, and at least one provider willing to run it.
5. The RPC data service
Plain JSON-RPC access, metered and settled through the protocol. It is the least Graph-shaped item on the list and arguably the largest market: every application needs RPC, and almost none of it is paid for through anything resembling this.
What is in the way is the same pair as Substreams, an audit and an operator, and one thing more. RPC is a commodity with entrenched incumbents and thin margins, so the question is not whether it can be built but whether anyone will serve it here.
6. A multi-product Studio experience
One surface for subgraphs, Substreams, the Token API and whatever comes next, rather than a product per thing.
What is in the way is mostly not engineering. Subscription billing needs a legal entity that can invoice and be invoiced, and running managed indexing pipelines is an operating commitment rather than a feature.
7. Chain integrations data service
Turning chain integration into a metered service, so that the work of supporting a new chain is paid for through the protocol rather than absorbed.
What is in the way is honestly stated in its own risk note as a business-model and governance problem rather than an engineering one. Somebody has to decide what an integration is worth and who pays, and no amount of contract code answers that.
8. Institutional audit layer
Selective disclosure over signed receipts: letting a counterparty verify that a particular transfer happened without publishing the whole history around it. The target is institutions with regulatory reporting duties.
What is in the way is entirely institutional. SOC 2, a design partner willing to be first, and a legal entity that can hold attestation authority and be sued if the attestation is wrong.
The pattern worth noticing
Read the eight blockers together rather than one at a time.
Barely any of them is “the code does not exist yet”. They are a governance vote that has not happened, indexers who have not been persuaded, audits nobody has paid for, providers who have not volunteered, a value-capture policy nobody has set, and company formation. Even Memory for AI, the most research-shaped item, is blocked on a design tradeoff rather than on implementation effort.
That observation cuts in more than one direction, and it is worth being even-handed about. It is a caution against assuming issuance converts straight into delivery. It is equally a caution against assuming that working software from any quarter converts straight into adoption: a data service nobody provisions and a gateway no indexer trusts are both complete and both inert.
Independent assessments
There is at least one public attempt to score how much of this the community has already built, published by The Night’s Watch on 2026-08-28 with an offer to collaborate, alongside repositories that are public and MIT or Apache licensed.
It is worth reading and worth reading for what it is: an argument by a party with an interest in the answer, about their own software, scored by a method they chose and describe as editorial. The repositories are checkable, which is more than most claims in this ecosystem offer. The scoring is a position. This site does not carry either side’s numbers, because a coverage percentage is a judgement dressed as a measurement, and the same test applies to a Foundation progress report.
Before reading on: if the code mostly exists, why is so little of it in production?
Because a data service is not a program, it is a market with a program in it.
Every item on this list needs at least two parties who do not report to each other. A gateway needs indexers who trust its receipts. A data service needs providers willing to provision stake against it and consumers willing to pay. An audit layer needs an institution prepared to be the first to rely on it. None of those are things the author of the software can supply, and all of them are slower than writing the software was.
This is the recurring shape of the whole protocol rather than a quirk of this roadmap. Indexing rewards exist because somebody had to be paid to show up before there was demand. Curation exists because somebody had to signal what was worth serving before queries proved it. The protocol is largely a set of answers to “how do you get the second party to arrive”, and Catalyst is the current instance of that question rather than an exception to it.