Agenda (Oliver Zerhusen)
Figment joins The Graph as a Core Developer
GIP and Governance Updates
Streaming Fast Update
Semiotic AI Update
Decentralizing Core Developers (Eva Beylin)
The Graph is excited to bring on Figment to its core development team. Initially launching with Edge & Node as the core development team, the recent addition of Streaming Fast and Figment to The Graph family aligns multiple teams around The Graph’s mission of decentralization and supports long term goals.
Figment Joins the Graph (Joseph)
Figment joined The Graph as a core protocol developer focusing on growing the decentralized indexing layer for Web3. They have received a core development grant of 80M GRT to be distributed over 5 years.
“We’re thrilled to have Figment join The Graph as a core dev and grow the developer horsepower,” said Eva Beylin, Director of The Graph Foundation. “The Figment team's experience with node infrastructure and multi-blockchain is unparalleled and their expertise will be invaluable to the community’s work in building out the protocol and The Graph ecosystem.” Figment will join Edge & Node and Streaming Fast as core developers on The Graph.
Figment has 25+ years of combined experience providing services across 35+ proof-of-stake networks. There work on The Graph will include:
- Improving Indexing Performance
- Adding support for native network subgraphs
- Expanding multi-blockchain support
The integration of Figment as a core development team will streamline The Graph deployments on Solana, Polygon, Polkadot, NEAR and Celo as they continue to contribute to the evolution of the core protocol.
GIP and Governance Updates (Ariel Barmat)
Batch GNS Transactions (Deploy & Signal)
Currently when a subgraph developer deploys a subgraph, they must signal the subgraph in a separate transaction. This proposal will allow subgraph developers to deploy a subgraph and mint their initial signal in one transaction.
Subgraph Change of Ownership
There has been community feedback in the forum for a method to change ownership of a subgraph. This would allow subgraph developers to transfer their subgraphs to an official team – as a third-party developer – and improve flexibility.
There is a current investigation by the core development team into how to best implement subgraph change of ownership.
Stable Delegation Yield
The stable delegation yield proposal has been audited and is under review by indexers and delegators. An in-depth discussion was covered during Core Dev Meeting #5.
Reduce Curation Tax (Brandon Ramirez)
This proposal would reduce the curation tax parameter from 2.5% to 1%. Curation tax can be burdensome for subgraph developers who are required to pay the tax on every upgrade due to exit and reentry of bonding curves.
The current tax rate is in place to discourage subgraph churn (leading to wasted indexer resources) and to prevent an economic attack referred to as the Subgraph Withholding attack. The current proposal focuses on lowering curation tax while preventing the Subgraph Withholding attack. An in-depth overview can be read here.
Feedback from curators and subgraph developers is highly encouraged in the forums.
Dynamic Curation Tax (Slim Chance)
There are currently three primary challenges the curation community is facing:
- Bots frontrunning newly deployed subgraphs
- High volatility around launch, discouraging curators
- Bait and switch attacks
The proposal to reduce curation tax from 2.5% to 1% is a steady state curation tax. A dynamic curation tax would see curation tax decrease as time goes by. This would disincentivize bot frontrunning and give curators time to analyze subgraphs. Community feedback and ideas are encouraged in the forums.
Arbitration Charter
There are two obstacles blocking full adoption of the arbitration charter and API versioning:
- Graph node unable to produce POIs for failed subgraphs
- Feature detection
The community reached a consensus that indexers should be able to receive rewards for failed subgraphs, but graph node is unable to supply valid POI’s. The graph node team is also working on providing feature detection for subgraphs and integrating it into subgraph studio and the graph explorer’s UI. Both obstacles are being worked on and an update can be expected at the next core dev meeting.
Streaming Fast Update (Alexandre Bourget)
The Firehose is a files-based streaming-first approach to processing blockchain data. It reduces the risk of non-deterministic output and will lay some of the groundwork for future optimization of indexing performance.
Indexing Goals
Providing a way to index data which:
- is capable of handling high throughput chains (network bound)
- increases linear indexing performances
- increases backfilling performance & maximize data agility by enabling parallel processing
- reduces risks of non-deterministic output
- improves testability and developer experience when iterating on subgraphs
- simplifies the operations of an indexer by relying on flat data files instead of live processes like an archive node
- reduces the need for RPC calls to nodes
An in-depth reading of the Firehose approach can be read here. Streaming Fast is looking for feedback in the forums to identify any other pain points present and to review how the implementation of the Firehose is impacting indexers.
Semiotic AI Update (Sam Green & Matt Deible)
Reinforcement Learning Review (Sam Green)
Automated Prosocial Negotiations (Matt Deible)
Automated prosocial negotiations aim to maximize total utility received by network participants interacting with The Graph protocol. The research produced by Semiotic AI is a step toward rewarding pro social behavior during protocol negotiations. This is achieved by opening a dialogue between two neural network agents tasked with a simple negotiation. Depending on the factors – such as negative or positive reinforcement – decided by Semiotic AI, agents are trained to negotiate. Prosocial agents are trained to maximize benefit, but never at the cost of other network participants.
Semiotic AI’s work will impact in-network negotiations such as:
- Maximizing utility received by consumers and indexers
- Consumer-Indexer Direct Negotiations
- Single query prices
- Negotiate long-term Service Level Agreements
Reinforcement learning and goal driven reward functions will have a large impact on the future of the overall health and profitability of all network participants.
Questions for Semiotic AI
Is the average length of the dialog captured in the utility vectors used by each agent?
No, the utility vectors just define how each agent would value each clause if it’s included in the final deal. If the deadline is reached with no deal, then each agent just gets a flat, small punishment to disincentivize wasting of negotiation resources.
Where do you see this going? What do you plan on doing next?
Future questions about Semiotic AI’s research can be directed to:
Matt Deible - Research Scientist
Sam Green – CTO
Stay Tuned!
Join us next month for The Graph’s Core Devs Meeting #7!
Keep up to date by joining discussions in the forum, following The Graph on Twitter or joining the Discord server.
For further discussion on the topics covered in The Graph’s Core Devs meetings, attend Indexer Office hours in the Graph Stage voice channel every Tuesday.
Access a full transcript of Core Devs Meeting #6 here, or watch it on YouTube!