Horizon is not framed as a feature patch; the docs and GIP both describe it as a move toward a modular data-services protocol built from reusable staking and payments primitives.
The decisive mechanism is the provision: providers assign stake to a specific data service, and that service gets its own slashing, thawing, and registration logic on top of shared protocol primitives.
Delegation becomes service-specific. That is a real control-surface change because delegators now choose not only who runs infrastructure but which service-level risk model they are underwriting.
Payments are being generalized too. Horizon treats GraphTally as the first integrated payment collection system inside a broader payments protocol, not as a one-off query-fee hack.
The docs make clear that SubgraphService is only the first Horizon-native service. The GIP explicitly holds open other service types like Firehose, Substreams, SQL queries, LLM queries, oracles, and challenge systems.
The what changes page is especially useful because it shows concrete shifts in operator workflow: service-specific provisions, collateralized payment collection, GraphTally-only gateway receipts, and the intended move to long-lived allocations with freshness requirements.
Specific source signals worth retaining
Horizon overview: best concise articulation of the protocol’s move from tightly coupled subgraph indexing toward reusable data-service primitives.
GIP-0066: strongest source for the provision model, service/provider/payer/receiver vocabulary, and the explicit claim that Graph Horizon should support many service types beyond subgraphs.
What changes docs: strongest source for the operator-facing consequences of the redesign, especially per-service delegation, service-specific registration, payment collateralization, and flexible slashing/arbitration.
Why this entry belongs in the corpus
It gives the corpus a clearer higher-layer comparison point for The Graph-related sub-protocols already cataloged, especially Graph Tally and the Rewards Eligibility Oracle.
It turns The Graph infra into a more decomposed map: lower-layer payment collection, quality-gating oracle policy, and a broader shared staking/payments substrate for future data services.
The durable insight is that a protocol can try to commoditize many service markets by standardizing stake and payment primitives, while still pushing real power into service-specific registration, slashing, dispute, and fee-routing rules.