# The Graph Horizon’s Incentive Evolution

## Introduction

The Graph Network is preparing to transition to [Horizon](https://github.com/graphprotocol/graph-improvement-proposals/blob/main/gips/0066-graph-horizon.md), its most significant protocol upgrade aimed at clearing technical debt, adding new data services and addressing years of accumulated feedback. Horizon is also the stage on which long-running debates about incentive design will be resolved. When The Graph began its journey, it faced no serious competition. Now challengers arrive with radically different architectures and critics increasingly label The Graph's design as over-engineered for the task of querying blockchain data. But The Graph has a bold thesis and standpoint for decentralized data.

Yet the more revealing story lies in economics. In a decentralized network, incentive design determines everything—and changing incentives reveals how governance actually operates. This piece traces three Graph Improvement Proposals that tackled the same economic friction from different angles. Two proposals [GIP-58](https://github.com/graphprotocol/graph-improvement-proposals/blob/main/gips/0058-Replacing-Bonding-Curves-with-Indexing-Fees.md) and [GIP-63](https://github.com/graphprotocol/graph-improvement-proposals/blob/main/gips/0063-optimizing-query-fees-sent-to-indexers.md) identified fundamental inefficiencies but never reached voting thresholds. A third proposal [GIP-81](https://github.com/graphprotocol/graph-improvement-proposals/blob/main/gips/0081.md) builds on their analysis and advances through governance.

I'll compare them side-by-side across technical implementation, economic mechanisms, and governance dynamics to develop a mental model for how the protocol’s economic surface is evolving.

## TL;DR

- **GIP-58** aimed to swap bonding-curve [curation](https://thegraph.com/docs/en/resources/roles/curating/) for posted network-wide prices backed by collateral and [POI](https://thegraph.com/docs/en/indexing/overview/#what-is-a-proof-of-indexing-poi) verification, promising predictable indexer income; it surfaced the need for better cost telemetry and a concrete contract stack before such a shift could land.
- **GIP-63** suggested setting curator and protocol cuts on Arbitrum query fees to zero, immediately boosting developer/indexer capture; it raised the open question of how the network would reward information discovery once curator incentives vanished.
- **GIP-81** reintroduces per-work payments through gateway-authored agreements with escrowed vouchers and an MVP rollout, clarifying operational steps; it still must address how gateway-centric coordination and Horizon dependencies maintain decentralization at scale.

## Original mechanisms (58, 63) — what they proposed, why they may have failed
GIP-58 shifted incentives to posted network-wide prices per "subgraph gas" unit, with POI-based settlement and consumer-side selection; it simultaneously pulled issuance reform into Graph Horizon and initially removed bonding curves entirely. The proposal admitted core parameters (subgraph gas estimation, selection algorithms) were uncertain, noted that the collateral escrow and issuance modules tied to Horizon would be specified later, and offered [no smart-contract specification](https://github.com/graphprotocol/graph-improvement-proposals/blob/main/gips/0058-Replacing-Bonding-Curves-with-Indexing-Fees.md#detailed-specification), leaving governors to underwrite [undefined execution risk](https://github.com/graphprotocol/graph-improvement-proposals/blob/main/gips/0058-Replacing-Bonding-Curves-with-Indexing-Fees.md#so-what-about-graph-horizon).

The authors were pushing against time and telemetry constraints, and an MVP could have given stakeholders the confidence to iterate further.
Amendment later backpedaled to run alongside curation and to drop issuance changes, signalling coalition resistance and diluting the original "replace bonding curves" narrative—this likely eroded clarity on objectives.

GIP-63 proposed to remove the 10% curator share and 1% burn on [Arbitrum](https://thegraph.com/docs/en/archived/arbitrum/arbitrum-faq/) query fees via parameter calls, explicitly acknowledging curator revenue would disappear and many would exit. The draft never advanced to a vote; a plausible explanation for the stall is value transfer and coordination gaps: curators faced an immediate haircut, while [delegators](https://thegraph.com/docs/en/resources/roles/delegating/delegating/) would have enjoyed higher query-fee participation because the full 11% haircut flowed back to indexers and into their shared-revenue agreements, yet they lacked clear telemetry to validate those gains, and indexers still confronted undefined pricing telemetry and Horizon dependencies. That combination made it hard for large stakeholders to back the package without management-provided roadmap or shared data (assumption grounded in the cited incentive shifts).

| Mechanism | Low telemetry / uncertain cost proxies | High telemetry / observable cost proxies |
|----------------------------------------|-------------------------------------------------------------------------------|--------------------------------------------------------------------|
| **Legacy bonding curves + fee haircuts** | Indexer: thin margins, participation at risk | Indexer: modest surplus if demand steady; curator able to justify bonding |
| **GIP-58 posted prices + POI enforcement** | Indexer: negative unless collateral priced via conservative oracle | Indexer: positive surplus; curator accepts posted price if discovery premium |
| **GIP-63 zero protocol/curator tax** | Indexer: surplus improves but curator exits, raising selection search costs | Indexer: high surplus but curator participation breaks without new bounty signal |

*Table 1: Indexer payoff vs. telemetry quality (participation binds near zero surplus)*

Participation constraints bind where indexers cannot validate costs (low telemetry case) or where curators lose upside (fee reset column). GIP-58 relaxes the indexer constraint only when telemetry improves, whereas GIP-63 relaxes indexer margins immediately but violates curator participation unless alternative discovery rewards appear.

Taken together, GIP-58 and GIP-63 mapped the economic terrain the network needed to understand. They introduced unit pricing anchored in verifiable work, quantified how much value leaked through legacy fee splits, and framed the issuance-versus-cost debate Horizon now inherits. Their authors ran into practical constraints—limited telemetry, compressed specification timelines—but the conceptual breakthroughs defined the design space subsequent proposals operate within.

## New proposal — what changed, why it might fare better

GIP-81 reframes the mechanism around gateway-authored indexing agreements with cancellable on-chain contracts, POI-linked payouts, and public price disclosures, preserving the familiar "price per unit of work" narrative without forcing immediate curation deprecation. It introduces a concrete escrow/voucher architecture that removes bilateral trust: gateway-funded escrows, signed vouchers, and Subgraph Service-controlled releases tied to POIs and stake checks.

An off-chain MVP ("Dipper") gives management a phased rollout to learn pricing data, aligning governance expectations with demonstrable operational experience before on-chain changes.

Measurement is operationalized through per-block and per-entity pricing, with acknowledged need for future research toward a subgraph-gas metric, signalling awareness of telemetry gaps that plagued GIP-58. The design inherits the earlier proposals’ insights and layers in a richer instrumentation loop so pricing data, QoS expectations, and dispute signals can evolve before the protocol ossifies around them.

Dependencies on Horizon/Subgraph Service are explicit, allowing governors to sequence approvals with the wider program and increasing management alignment (assumption based on dependency listing). In effect, GIP-81 operationalizes the prototypes from GIP-58/63: it keeps per-work pricing, adds an agreement workflow that can absorb iterative feedback, and packages the rollout in stages that reduce coordination risk.

## Side-by-side comparison — continuities and divergences

### Continuities
- Per-work pricing and POI-based verification remain the backbone across proposals.
- Each design relies on collateralized guarantees so payers and indexers can transact without mutual trust.
- Horizon-era components are assumed in all three paths, whether implicit (GIP-58/63) or explicit (GIP-81).

### Divergences
- Incentive distribution: GIP-58/63 concentrated value with indexers at the expense of curators; GIP-81 routes incentives through gateway-led agreements, raising centralization considerations.
- Deployment path: The earlier package lacked contract specs and tied changes to issuance reform; GIP-81 offers an MVP and defined contract responsibilities for escrow, vouchers, and Subgraph Service integration.
- Measurement: GIP-58 gestured toward future “subgraph gas” research without an interim proxy; GIP-81 adopts block/entity proxies today and plans iterative refinement.
- Governance feasibility: The early combo demanded immediate overhaul and simultaneous fee redistribution; the later proposal sequences change through Horizon dependencies and MVP learnings, improving odds of meeting voting thresholds.

## Scenario analysis

**Gateway concentration:** Escrow-backed gateways consolidate bargaining power, indexers rely on voucher pipelines, and curation discovery fades. Reintroducing a protocol tax on gateway spreads—i.e., re-targeting the Subgraph Service treasury cut that supersedes `L2Staking.setProtocolPercentage` back into a 3–5% band so voucher payouts [skim to the protocol](https://github.com/graphprotocol/graph-improvement-proposals/blob/main/gips/0063-optimizing-query-fees-sent-to-indexers.md?plain=1#L32)—and raising penalty weights on missed POIs by increasing the Horizon Dispute Manager’s POI fault weight (the slashing coefficient tied to voucher settlement) shift surplus back toward network-wide enforcement, while reinstating curator bounties on scarce coverage through the Subgraph Service discovery hooks reduces gateway gatekeeping.

**Diverse competition:** Multiple gateways publish prices, indexers self-orchestrate, and curator-sourced intelligence stays rewarded. Lower penalty weights keep experimentation cheap, while optional protocol tax toggles remain off. If competition slips, governors can nudge curator bounty multipliers upward or temporarily tighten penalty weights to deter free-riding, keeping the market elastic without reverting to heavy-handed centralization.

**Risk & compliance:** Gateway-led coordination generates centralization pressure, and voucher redemption flows invite Sybil routing unless stake-weighted identity and rate limits are enforced. Detailed performance reporting can leak sensitive deployment data. Mitigations include quota caps per gateway, anonymized telemetry aggregation, and staking-weighted challenge audits that preserve privacy while maintaining Sybil resistance.

## Conclusion

Taken together, these proposals sketch a continuous arc: bold economic hypotheses in GIP-58 and GIP-63 exposed where curation-based incentives were leaking value, while GIP-81 translates their scaffolding into a roadmap the network can realistically ship under Horizon. The decisive question now is whether the MVP can validate per-work pricing with enough fidelity and transparency to keep participants coordinated. If the Dipper rollout demonstrates that pricing proxies map to real costs, that gateway-led agreements can be monitored without eroding decentralization, and that curators and delegators still have space to surface information, then the ecosystem will have proven that the earlier experiments were not failures but foundational steps.

The task for governors, operators and community is to nurture these feedback loops—collecting metrics, publishing accountability reports, and probing for centralization risks—so the network can iterate on solid ground rather than resetting the mechanism design cycle yet again.