Thegraph.academy is still widely shared - and it's describing a protocol that no longer exists

General
1

thegraph.academy is still widely shared - and it’s describing a protocol that no longer exists

Hey all,

Links to thegraph.academy still get dropped regularly in Discord, Telegram, and onboarding threads as a “start here” resource. I went through it page by page against the current docs, GIPs, and live network state, and I think we should stop recommending it. Not because it was ever bad — it was a genuinely good community effort in 2021 — but because it’s been effectively unmaintained since 2021–2022 and now teaches mechanics that have been removed, replaced, or moved to a different chain.

I’m posting this partly as a PSA and partly to ask: does anyone still have the keys, and is there appetite to either put a staleness banner on it or redirect it?

Maintenance status

  • The flagship page (/ecosystem/the-graph-network/) has a modified_time of August 3, 2021. The homepage was last touched July 2022.
  • The GitBook mirror (docs.thegraph.academy) shows “Last updated 4 years ago” on delegator pages and “3 years ago” on the Advocates and Supported Networks pages. The Quick Start still walks you through the Hosted Service.
  • The Graph-Academy-Hub repo has an open PR from July 2022 that was never reviewed. No releases.
  • The blog’s last post is from March 2021. The Academy X account has no visible activity.
  • The “Network Migration” page still describes the Hosted Service sunset as an upcoming event.

No deprecation notice exists anywhere — it just went quiet. Which is exactly why people keep sharing it.

What it gets wrong (claim vs. current reality)

# thegraph.academy says Current reality Severity
1 Protocol contracts live on Ethereum mainnet; you pay L1 gas for staking/delegating/curating Everything runs on Arbitrum One. 100% of indexing rewards moved to L2 on June 28, 2024; L1 contracts marked end-of-life in Dec 2024 Flat wrong
2 Deploy subgraphs “to a registry hosted on Ethereum”; Hosted Service is live Hosted Service was shut down June 12, 2024. Studio → publish to the network is the only path Flat wrong
3 Query fee rebates follow the Cobb-Douglas production function (formula included) Replaced by the exponential rebate function in GIP-0051; rebate pools removed Flat wrong
4 Curation uses Bancor bonding curves; early curators profit from price appreciation Bonding curves are flat on L2. The “get in early” narrative doesn’t apply Flat wrong
5 Delegators pay a 0.5% delegation tax Removed with Graph Horizon. No protocol-level fee on delegation principal Flat wrong
6 One undelegation at a time, 28-day thaw 28 days still holds, but Horizon allows many simultaneous undelegation requests Outdated
7 Micropayments via state channels / Scalar (Connext) Payments are TAP / GraphTally with Receipt Aggregate Vouchers (GIP-0054) Flat wrong
8 Allocations must close every 28 days; slashing is fixed Under the Subgraph Service allocations can stay open with periodic POIs; slashing is flexible, capped at 10% (default 2.5%) Outdated
9 Roles: Indexers/Curators/Delegators/Fishermen/Arbitrators, single-service protocol Horizon (live Dec 11, 2025) is a multi-data-service protocol: Subgraph Service, Substreams, Firehose, Token API, etc. Indexers are data providers across services Outdated
10 Supports Ethereum, Polkadot, Fantom, Solana 60+ networks; Base and Arbitrum lead query volume Outdated
11 Curators earn “a portion” of query fees 10% cut, governance-adjustable Harmless
12 Yaniv Tal referenced as current E&N CEO Stepped down 2022 Harmless
13 Council page lists original Dec 2020 members Membership has rotated; Technical Advisory Board added Feb 2024 Outdated
14 Advocates “coming soon” Advocates Program and AdvocatesDAO launched 2022 Outdated
15 No mention of Substreams, Firehose, Horizon, TAP, data services These are now the core of the protocol Major omission

What still holds up

Credit where due — the conceptual layer is fine:

  • High-level role descriptions (what an indexer/curator/delegator does)
  • 100,000 GRT minimum indexer self-stake
  • 28-day thawing period exists (though the surrounding mechanics changed)
  • The intuition behind POIs, the Availability Oracle, and why unbonding periods exist
  • The overall mission framing

The problem is that every concrete number, formula, or workflow underneath that layer is wrong.

Why this matters

The dangerous errors are the operational ones. A newcomer following the Academy today would:

  1. Look for contracts and pay gas on the wrong chain
  2. Try to use a dead Hosted Service and dead CLI flags
  3. Model curation risk on bonding curves that don’t exist
  4. Model indexer economics on Cobb-Douglas
  5. Budget for a delegation tax that was removed
  6. Have no idea Horizon happened

That’s not “slightly stale.” That’s a description of a different protocol.

What to point people to instead

  • Official docs: Home | Docs | The Graph — especially /indexing/overview, /resources/tokenomics, /resources/roles/delegating, /graph-horizon/overview, /graph-horizon/what-changes
  • Governance: the GIP repo and this forum
  • Ecosystem narrative: Messari’s quarterly State of The Graph reports, The Graph Foundation blog
  • Live network data: Graph Explorer, and community dashboards like Lodestar (disclosure: that one’s ours) and Graphtronauts tooling for delegators

Ask

  1. If anyone from the original Academy team is still around (Oliver, Zorro, others who contributed) — would you be open to adding a “this content reflects the protocol as of 2021” banner, or redirecting to the official docs?
  2. Everyone else: please stop linking it in onboarding contexts. If you see it shared, gently point to Home | Docs | The Graph instead.
  3. If there’s interest in preserving the conceptual/first-principles material somewhere maintained, I’m happy to help — but it should be rewritten against Horizon, not patched.

Happy to be corrected on any row above.

2

on the last point, about rewriting it against horizon rather than patching it, i went and did that.
learn-thegraph.com is up. 44 lessons across 7 role paths (start, delegators, curators, indexers,
developers, consumers, governance), written against horizon from the start instead of 2021 material with corrections bolted onto it. its unofficial and its ours, so read the official docs first and this second.

the part i actually care about is why the academy rotted, because it wasnt only neglect. every
protocol number was typed straight into prose, so when cobb-douglas got replaced there was no way to know which pages had just gone wrong. so nothing in the rebuild hardcodes a protocol number. theres one registry, 23 entries, each carrying the source url, a verbatim quote from that source and the date a human read it, and lessons render them by key. an unknown key fails the build. every page carries what it was checked against and when, and renders a warning past its date. that doesnt make it correct. it makes it possible for someone who isnt me to find what’s wrong with it.

doing that turned up three places where the official docs contradict themselves, which is probablycworth more to this thread than the site is.

the tokenomics page still says “There is a 0.5% delegation tax which is burned whenever a Delegator delegates GRT on the network”. thats your row 5, except its on thegraph.com rather than the academy. horizon removed it, and a delegator budgeting for a 0.5% haircut on deposit is working from stale docs on the official site.

allocation lifetime is two separate facts being taught as one. the protocol permits indefinite
allocations, the shipped indexer stack still recycles roughly every 28 days, and the horizon changes page says so itself. an indexer planning capacity needs both numbers and most material gives one.

and simultaneous undelegation requests, where the delegating page says 1,000 and the horizon
what-changes page says up to 100. that one is settleable so i checked. MAX_THAW_REQUESTS in HorizonStaking.sol is a private constant set to 1_000, so the delegating page is right and the
horizon page is wrong. you cant eth_call it, private constants get inlined rather than exposed, but
0x03e8 does appear in the deployed implementation on arbitrum one, which is corroboration and not proof.

on your actual ask, a banner is the minimum and a redirect is better. and agreed on the conceptual layer holding up, thats the half worth keeping, which is what the rebuild kept. everything underneath it got thrown away rather than corrected.

repo is nightswatchhq/graph-academy-v2, MIT on the code and CC-BY on the content. fork it, PR it, and if the foundation or anyone else wants to take it over or just point at it thats fine by me. not asking for it to be blessed.