SN 58. Rank 21 of 32 by emission
greevils
A Bittensor subnet where autonomous trading agents go head-to-head with human traders on live Hyperliquid markets.
In their own words · greevils.aiSN58 Greevils is a trading firm's tryout desk where robots and human traders sit side by side, each trading their own money on a live exchange, ranked on one board, with the robots locked in sealed booths.
The commodity, explained from zero
The front page says it in two lines: "Agents against the humans. Real money on the line." Greevils is "A Bittensor subnet where autonomous trading agents go head-to-head with human traders on live Hyperliquid markets." Hyperliquid is a crypto exchange for perpetual contracts, bets on a coin's price that never expire. An agent is a program a miner writes that places those bets on its own, the robot at the next seat.
The subnet produces the tryout's scorecard, a ranked record of trading skill where every account, robot or human, is measured on the same five factors: return, real profit, drawdown (how far the account fell from its peak), losing-day volatility, and consistency.
Who pays: the materials name no buyer of the agents or their signals. The only money changing hands beyond the traders' own is a builder fee of about 5 basis points, 0.05 percent, that every trade must route through the Greevils app, stated only in the validator README. What a third party would buy is not said.
Why it is on Bittensor at all
An ordinary tryout trusts the firm running it. Greevils' answer is sealed execution: "Every agent executes inside a Confidential Space TEE," a locked booth on a Google Cloud machine proving the code ran unchanged, holding "trade-only keys" that "can move the book but can never sign a withdrawal." Here the comparison breaks: at a real tryout the firm's money is on the desk and the firm pays whoever passes, while here every entrant trades their own money, the prize is subnet emission, and the agents' share goes unclaimed today.
How the work gets done
Miners package an agent, a directory with an entry.py file, and deploy it to a confidential virtual machine they rent, the sealed booth; the agent trades by HTTP calls to a harness that holds the key. Every account must stay above a $1,000 floor; falling below it or skipping the builder fee means permanent disqualification. Validators pull each account's daily equity from Hyperliquid, run a pairwise tournament over overlapping windows, and set weights, the numbers that decide who gets the subnet's share of newly minted TAO. Agents earn only once they are open-sourced, approved on chain by the highest-staked validator, and past 60 days of eligible trading; "Anything short of all three earns 0." The validator README adds that the agent classifier is not deployed, so "the agent lane is inert and every account is classified human" for now, so every seat is scored as a human's.
How you would know it works
The live leaderboard at https://app.greevils.ai/leaderboard read zero ranked competitors and zero agent submissions when fetched. The front page's leaderboard appears illustrative (inferred from the empty live app). No dated result exists yet, in those words.
What is missing
The chain identity is 4 of 7: no description, no Discord, nothing in the additional field. The contact atlas@greevils.ai matches a commit author and appears nowhere on the site. The incentive document is thorough but sits in a third repository. The CLI repository was last pushed 2026-07-06 and has no releases. There is no status page.
Go deeper
- Front door with plain paragraph: https://www.greevils.ai
- Incentive document with scoring rules: https://raw.githubusercontent.com/greevils-ai/greevils-validator/main/incentive.md
- Overview of the four repositories: https://raw.githubusercontent.com/greevils-ai/greevils-docs/main/README.md
Sources for this explainer
- https://www.greevils.ai
- https://app.greevils.ai/leaderboard
- https://raw.githubusercontent.com/greevils-ai/greevils-docs/main/README.md
- https://raw.githubusercontent.com/greevils-ai/greevils-cli/main/README.md
- https://raw.githubusercontent.com/greevils-ai/greevils-validator/main/README.md
- https://raw.githubusercontent.com/greevils-ai/greevils-validator/main/incentive.md
Metaphor: a trading firm's tryout desk with robots at the next seat. Every claim is drawn from the evidence set or the subnet's own materials; "(inferred)" marks a conclusion rather than a quote. Corrections.
How we scored it
Four audiences, four questions each, scored on what a first-time reader can find in five minutes. Method in the rubric.
- subnet_name
- greevils
- github_repo
- github.com
- subnet_contact
- atlas@greevils.ai
- subnet_url
- greevils.ai
- discord
- not set
- description
- not set
- additional
- not set
Stakers and validators●●●●● 2.8
Should I allocate here? · rank 23 of 32 for this audience
Q1What it is●●●●● 3
A Bittensor subnet where autonomous trading agents go head-to-head with human traders on live Hyperliquid markets.
output is trading skill, no unit, 5 bps builder fee only in the validator README
Q2Who it is for●●●●● 2
No competitors yet
with 0 ranked on the live leaderboard and 158 agent submissions mostly STALE or UNHEALTHY, while the front page shows a leaderboard with illustrative names and placeholder zeros
Q3How it resists gaming or fails●●●●● 4
named incentive document with eligibility, permanent disqualification, a builder fee pre-gate and a human cap, three clicks from the front page, validator README says the agent lane is inert today
Q4Identity and documentation●●●●● 2
identity 4 of 7, description, discord and additional not set, github and url resolve, contact matches the commit author only, site Discord is an Opentensor channel, no staker page
Miners●●●●● 3.5
Can I compete, and what wins? · rank 15 of 32 for this audience
Q1What it is●●●●● 4
Your agent is a directory that MUST contain entry.py
trades by HTTP calls to a loopback harness, package, submit, deploy, commit commands, an example-agent folder, harness API schema in another repo
Q2Who it is for●●●●● 3
Drop below the $1,000 floor and you're out
is the capital entry cost, the CLI README requires your own GCP Confidential Space VM, registration cost not stated (chain burn 0.1 TAO today)
Q3How it resists gaming or fails●●●●● 4
pairwise tournament scoring in prose with five factors, a constants table, elimination and permanent DQ rules, front page example: a 15 percent drawdown costs as much as failing to double your money
Q4Identity and documentation●●●●● 3
step by step CLI guide, repo pushed 2026-07-06, no releases or changelog, the CLI defaults to http://localhost:8000 for the backend, the live app shows most submissions STALE
Buyers and enterprises●●●●● 2.3
Can I use this today? · rank 18 of 32 for this audience
Q1What it is●●●●● 3
Trade Portfolio Vaults Leaderboard Mining Support
a live Hyperliquid trading terminal behind a wallet connect, what a buyer gets from the agents is not stated, no example
Q2Who it is for●●●●● 2
Every fill must be routed through the Greevils builder app
carrying about 5 bps, the only price and it sits in the validator repo, human traders and agent builders are the named participants
Q3How it resists gaming or fails●●●●● 2
no status page or SLA, the site footer offers Discord (an Opentensor channel), X and GitHub, the app footer has a Support link, no response expectation
Q4Identity and documentation●●●●● 2
no API reference for buyers, the harness API is for agents, the chain contact atlas@greevils.ai appears nowhere on the site
Newcomers●●●●● 3.0
What is this and why does it matter? · rank 22 of 32 for this audience
Q1What it is●●●●● 4
Agents against the humans. Real money on the line.
followed by a plain paragraph on the front page, TEE and on-chain unexplained, chain description not set so nothing to match
Q2Who it is for●●●●● 3
Verifiable enough to trust. Sealed enough that no one can reach in.
sealed execution, trade-only keys and on-chain settlement stated, no comparison with a centralized trading competition
Q3How it resists gaming or fails●●●●● 2
the live leaderboard reads no competitors yet, the front page leaderboard uses illustrative names and placeholder zeros, no dated results
Q4Identity and documentation●●●●● 3
subnet_name greevils matches the site, url resolves, description not set on chain, the front page sections serve as a learn page, the Documentation link goes to a GitHub README
How this score came to be. Verified means the scorer fetched the page and the words are on it; inferred means concluded from code, absence, or a third party. The link badge is a separate automated check made before publication.
Something wrong? Corrections of fact are applied as they arrive during the window; score disputes are batched at its close. How to file one · GitHub issue · email.