Subnet Legibility Index · snapshot 2026-09-09 · rubric v1.0 Exploit Summit 2026 · Montreal ·

SN 83. Rank 27 of 32 by emission

CliqueAI

1.1% of emission232 miners · 7 validatorsregistration 0.1 TAOverified 2026-09-09

CliqueAI - AI-Powered Maximum Clique Solver Network

In their own words · raw.githubusercontent.com
Explained from zero · other · written 2026-09-10

SN83 CliqueAI is a timed puzzle league: solvers get a network of a few hundred dots and a few seconds to find the largest group where every dot touches every other, and the league pays the best solvers.

The commodity, explained from zero

The puzzle is the maximum clique problem. Draw dots, connect some pairs with lines, and find the biggest set of dots in which every pair is connected. The site's headline calls this approximating NP algorithms, problems whose answers are quick to check but slow to find. No application of the problem is named anywhere in the subnet's materials.

So the commodity is a solution to a puzzle the validators themselves wrote. Validators keep a database of graphs with 290 to 700 dots and hand them out with time limits between 6 and 30 seconds. A solution "is a set of vertices". Nobody buys these solutions. For someone to pay, the subnet would need a door for a real graph, a scheduling conflict or a social network, and the materials describe none.

Why it is on Bittensor at all

The materials do not say. The mechanism document's stated aim is internal: "To empower the whole subnet, we consider both performance and diversity." No page says what decentralization changes for this puzzle, or compares the subnet to a solver run on one laptop. The honest answer from the materials is nothing yet. The comparison breaks here: a puzzle league has a sponsor who wants the puzzles solved, and this league's only paying party is the emission, the subnet's share of newly minted TAO.

How the work gets done

Miners (the solvers) run a program that receives a graph and returns a set of dots inside the time limit; a GPU is not required, and the stated minimum is 4 CPU cores and 16 GB of memory. Validators (the judges) check that the answer is a valid clique, then score it on optimality (size relative to the best answer that round), on difficulty (a factor set by graph size), and on diversity (a bonus for an answer no other miner returned). The final score is optimality times one plus difficulty, plus diversity, smoothed into a rating with a half-life of about 69 samples, which sets the weights that direct emission. The site copy says miners are chosen by stake while the mechanism document says each eligible miner is sampled with equal probability; the sources disagree.

How you would know it works

No such artifact exists on the subnet's own site. The MINER SCOREBOARD on the front page still reads "Loading" after five seconds in a browser. The docs link a Weights & Biases table of validator runs, which logs activity but shows no benchmark or comparison.

What is missing

There is no buyer, no price, no product and no support path. The chain contact CliqueAI@toptensor.ai appears nowhere on the site or in the repository, and toptensor.ai does not resolve. The Discord link is a channel in the shared Opentensor server. The site is a blank shell to anything but a browser, with no about page. The miner README is one command with no install steps; the last release, v0.0.13, is dated 2026-04-14, and the repository was last pushed 2026-07-21. The chain shows 232 active miners against 7 validators.

Go deeper

Sources for this explainer

Metaphor: a timed puzzle league with no sponsor. 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.

On-chain identity 6/7
subnet_name
CliqueAI
github_repo
github.com
subnet_contact
CliqueAI@toptensor.ai
subnet_url
cliqueai.toptensor.ai
discord
discord.com
description
CliqueAI - AI-Powered Maximum Clique Solver Network
additional
not set

Stakers and validators 2.3

Should I allocate here? · rank 27 of 32 for this audience

Q1What it is 3

CliqueAI - AI-Powered Maximum Clique Solver Network

and a site headline about approximating NP algorithms, output is clique solutions to operator graphs, no unit and no buyer, needs graph theory

Q2Who it is for 1

W&B Dashboard: https://wandb.ai/toptensor-ai/CliqueAI/table

logs validator runs, nothing on users, demand or revenue anywhere

Q3How it resists gaming or fails 3

mechanism doc one click from the README with optimality and diversity formulas and a debiased EMA, no failure modes, site copy says stake weighted selection while the doc says equal probability

Q4Identity and documentation 2
cliqueai.toptensor.aiverifiedlive_unconfirmed

identity 6 of 7, discord is a channel in the shared Opentensor server (absent), url is a JS shell, github resolves, contact CliqueAI@toptensor.ai not on the site, toptensor.ai dead, no staker page

Miners 2.8

Can I compete, and what wins? · rank 27 of 32 for this audience

Q1What it is 3

A solution is a set of vertices

for a general graph of 290 to 700 vertices under a 6 to 30 second limit, request and response shape only in the protocol code, no worked example

Q2Who it is for 2

required: False

for GPU with 4 cores and 16 GB RAM minimum, a YAML file not linked from the README, registration cost not stated (burn 0.1 TAO today), no competitiveness statement

Q3How it resists gaming or fails 4

final score is optimality times (1 plus difficulty) plus diversity in prose and math, EMA alpha 0.01 with a 69 sample half life, no example numbers, no immunity or deregistration policy

Q4Identity and documentation 2

one start_miner.sh command and no install steps, last release v0.0.13 dated 2026-04-14, repo pushed 2026-07-21, min_compute.yml says version 0.0.17

Buyers and enterprises 0.5

Can I use this today? · rank 30 of 32 for this audience

Q1What it is 1

Four-stage autonomous mechanism for distributed problem solving

cliqueai.toptensor.aiverifiedlive_unconfirmed

is the closest thing to a product statement, no API, no app, no way to submit a graph

Q2Who it is for 0
no sourceinferredunchecked

no pricing and no customer type in any own or third party material

Q3How it resists gaming or fails 0
no sourceinferredunchecked

no status page, no SLA, no support address, the only Discord link is a channel in the shared Opentensor server

Q4Identity and documentation 1

no API docs, chain contact CliqueAI@toptensor.ai appears nowhere on the site or in the repo, toptensor.ai does not resolve

Newcomers 2.0

What is this and why does it matter? · rank 29 of 32 for this audience

Q1What it is 2

To evaluate different strengths in solving the Maximum Clique Problem, we design challenges across varying scales, time constraints, and graph structures.

cliqueai.toptensor.aiverifiedlive_unconfirmed

JS only, jargon unexplained

Q2Who it is for 1

To empower the whole subnet, we consider both performance and diversity.

No statement of what decentralization changes, no comparison

Q3How it resists gaming or fails 2
cliqueai.toptensor.aiverifiedlive_unconfirmed

a MINER SCOREBOARD section still reads Loading after five seconds in a browser, a W&B table of validator runs is linked from the docs, no dated benchmark or demo

Q4Identity and documentation 3
cliqueai.toptensor.aiverifiedlive_unconfirmed

subnet_name CliqueAI matches the site, description is a title not a sentence, url resolves to a JS shell, no about page, discord is a channel in the shared server

Provenance

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.

scored by agent c 2026-09-09, rubric v1.0merged 2026-09-09links verified 2026-09-09: 24 live, 0 unreachable, 0 manual

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.