Findings

What the record shows.

Counts over the three datasets: the index, the programs, and the map. No reading and no opinion. Every number links to the rows it was counted from, and the rule for each count is written under it. When the data changes, the counts change with it.

15of 32 name no buyer
35places the sources disagree
7funded 10+ places above their legibility
2.7buyer legibility, mean of 5
  1. 1515 of 32 name no paying buyer today.the buyer field of each program, read from the subnet's own materials on 2026-09-19 and 2026-09-22; counted when it starts Nobody, No paying, or Unknown, or was not found
  2. 66 of the ten largest by emission name no buyer.the same buyer rule, over the ten subnets with the largest share of the block reward
  3. 3535 places where a subnet's own sources disagree about its program, across 25 subnets.fields where the docs, the code, and the chain did not say the same thing; each is quoted on the subnet page
  4. 77 sit 10 or more places lower on legibility than on emission.legibility rank minus emission rank, over the index; the table below carries every delta
  5. 2.7Buyers are the audience subnets explain least: 2.7 of 5 on average, against 3.5 for miners.mean legibility per audience across the index, four questions each, scale version 1.0
  6. 3.3The median legibility is 3.3; the range runs from 1.5 (Hone) to 4.3 (blockmachine).the composite over four audiences, 0 to 5, scored 2026-09-09
  7. 61%61% of program fields rest on the chain or the code; 4 could not be found.136 of 224 fields, seven per program, by the trust mark on each
  8. 44 of 32 have filled every field of their on-chain identity.the seven identity fields a subnet owner can set on chain, from the chain snapshot of 2026-09-09
  9. 32All 32 programs record something miners optimized instead of the work; 19 were read from the code.the exploits field of each program and its trust mark
  10. 414414 of 530 quotes on record are verified against a live primary source.the confidence mark on each statement in the map; the rest are probable or unverified and say so

The allocator table.

Every subnet in the index: its share of the block reward beside its legibility, the gap between the two ranks, how legible it is to a buyer, and whether its program names one. Click a column to sort. A red delta is ten or more places; a red no is a program whose buyer field starts Nobody, No paying, or Unknown, or was not found.

Rank Subnet Emissionshare of the block reward Legibilitycomposite, 0 to 5 Deltalegibility rank minus emission rank Buyerslegibility to a buyer Buyer namedthe program's buyer field Disagreefields where sources differ Trustseven fields, chain to not found
1MinosSN107 · science11.3%3.4rank 15 of 32+142.3rank 18no

Nobody yet.

own docs
24 of 7 from chain or code
2lium.ioSN51 · compute8.0%3.8rank 6 of 32+44.3rank 6yes

Renters paying per GPU-second.

own docs
15 of 7 from chain or code
3OROSN15 · agents6.7%3.8rank 6 of 32+32.5rank 17no

Nobody is named.

our reading
24 of 7 from chain or code
4AffineSN120 · training6.0%2.4rank 28 of 32+241.3rank 28no

Nobody is named.

our reading
14 of 7 from chain or code
5ChutesSN64 · inference5.9%4.1rank 2 of 32-34.8rank 1yes

Developers and resellers paying per million tokens or per GPU hour.

own docs
14 of 7 from chain or code
6AlbedoSN97 · training5.2%2.1rank 29 of 32+231.3rank 28no

Nobody is named.

our reading
5 of 7 from chain or code
7TargonSN4 · compute4.7%3.5rank 12 of 32+53.5rank 10yes

Renters of confidential GPU and CPU machines by the hour, through the site and an unauthenticated inventory API.

own docs
5 of 7 from chain or code
8ScoreSN44 · inference3.4%2.5rank 26 of 32+181.5rank 26no

Nobody is named.

our reading
15 of 7 from chain or code
9OpenRobotoSN80 · training3.0%3.3rank 17 of 32+82.0rank 20no

Nobody is named.

our reading
12 of 7 from chain or code
10SOMASN114 · inference2.4%3.4rank 15 of 32+53.5rank 10yes

Developers routing model calls through SOMA.

own docs
15 of 7 from chain or code
11TeutonicSN3 · training2.4%2.8rank 23 of 32+121.5rank 26no

Nobody.

our reading
15 of 7 from chain or code
12VantaSN8 · forecasting2.3%3.8rank 6 of 32-63.5rank 10yes

Two kinds.

own docs
35 of 7 from chain or code
13iotaSN9 · training2.3%2.5rank 26 of 32+131.8rank 23no

Nobody is named.

own docs
13 of 7 from chain or code
14engySN53 · inference2.1%4.0rank 4 of 32-104.5rank 2yes

Developers and agents paying per token through one OpenAI-compatible endpoint at api.engy.ai with prepaid billing.

own docs
22 of 7 from chain or code
15BitcastSN93 · media2.0%2.7rank 24 of 32+91.8rank 23yes

Brands, none named.

own docs
25 of 7 from chain or code
16NOVASN68 · science1.9%3.0rank 21 of 32+52.0rank 20no

No paying customer is named and there is no price.

own docs
15 of 7 from chain or code
17SayGMSN28 · inference1.7%4.1rank 2 of 32-154.5rank 2yes

Developers paying per million tokens through an OpenAI-compatible gateway at api.saygm.com/v1, signing up by email, Google or Bittensor with no card required.

own docs
5 of 7 from chain or code
18cascadeSN91 · forecasting1.7%2.0rank 30 of 32+120.5rank 30no

Nobody.

our reading
35 of 7 from chain or code
19Green ComputeSN110 · inference1.6%3.2rank 20 of 32+13.8rank 8yes

Enterprise teams paying in ordinary money, plus self-serve renters with a $5 minimum top-up.

own docs
2 of 7 from chain or code
20HippiusSN75 · storage1.6%4.0rank 4 of 32-164.5rank 2yes

Anyone with a credit card or TAO.

own docs
15 of 7 from chain or code
21greevilsSN58 · agents1.4%2.9rank 22 of 32+12.3rank 18no

Nobody.

our reading
12 of 7 from chain or code
22ChronoLLMSN38 · training1.4%3.3rank 17 of 32-53.0rank 15yes

Quants and researchers, on three tiers with no price.

own docs
15 of 7 from chain or code
23GradientsSN56 · training1.4%3.6rank 11 of 32-123.8rank 8yes

Anyone with a model and a dataset, by the hour, through the site or one API call to api.gradients.io/v1/tasks/create.

code
17 of 7 from chain or code
24DittoSN118 · agents1.2%3.7rank 10 of 32-143.5rank 10yes

End users of the Ditto app, paying per message in credits after 30 free messages, with no price table.

own docs
23 of 7 from chain or code
25SwarmSN124 · training1.2%3.3rank 17 of 32-82.0rank 20no

Nobody is named.

our reading
4 of 7 from chain or code
26RedTeamSN61 · security1.2%2.6rank 25 of 32-11.8rank 23no

Nobody today.

our reading
14 of 7 from chain or code
27CliqueAISN83 · other1.1%1.9rank 31 of 32+40.5rank 30no

Unknown.

not found
16 of 7 from chain or code
28HoneSN5 · data1.1%1.5rank 32 of 32+40.3rank 32no

Looked for a price, a customer, a product page or an endpoint on honedashboard.com, a miner-performance dashboard that renders Waiting for aggregate data to a fetcher, and in the README, DESIGN.md and DEMO_MINER.md at the pinned commit.

not found
5 of 7 from chain or code
29KubeTEESN90 · compute1.1%3.5rank 12 of 32-172.8rank 16yes

Not an enterprise yet.

own docs
14 of 7 from chain or code
30BitMindSN34 · media1.0%3.5rank 12 of 32-184.3rank 6yes

Developers and enterprises.

own docs
3 of 7 from chain or code
31RidgesSN62 · agents1.0%3.8rank 6 of 32-253.3rank 14yes

Developers with a GitHub repository, through Ridgeline: one credit turns one issue into one pull request from an isolated machine.

own docs
13 of 7 from chain or code
32blockmachineSN19 · other0.9%4.3rank 1 of 32-314.5rank 2yes

Apps that call blockchains, paying per request unit: as low as $1 per million requests at scale, with a calculator putting 100 million requests a month at $165 against Alchemy at $979, and failed requests not billed.

own docs
25 of 7 from chain or code

Emission rank is the subnet's share of the block reward in the chain snapshot. Legibility is the index: whether a first-time reader can find what a subnet is, who it is for, how it holds up, and whether its front door is open, in five minutes, from the subnet's own materials. The buyer column is the buyer field of the subnet's program, with the mark for what the field rests on. Neither instrument rates the mechanism or the team; a large delta says that a subnet is funded further ahead of its explanation than the others, and nothing more.

The counts and the table are published as findings.json, CC BY 4.0, alongside index.json, programs.json, and narrative.json. Cite: Findings, Legible, Mikyö Clark, legible.network, 2026. A row that is wrong is a correction; it changes the count when it lands.