Validators lease a public coding challenge, rotation-deal it to half of the serving miners, commit the signed responses, reveal the hidden tests, grade every response in a local Docker sandbox, and submit the resulting weights on chain.
SN 5. Rank 28 of 32 by emission
Hone
Emission1.1%of the block reward · rank 28 of 32
Participants243 miners · 9 validators
Registration0.1388 TAO
Verified2026-09-09links checked and scores merged
The programdata · github.com@34de7f6 · read 2026-09-19
The work
A program that solves a coding problem. A validator posts a signed task to the miner's HTTP endpoint and the miner returns a Python or Rust solution before the deadline. The included demo miner forwards the task to a hosted GLM-5.2 model. What is paid for is a solution that passes every hidden test case, which the miner never sees.
own docs · raw.githubusercontent.com
Validators lease a public coding challenge, rotation-deal it to half of the serving miners, commit the signed responses, reveal the hidden tests
How it is scored
All or nothing, then speed. The validator commits the exact signed answers to the problem server before it may see the hidden cases, then runs each answer in a one-CPU, 256 MiB Docker container with no network. The fastest fully correct answer is paid 1.0; every three minutes of extra latency halves the part above 0.95. Anything short of the full suite pays 0. Peer pass rates do not matter.
code · raw.githubusercontent.com · rlvr/scoring/payment.py
DEFAULT_SPEED_HALF_LIFE_MS = 180_000.0; DEFAULT_SPEED_FLOOR = 0.95
How the pay splits
Proportional. A miner's score is the mean of its last 200 completed-problem payments, with a floor divisor of 4 so a couple of early wins cannot show as a full score. Validators normalize those means to sum to one and submit them as weights with no owner reservation; the policy refuses a non-zero owner burn share. A miner absent when sampled records a zero. History is keyed by hotkey.
code · raw.githubusercontent.com · rlvr/scoring/eval_engine.py
Count-bounded score histories -> L1-normalized weights.
Who judges
Nine validators were active on the 2026-09-09 chain snapshot. Commit-reveal is off. Each validator is a whole examiner: it leases problems from a private problem server outside the repository, deals each to half of the serving miners in rotation, commits the signed replies, receives the hidden cases and grades every reply itself in a sandbox. The server can pace validators with Retry-After and gates the reveal.
chain · hyperparams-2026-09-19.json
commit_reveal_weights_enabled False, weights_rate_limit 100, weights_version 803; chain-2026-09-09.json active_validators 9
How often it turns
Emission settles every 360 blocks, about 72 minutes. A validator attempts a challenge every 38 blocks, about 7.5 minutes, and submits weights every 180 blocks, about 36 minutes, never below the chain rate limit plus 20. A miner is judged on a rolling window of its last 200 problems. A new registration is immune from deregistration for 5,000 blocks, about 16 hours.
chain · hyperparams-2026-09-19.json
tempo 360, immunity_period 5000; policy.py round_interval_blocks: int = 38, weights_interval_blocks: int = 180
Who buys
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. None found. The README says problem construction is not part of this repository; the only hint of an output beyond weights is that partial-credit rewards affect exported dataset labels.
not found
What has been gamed
Leaking hidden tests into a prompt: the signed answer set is committed before the cases are revealed. Replaying a captured request: an 8-second UUID cache. Inheriting a good history: scores are keyed by hotkey and cleared on change. Copying another miner's solution: not addressed; the code says AST fingerprints were rejected because they reward irrelevant structural changes, so V1 has no duplication scoring. Timeouts and invalid replies score zero.
code · raw.githubusercontent.com · rlvr/scoring/payment.py
TODO(V2): Revisit duplication-resistant scoring only with a signal that cannot be cheaply gamed.
5 of 7 fields rest on the chain or the code. A manifest is the site's reading of the subnet's program, not a review; each field links its source. All programs · What a field means · Correct a field
What it says
The subnet in its own words: the line the index took from its front door, and what the team has said on the record.
What it shows
Explained from zero by the index, from the subnet's own materials, for someone who has never heard of Bittensor.
SN5 Hone is a programming exam with sealed answers, where an exam board mails out a coding problem, the examiners lock in every candidate's answer before the hidden test cases are revealed, and full marks are the only marks.
The commodity, explained from zero
What the subnet produces is a record of coding problems, the programs written to solve them, and a verified pass or fail for each. The README says it holds "the validator for RLVR on Bittensor Finney, NETUID 5" and leaves RLVR unexpanded. Problems arrive from "a private problem server"; the README says plainly that "Problem construction is not part of this repository". The design document mentions "exported dataset labels", which is the nearest the materials come to naming an output (inferred from docs/DESIGN.md).
Why such a record would have a price, and who would pay, is not stated. There is no product page, no price, no customer and no contact address. The chain description is the two-word tag "Hone training". Here the exam comparison breaks: a real exam board exists to certify students for someone, while the materials do not say who reads Hone's graded papers or what they are for.
Why it is on Bittensor at all
The materials do not say what decentralization changes here; on that question the honest answer is nothing yet, or nothing stated. What the design document does describe is a trust model: each validator "accepts no remote grading claim" and runs every submitted program in its own sandbox, so neither the problem server nor any miner can set a score.
How the work gets done
A validator, the examiner, leases a public coding problem from the problem server and sends it, signed, to half of the miners currently serving. Miners, the candidates, return a Python or Rust program; the included demo miner simply forwards the problem to a hosted GLM-5.2 model. The validator commits the exact set of signed answers to the server, and only then receives the hidden test cases, so no test can leak into a prompt. It runs every answer in a locked-down container with no network access, and a program that passes the whole hidden suite scores between 0.95 and 1.0 with speed as the tiebreaker; anything else scores zero. Each miner's last 200 results feed the weights that split emission, the subnet's share of newly minted TAO, and the owner's cut is set to 0%.
How you would know it works
The only artifact is the dashboard at honedashboard.com, titled "Hone Network · Miner Performance", which promises median accuracy, response rate and latency per validator. To a fetcher it reads "Waiting for aggregate data" and "0 registrations"; a reader needs a browser. There is no dated benchmark, sample output or usage figure.
What is missing
There is no miner guide, only a demo that calls a paid third-party model and defaults to the test network, plus five sample challenges. No hardware or competitiveness statement exists for a miner that wants to win; the registration cost is not stated (about 0.14 TAO on chain on 2026-09-09, with 243 active miners). The chain contact and discord fields are empty. The README calls the project "RLVR subnet" while the chain and dashboard say Hone. There is no plain-language paragraph anywhere, and the design document marks copy-resistant scoring as a TODO.
Go deeper
Sources for this explainer
- https://honedashboard.com
- https://raw.githubusercontent.com/hone-subnet-org/hone-subnet/main/README.md
- https://raw.githubusercontent.com/hone-subnet-org/hone-subnet/main/docs/DESIGN.md
- https://raw.githubusercontent.com/hone-subnet-org/hone-subnet/main/docs/DEMO_MINER.md
- https://raw.githubusercontent.com/hone-subnet-org/hone-subnet/main/examples/sample_challen
Metaphor: a programming exam with sealed answers. Every claim is drawn from the evidence set or the subnet's own materials; "(inferred)" marks a conclusion rather than a quote. Corrections.
How legible it is
Four audiences, four questions each, rated on what a first-time reader can find in five minutes. The scale measures findability, not quality.
On-chain identity 4/7- subnet_name
- Hone
- github_repo
- github.com
- subnet_contact
- not set
- subnet_url
- honedashboard.com
- discord
- not set
- description
- Hone training
- additional
- not set
Stakers and validators●●●●● 1.8
Should I allocate here? · rank 32 of 32 for this audience
Q1What it is●●●●● 2
This repository contains the validator for RLVR on Bittensor Finney, NETUID 5.
output is graded coding solutions, inferable from the protocol diagram, no unit or buyer, chain says Hone training
Q2Who it is for●●●●● 0
no users, customers, revenue or usage anywhere in own materials, the dashboard shows miner accuracy only and renders Waiting for aggregate data to a fetcher, third-party pages describe an older Hone
Q3How it resists gaming or fails●●●●● 3
commit-before-reveal, sandboxed grading, zero for invalid or timed-out responses, hotkey-keyed history, a private problem server as a dependency, duplication-resistant scoring marked TODO
Q4Identity and documentation●●●●● 2
identity 4 of 7 (contact, discord, additional not set), github and url resolve, the README calls the project RLVR subnet while the chain and dashboard say Hone, no staker page, no about page
Miners●●●●● 2.5
Can I compete, and what wins? · rank 29 of 32 for this audience
Q1What it is●●●●● 3
accepts signed POST /solve requests from authorized validators
coding task in, a Python or Rust solution out, five sample challenges in examples, no miner repo, demo miner calls a hosted GLM-5.2
Q2Who it is for●●●●● 2
a Z.ai API key with access to glm-5.2
and a public port are the demo requirements, no hardware statement for a competitive miner, registration cost not stated (chain burn 0.14 TAO), 243 miners
Q3How it resists gaming or fails●●●●● 3
Reward and weights: a fully correct submission scores 0.95 to 1.0 with speed as tiebreaker, anything else zero, rolling window of 200 observations, history cleared on hotkey change, no immunity
Q4Identity and documentation●●●●● 2
no miner guide beyond the GLM demo miner and sample challenges, the README is a validator runbook, repo pushed 2026-09-05, no releases, the demo miner defaults to the test network
Buyers and enterprises●●●●● 0.3
Can I use this today? · rank 32 of 32 for this audience
Q1What it is●●●●● 1
nothing to buy or call, the README says problem construction is not part of this repository and names a private problem server, dataset labels mentioned in passing, no product statement, no date
Q2Who it is for●●●●● 0
no sourcesecondhandunchecked
no pricing and no customer type anywhere
Q3How it resists gaming or fails●●●●● 0
no sourcesecondhandunchecked
no status page, SLA, support address or support channel, discord and contact not set on chain
Q4Identity and documentation●●●●● 0
no API docs, no contact on the site or in the repo, subnet_contact not set
Newcomers●●●●● 1.5
What is this and why does it matter? · rank 32 of 32 for this audience
Q1What it is●●●●● 2
Validators lease a public coding challenge, rotation-deal it to half of the serving miners
technical README opening, chain description Hone training is a two word tag, no plain paragraph anywhere
Q2Who it is for●●●●● 0
no sourcesecondhandunchecked
no statement of what decentralization changes in own materials, third-party explainers describe an older pretraining design and were not fetched
Q3How it resists gaming or fails●●●●● 2
live dashboard of miner accuracy and latency exists but renders Loading and Waiting for aggregate data to a fetcher, no dated benchmark or sample output
Q4Identity and documentation●●●●● 2
subnet_name Hone matches the dashboard title Hone Network, description Hone training is not a sentence, url resolves, no about or learn page, the README names the project RLVR subnet
Provenance
How this rating came to be. Verified means the scorer fetched the page and the words are on it; secondhand means concluded from code, absence, or a third party. The link badge is a separate automated check made before publication.
scored by agent d 2026-09-09, rubric v1.0merged 2026-09-09links verified 2026-09-09: 23 live, 0 unreachable, 0 manual
Something wrong? Corrections of fact are applied as they arrive during the window; rating disputes are batched at its close. How to file one · GitHub issue · email.