Daily digest of the Bittensor ecosystem

中文/EN

Daily edition
Oct 20, 2026

Bittensor Daily A 1T1B product
1T1B ↗
BITTENSOR DAILY
Network upgrades·eTao·

24H Market

01/10

Gainers
--Loading…--
Losers
--Loading…--

THE DAILY BRIEFING

Bittensor Daily — October 20, 2026

A dispute over how emissions are distributed on SN82 (Noah) dominated community discussion this week, after public wallet and extrinsic data surfaced showing heavy concentration among a small number of addresses. Elsewhere, SN92 (REAX) tightened its scoring rules to close a loophole that let a near-identical model win on calibration alone, while SN55 (NIOME) and SN44 (Score) both confirmed and fixed validator-side problems affecting miners.

Today's highlights

  1. 01

    Community members reviewing SN82 (Noah) validator logs and taostats extrinsic links found that ten wallets held 97.89% of the subnet's counted stake while it received roughly 165-175 TAO per day in emissions, and some wallets appeared registered before the subnet's public launch.

    The findings raised questions in the Discord about whether emissions distribution on SN82 (Noah) reflects independent mining activity, with no response yet from the team.

  2. 02

    SN92 (REAX) activated Eval v3, adding a rule effective Oct 20 that requires challengers close to the champion's weights to also beat it on correctness, and confirmed its engine track will open Oct 20 with a reported 6x inference speedup over the reference server baseline.

    The change closes a gap where a model could win the crown on calibration alone, and the engine track opening adds a new, officially confirmed competition for miners.

  3. 03

    SN55 (NIOME) validators skipped submissions from all 256 uids as "not found" while still scoring roughly 100 uids tied to 8 coldkeys, which the team attributed to new compression logic and said it fixed.

    The bug affected how scores were recorded across the subnet before the team applied a correction and pushed the next competition phase to testnet.

  4. 04

    SN44 (Score) confirmed a miner exploited an archive-based tie-break flaw to retain top-miner status while running a copied model, and the team fixed the exploit and revised the logic to compare the last 10 common non-zero challenges.

    The confirmed fix changes how top-miner status is determined on the fallen-person detection track, addressing a method used to avoid validation challenges.

  5. 05

    SN1 (Apex) confirmed that ten submissions tied at 639/640 at the round 4 boundary of its NIST-Lite Assembly competition, saturating the current scoring rule, and said a harder v0.2 format with multi-peg sequential assembly is in development with no firm timeline.

    The saturation shows the current competition format has reached its scoring ceiling, prompting the team to redesign it rather than patch the existing rules.

Subnet Community Pulse

5

SN92 24h −11.02%

REAX

What happened
SN92 (REAX) activated Eval v3 for its DF and RA specialist tracks, adding private real-data items, a generalisation penalty and a tougher crowning threshold (2.75 SE over draws). A new rule takes effect Oct 20 18:00 UTC requiring challengers close to the champion's weights to also beat it on correctness, closing a gap where a near-identical model won on calibration alone. The REAX engine track is set to open Oct 20 18:00 UTC (fallback Oct 21), with the team reporting a 6x inference speedup over the reference server baseline. UID 67 passed its confirmation draw and became the new RA champion, pending copy check and approval.
Community
Miners questioned the fairness of the prior DF crowning, arguing the new champion was effectively a copy of the previous one with only calibration differences. Separately, a miner noted that current RA models appear to be derivatives or transforms of two base models, raising concern about originality, though they added this may not be an issue if the models perform better.
Official
The team confirmed Eval v3 went live at 18:08 UTC and that the engine track opens Oct 20 18:00 UTC with a fallback to Oct 21. It also corrected an earlier post implying the engine competition had already started, clarifying that submissions only begin counting once the track opens.

In debate · SN92 registration fees and multi-competition mining · 14 participants

  • The team argues the fee cannot be lowered or more entrants would crowd out miners doing good work.

    “We can't lower it, otherwise even more people would sign up and kick out other miners that do good work”
  • Some miners argue the fee is actually too low and should be raised sharply.

    “Reg fee is too cheap now, Kindly increase to 100 TAO”
  • Other miners want the ability to run one hotkey across multiple competitions simultaneously, as on SN44.

    “Please allow participation in multiple competitions simultaneously using a single hotkey.”

SN55 24h +24.25%

NIOME

What happened
Community members reviewing a validator's public Weights & Biases logs on SN55 (NIOME) found that submissions from all 256 uids were being skipped as "not found," while scores were still recorded for roughly 100 uids tied to 8 coldkeys. The team later confirmed validator-side submission and benchmarking problems, attributing them to new compression logic, and said a fix had been applied. The next competition phase went live on testnet, and validators were separately reported to have issued no tasks for about 15 hours. NIOME also announced a GENOME-to-NIOME token migration starting October 28, which it said could bring roughly 1,500-2,000 TAO into the ecosystem.
Community
Community members posted log excerpts questioning why scored uids matched those marked as having no submission, and one review raised concerns about whether duplicate-owner protections were properly enforced across multiple uids. Others reported submission uploads failing and asked why mainnet and testnet appeared to be down, with some waiting for the team's response.
Official
The team said it investigated the validator codebase after miner reports, identified and fixed submission and benchmarking issues caused by compression logic, and said v4 would be released on testnet before mainnet.

SN44 24h +1.51%

Score

What happened
A miner was found exploiting the tie-break logic on the fallen-person detection track, repeatedly avoiding certain validation challenges to retain top-miner status despite running a copied model. The team fixed the underlying archive exploit and updated the tie-break logic to compare the last 10 common non-zero challenges instead of simply the last 10 challenges overall. Separately, the team introduced a formal process for miners to request review and exclusion of soccer-track challenges demonstrated to be mislabeled against the annotation rules. Multiple miners also reported extended downtime on the Petrol station, Beverage detection, and soccer private track elements.
Community
Miners repeatedly reported the exploited hotkey and the extended outages on the Petrol station and Beverage detection elements, asking the team to investigate and, in some cases, blacklist the hotkey. A separate debate ran over soccer-track annotation consistency, with miners arguing that ground truth for events such as aerial duels was applied unevenly across clips.
Official
The team confirmed it fixed the archive exploit that allowed miners to selectively avoid challenges and revised the tie-break logic to compare common non-zero challenges. It also announced a review process for soccer-track challenges with demonstrably incorrect annotations, and said on X that fallen-person detection accuracy rose from a 43.6% baseline to 98% on crowded street footage.

In debate · SN44 soccer GT annotation inconsistency · 18 participants

  • Miners argued that ground-truth labeling for events like aerial duels was inconsistent across clips, undermining fair scoring.

    “aerial duel situation in one clip is annotated as "aerial duel" in GT, same situation is not annotated as "aerial duel" in other clip”
  • The team acknowledged recurring inconsistencies but said it relies on the external annotation provider to fix its process.

    “we need to contact eyeball so they straighten it out their different annotating groups”
  • A miner warned that persistent annotation inconsistency was turning the soccer track into a labeling contest rather than a model competition.

    “this kind of inconsistency will lead soccer element to wrong direction”

SN1 24h −0.98%

Apex

What happened
SN1 (Apex) moved production to v0.1.5 of its NIST-Lite Assembly competition at the round 4 boundary. In that round, ten submissions tied at 639/640, saturating the scoring under the current flat 1% improvement rule and leaving the top spot unable to be overtaken. The team said it is now developing a harder v0.2 NIST-Lite format with multi-peg sequential assembly and partial credit, with no firm timeline yet. Separately, miners solved a NIST robot-assembly peg-insertion challenge at 99.9% success within 15 minutes of it going live.
Community
Community members asked whether the saturated competition should be ended and replaced, and debated whether the 1% margin rule had worked as intended. Some asked for an accelerated v0.2 release, and others questioned the reliability of information shared by a team-linked bot account in the channel.
Official
The team confirmed the round 4 saturation at 639/640 across ten submissions, said v0.2 is in development with no firm ETA, and announced that all new competitions will use a 1-day reveal period instead of 4 days to release replays and submissions faster.

SN54 24h −12.02%

Yanez

What happened
SN54 (Yanez) officially started Phase 5, Cycle 1 execution, citing over 2 million face variations and 12,000 real screen-replay images collected during the prior cycle. The team also scheduled changes to a planned Pulse workshop for agent builders, moving it from Oct 21 to November, with a new date to be announced. Separately, the official account published commentary on the Financial Stability Board's Oct 15 report on AI in financial services, noting its emphasis on tracing whether an agent acted within its permissions.
Community
Several miners repeatedly asked when UAV rewards from the prior cycle would be paid, with one later reporting they had not received any rewards despite a decreasing score. One community member described the Yanez technology's potential use cases positively.
Official
The team explained that UAV rewards are released gradually and decay over time as validators update their code, and pointed miners to the dashboard to check their UAV score.

More subnet community coverage every day in the Telegram channel →

Ecosystem debate

62 participants

SN82 (Subnet 82) emissions dispute

Starting mid-October, community members in the Bittensor Discord raised concerns about SN82 (Subnet 82), alleging that owner-linked wallets were registered before the subnet's public launch and that insiders were receiving a large share of emissions rather than independent miners. Validator logs and taostats extrinsic links were shared as evidence, and d898 reported the subnet receiving 175 of 3600 tao per day (5% emissions) while, in the poster's view, producing no output. The discussion widened into a broader debate about whether such behavior should be tolerated or regulated across the ecosystem.

Positions

  • Some participants said owner-linked accounts registered wallets ahead of public participation, giving insiders an advantage before independent miners could compete.

    “owner / insider register with 3 wallets using 15t collateral registration fees”
  • Others argued the subnet's behavior, even if concentrated among insiders, falls within free-market norms and does not require intervention.

    “A free market cuts both ways.”
  • A number of community members called for a formal mechanism to deregister subnets accused of this behavior, appealing directly to Opentensor.

    “I want [mention] to implement a forced deregistration mechanism in Bittensor and deregister this subnet.”
Official response
No message from the SN82 team or Opentensor directly addressing these allegations appears in the thread. Community members later noted in d1036 that the subnet appeared to have switched off, and in d1067 that its emissions had been parked, but these are observations from participants, not a team statement.
What to watch
Whether SN82's emissions status remains changed and whether Opentensor responds to the community's request for a forced deregistration mechanism.

Voices on X

  • @taostats

    Taostats announced eTao live on-chain, with a default 18:41:41 alpha emission schedule that subnet owners can now adjust, plus a rules and simulator link.

  • @conjectures_io

    Conjectures said its miners found a solution to Erdős Problem 1057, proving C(x)=x^(1−o(1)) for Carmichael numbers, verified in Lean.

  • @SwarmSubnet

    SwarmSubnet gave a quarterly update, saying its thesis of open competition plus a physical lab has produced customers, with two contracts signed in two months.

  • @TensiaBot

    TensiaBot noted SN103 (Consted), SN70 (Consted 2.0), SN35 and SN36 (Epago) raised owner cuts to 33% within 12 minutes, questioning whether they share a team.

  • @webuildscore

    webuildscore said SN44 miners raised fallen-person detection accuracy from a 43.6% baseline to 98% on crowded street footage, prompting a higher difficulty target.

  • @TaoOutsider

    TaoOutsider reported that ten wallets held 97.89% of SN82 (Noah)'s counted stake, that it received roughly 165 TAO/day in emissions, and that mining rewards depend on financial contributions.

  • @almnc_ai

    almnc_ai said SN41 (Almanac) is shifting its emissions split so miners rise from 41% to 50% while the owner cut falls from 18% to 10%.

  • @reaxlabs

    reaxlabs said its REAX engine serves Perplexity Decider v1.1 up to six times faster than a reference server, offering it free to individuals and small companies.

◆2026-10-20 · End of this edition

Protocol

New release · v481 Released

Subtensor v481 lets each Subnet Owner choose how their subnet's Emission is split between owner, Validators and Miners, while Root stakers keep their baseline share.

Until now every subnet split its Emission the same way: 18% to the owner, 41% to Validators and 41% to Miners. With this release, each Subnet Owner can set their own split within fixed limits, at most once every 50,400 blocks (about 7 days). Each block, Root stakers are paid their baseline share of every subnet's Emission before anything else comes out of the Validator portion, so an owner's choice cannot raise or lower what Root receives. Existing subnets keep their current split until their owner changes it. The one exception is subnets that had the owner cut turned off, where Root's share drops to the standard baseline.

Read the upgrade report →

Recent editions

Past editions +