Daily digest of the Bittensor ecosystem

中文/EN

Daily edition
Nov 20, 2026

Bittensor Daily A 1T1B product
1T1B ↗
BITTENSOR DAILY

24H Market

01/10

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

THE DAILY BRIEFING

Bittensor Daily — November 20, 2026

The day's substance ran through the subnet cards rather than the ecosystem channels: Green Compute's emission shutdown drew scrutiny of an owner-linked wallet, NIOME's scoring mechanism changed, and Carbon logged another round of intake outages. #General and #Rao stayed at their quietest, with no tokenomics or governance thread carried over from the week's earlier debates on registration pricing or key custody.

Today's highlights

  1. SN110 (Green Compute) has its emissions turned off, and community members question a dispersal wallet tied to the subnet's owner.

    (1) A commentator said the shutdown follows a pattern that had previously affected several other subnets before being reversed, this time with direct involvement from Const. (2) Community members noted that on-chain records since August show the dispersal wallet has paid only three wallets, all staking with the owner's validator, and asked whether other subnets with dispersal wallets face the same scrutiny. (3) The team posted a personal update thanking the community for its support while working through the issue, expressing confidence it would be resolved soon; no resolution had landed by day's end.

  2. On SN55 (NIOME), the team's buyback plans surface the same day a community investigation into seed-matching among top submissions remains unaddressed.

    @NiomeAI said the team behind NIOME has earned about $120,000 from clients, including a Scottish Government collaboration, and plans to grow buybacks. (1) This follows the seed-matching analysis reported on Nov 17, where rank #1 submissions matched round-specific seeds at an 84% rate across 25 rounds. (2) No further detail on an audit or official response to that allegation appears in today's material. (3) The buyback announcement stands as separate, positive news alongside the still-open scoring question.

  3. SN116 (Carbon) suffers another round of submission-intake outages, traced this time to an upload-handling bottleneck under high miner volume.

    (1) The team identified the cause as a code bottleneck handling upload chunks under high volume, with 94 miners uploading almost 3,000 chunks in 4 hours, and deployed a fix. (2) This continues the backlog first flagged on Nov 14 and partly addressed on Nov 16 with parallel grading; the team said validator restarts now take about 2 minutes versus up to an hour previously. (3) Separately, the team raised the per-hotkey graded submission allowance for Motor 1.8 to up to 10 per day effective Nov 19, with a further increase to 20 mentioned. The token fell 7.19% over 24 hours.

  4. Both #General and #Rao logged their quietest stretch of the week, with routine triage the only activity.

    My miner is online and running, but validators can't connect to my API. I've checked my configuration, but the issue persists. Is this a network connectivity issue or a Bittensor configuration problem? was answered only with It's not a Bittensor configuration issue. You should ask in the subnet channel. (1) #General logged 15 messages, close to Nov 18's low of 11, and #Rao had no substantive messages at all, down from 7 on Nov 18 and 50 on Nov 16. (2) #Announcements carried a single logistical notice about a Novelty Search call, the same series #General asked to reschedule, with no cross-reference between the two channels. (3) No tokenomics, governance or protocol-mechanism question surfaced in either ecosystem channel today.

Discord conversations worth following

#General

1 REPORT

Other

Other: a hacked-account recovery question, a zero-investment onboarding request, a testnet TAO ask and an unresolved miner-connectivity question filled a routine #General day

Opening ·

A member recovering a hacked account asked can you help me with this? and was told You would have to contact the mods on the DTAO server, with a further reply explaining You can join the server with a secondary account and make an appeal with the mods there to unban your main.

Onboarding ·

A newcomer asked I'm new to Bittensor and would like to explore the ecosystem without any initial investment. Could anyone recommend a subnet that supports the testnet and can run on a standard PC without a GPU? and received the standard caution against DMs before being told There are many subnets, and their requirements and supported environments are constantly changing, so we're unable to recommend specific subnets. I'd suggest exploring Taostats, TAO Marketcap, and Tao.app.

Support ·

A developer asked Hello, I am a developer and requesting testnet tao, where do I go and who do I ask? and was pointed to a pinned FAQ post rather than given a direct answer. Separately, a member reported My miner is online and running, but validators can't connect to my API. I've checked my configuration, but the issue persists. Is this a network connectivity issue or a Bittensor configuration problem? and was told It's not a Bittensor configuration issue. You should ask in the subnet channel.

Closing ·

A moderator repeated the channel's standing warning in full, including People claiming to be mods, admins, Opentensor Foundation members, subnet support, or any kind of "support" sending you a direct message or friend request -- they're imposters, block them!, and the day ended with a scheduling request for Novelty Search asking Would you please schedule NS at 21:00 UTC (instead of current 22:00 UTC)?

3 things to read from it

  1. Volume ticked up slightly to 15 messages from Nov 18's low of 11, but every item was routine triage -- account recovery, onboarding, support redirects and the standing scam script -- with nothing touching tokenomics, governance or protocol direction.
  2. The miner-connectivity question was answered only by ruling out a Bittensor-level cause and redirecting to the relevant subnet channel, leaving the actual issue undiagnosed in this channel.
  3. This continues the week's pattern of #General carrying almost no ecosystem substance, in contrast to earlier weeks when registration-cost and wallet-migration questions at least opened mechanism discussions.

Where it stands

All items closed through redirect, FAQ pointer, or the standing scam warning; none required or received further follow-up in the channel today.

#Rao

0 REPORTS

Nothing substantive today.

#Announcements

1 REPORT

Other

The only #Announcements message today was a last-minute Novelty Search call invite

Notice ·

The channel's single message today announced an upcoming call: We have a NS call this week which scales a 5 teams at once! with the five participating teams tagged and a closing Going to be great See you in 10 mins.

2 things to read from it

  1. The channel produced one logistical notice and nothing else, continuing the near-silent run #Announcements has shown through most of this stretch, broken only by a one-line cancellation on Nov 12.
  2. This is the same Novelty Search series #General asked about scheduling today, when a member requested the call be moved an hour earlier -- the two channels did not cross-reference each other.

In the week’s context

  • On Nov 12, #Announcements' only message was a one-line cancellation notice for the same Novelty Search series.

Where it stands

Closed; a routine scheduling announcement with no replies logged in the channel.

Subnets

5 REPORTS

SN99 24h +2.74%

Axonry

What happened
SN99 (Axonry) approved its first automated champion model, based on UID 250's pull request, reporting gains of roughly +10 to +25 on Math-500 and +6 on GSM8K from a 0.8B base model, with performance described as close to Qwen3.5-2B thinking. The team also pointed to a post from its official account framing Axonry as a reproduction of an MIT weight-editing technique that raised a small model's MATH-500 score from 7.9% to 48.1%, now run as an open competition on the subnet. Separately, validators briefly diverged on weights after the champion was set, and several miners reported submissions failing with EVAL_ERROR, including a write-once submission.json bug blocking re-evaluation and an OSError tied to a newly added evaluation machine.
Community
Miners discussed the correct order of commit, push, and PR submission to avoid being flagged as copying another miner's work, and one community member asked the team to ban a miner over a submission resembling another's PR. Miners also reported the EVAL_ERROR issues directly to the team, with some providing technical detail on the underlying causes; nearly 20 miners had made payment ahead of the upcoming competition.
Official
The team confirmed the validator weight divergence and said it was being addressed via direct message, announced the first champion model with performance figures, and linked a post from its official account describing the MIT-derived technique and the open competition on SN99.

SN44 24h −6.87%

Score

What happened
SN44 (Score) moderators removed several ground-truth annotation blocks across the Vehicle, Petrol Station, PPE, Fallen Person, and Room Dirtiness challenges after miners flagged errors. Enforcement action was confirmed against miners using clustered or similar hotkeys, OCR, or hidden databases, with repeat offenders blacklisted. A scoring change was announced so that aerial-duel events will be scored once per frame to address recurring edge cases in football annotation. The team also announced a one-month warning that the subnet will undergo major changes to become more ambitious, transparent, and open-source.
Community
Miners across several tracks reported what they viewed as ground-truth annotation mistakes, including missed objects in petrol, PPE, vehicle, and football challenges, and some raised concerns that certain miners were exploiting these GT gaps to boost scores. Discussion also covered confusion over how aerial-duel scoring should be applied to specific frames and why repeated annotation issues kept appearing in public elements.
Official
Moderators confirmed the removal of specific flagged blocks across multiple challenges and said an update to aerial-duel scoring would follow. The official account reported that vehicle detection accuracy rose from 90.2 to 92.7 and worksite PPE detection reached 91.8 over ten days, and separately announced a one-month warning before broader subnet changes.

In debate · Ground-truth annotation accuracy and fairness debate · 8 participants

  • A community member argued that some miners were exploiting ground-truth errors rather than improving their models to inflate scores.

    “Some miners aren’t actually trying to improve their models; they’re spotting errors in the ground‑truth images to artificially boost their average scores”
  • A miner defended reporting GT errors as a fairness measure, noting some reports helped other miners and some corrected their own mistakes.

    “I fully agree that evaluations should use correct GT, and that's the only reason I report blocks.”
  • A miner questioned how others achieved higher scores under the current aerial-duel ground truth despite apparent inconsistencies.

    “But some miners got high score for this challenge, how can they get higher score?”

SN110 24h −12.02%

Green Compute

What happened
SN110 (Green Compute) emissions were turned off, following a pattern a commentator said had previously affected several other subnets before being reversed, this time with direct involvement from Const. Community members raised concerns about a dispersal wallet tied to the subnet's owner, noting that on-chain records since August show it has paid only three wallets, all staking with the owner's validator, and questioned whether other subnets with dispersal wallets face the same scrutiny.
Community
Community members questioned how the centralization issue could be resolved given the subnet's ESG-related compute verification needs, with one member saying permissionless mining may not be compatible with the subnet's model. Others asked whether discussions with the team were ongoing and pushed for a resolution.
Official
The team posted a personal update thanking the community for its support while working through the issue, expressing confidence it would be resolved soon.

SN116 24h −7.19%

Carbon

What happened
SN116 (Carbon) experienced repeated submission-intake outages during the day, with the validator 'door' timing out and blocking new miner submissions for hours while grading continued running in the background. The team identified the cause as a code bottleneck handling upload chunks under high volume (94 miners uploading, almost 3,000 chunks in 4 hours) and deployed a fix. Separately, the team raised the per-hotkey graded submission allowance for Motor 1.8 to up to 10 per day effective Nov 19, later mentioning up to 20, while keeping the 1-per-epoch cap. The team also said it is developing a validation pipeline for a new 'custom algorithm' phase, inspired by Ridges, with details to follow before launch.
Community
Several miners reported persistent submission errors including construction_intake_unavailable, TRANSPORT_STALE, and idempotency conflicts, repeatedly asking the team to check direct messages and logs. Some miners expressed frustration at the system's instability, while others acknowledged the team's rapid iteration.
Official
The team confirmed the outages were caused by a single-file upload-handling bottleneck, stated submissions were healthy again after the fix, and said validator restarts now take about 2 minutes versus up to an hour previously.

In debate · Reliability of SN116's submission door during high-volume periods · 8 participants

  • A miner reported a stuck submission with multiple refusal types over several hours tied to a specific campaign and hotkey.

    “submissions from hotkey 5G7Nak8LbsfmTbp9ADb2kSEXizZRYnCzd6UczK5t5tztxp36 are stuck since 06:14 UTC on 19 Nov”
  • A miner called the system unstable and expressed frustration with recurring issues.

    “This system isn't stable. It's frustrating.”
  • The team explained the root cause as a code-level bottleneck in upload handling exposed by that day's volume.

    “our code accepted model uploads in a single-file line across the whole validator”

SN102 24h +0.44%

ConnitoAI

What happened
SN102 (ConnitoAI) changed prompt scoring from cycle 1400, now comparing generated courses against a reference course on accuracy, teaching approach, novelty, and clarity. The team also launched a public prompt dashboard showing per-cycle and per-miner scores with an API for programmatic access, and outlined plans for three additional upcoming tasks: earnings forecasting from news, reaction forecasting by demographic, and a search-agent prediction task. Separately, a validator update reportedly caused a weight/seed-hash issue that resulted in one UID receiving the entire emission in a recent cycle.
Community
Some members questioned the frequency of scoring rule changes, arguing it complicates dataset preparation and makes the competition feel too general; one suggested locking in a single goal before expanding and giving at least two cycles' notice before changes. Others raised concerns about one miner dominating emissions and asked for more transparency on miner rollouts, while another flagged the validator weight issue after the update, noting it had been reported before.
Official
A team member defended the current single-goal approach to course generation, said reference courses will not change until model quality approaches the reference, and confirmed the dashboard limits visible rollouts to the top 10% of models per prompt to manage database load.

More subnet community coverage every day in the Telegram channel →

Voices on X

  • @ridges_ai

    Ridges_ai announced Ridgeline, an AI agent fleet that turns GitHub issues into pull requests, with automated PR review and fixes and no local IDE required.

  • @TensiaFDN

    SN62 (Ridges) launched Ridgeline, where competing AI agents produce pull requests for approval; tasks cost 1 credit, reviews 0.1 credit, paid in TAO or by burning alpha.

  • @markjeffrey

    @markjeffrey reported a $2 million monthly buy-and-burn of SN51 (Lium) tokens.

  • @say_gm_

    @say_gm_ said agents can fund SayGM credits with Gamma minted on SN28, turning it into $1 of credit for Claude, GPT, Gemini, and open-weight models without a card.

  • @NATIXNetwork

    NATIX Network said growing data demand is driving talks for larger contracts, with top-tier Desktop App drivers earning USDC from December 1 and all drivers getting 4X points that month.

  • @404gen_

    @404gen_ said its latest competition produced over 20 Three.js-based games, showcasing varied environments, characters, and playable worlds.

  • @NiomeAI

    NiomeAI said the team behind NIOME has earned about $120,000 from clients, including a Scottish Government collaboration on rare neurological disease treatment, and plans to grow NIOME buybacks.

  • @conjectures_io

    @conjectures_io said TorchLean's winning kernel ran whole-model inference about 137× faster (117 µs vs 16.1 ms), from 226 submissions by 19 operators, with 221 passing the proof gate and 150 takeovers.

◆2026-11-20 · End of this edition

Protocol

4 PRs

New release · v492 Released

v492 tightens Coldkey swap protections, fixes Hotkey swap membership bugs and limits Operational priority to the sudo key

This release protects existing destination Coldkeys during Coldkey swaps, keeping their reward and lock-mode preferences intact. It fixes false Subnet membership entries after a full Hotkey swap and phantom lock conviction during Subnet dissolution. It also reserves the Operational transaction priority for the sudo key.

Read the upgrade report →

#3340 Breaking change

Make Yuma timelocked commit cap sudo-configurable

What
The maximum size allowed for timelocked commits used in Yuma consensus can now be changed by chain sudo rather than being fixed in the code.
Why
Gives chain governance the flexibility to adjust this limit as network needs change without requiring a full runtime code rewrite each time.
Who it affects
No direct effect on TAO holders; validators may see this cap adjusted by governance over time, but miners and ordinary holders are not affected.

#3343 Breaking change

Spend gamma from the deposit's hotkey

What
Gamma balances held in deposits are now spent by direct signature from the hotkey that owns the deposit, replacing the previous system where a separate assigned spender account made the payment.
Why
This closes a gap where a designated spender, rather than the actual owner, controlled the funds, tightening control over who can move gamma.
Who it affects
Validators and miners who use gamma deposits need to spend directly from the owning hotkey going forward; any previously configured spender accounts no longer have this authority, and existing data has been migrated accordingly.

#3337 Breaking change

Test alpha conservation for cross-owner childkey dividends

What
New regression tests were added to confirm that alpha balances remain conserved when dividends are paid out through child keys owned by a different account than the parent key.
Why
Protects against a bug class where cross-owner childkey payouts could create or destroy alpha incorrectly.
Who it affects
No visible effect for TAO holders, validators or miners; this is an internal safeguard for developers ahead of future releases.

#3342

Announce mainnet runtime and stable Python releases on Discord

What
Automated Discord announcements were added for three mainnet milestones: a runtime upgrade being proposed, that upgrade going live, and stable Python package releases reaching PyPI.
Why
Keeps the community informed of major mainnet events in a timely, consistent way without manual posting.
Who it affects
No effect on validators, miners or chain mechanics; TAO holders following the Discord channel will see earlier and more consistent notices of runtime upgrades and package releases.

Recent editions

Past editions +