Daily digest of the Bittensor ecosystem

中文/EN

Daily edition
Nov 17, 2026

Bittensor Daily A 1T1B product
1T1B ↗
BITTENSOR DAILY

24H Market

01/10

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

THE DAILY BRIEFING

Bittensor Daily — November 17, 2026

Today's substance ran almost entirely through #Rao, where a member's question about a registration-cost "race" widened into a multi-hour argument over how much of the ecosystem's reported demand for TAO is circular rather than new capital, even as the underlying lock parameters changed twice while members were discussing them. #General stayed thin and routine, and the subnet cards carried the day's other concrete news -- Carbon's grading fix and Swarm's upcoming test-field update -- while Release 491 shipped gamma-emission controls and a new protected-account feature at the protocol level.

Today's highlights

  1. A #Rao thread argues that much of what gets reported as TAO demand is really emission moving in circles, reframing a registration-cost complaint into a demand-quality debate.

    I am curious about is the under 100 TAO registration cost basically just fueling a registration race? drew basically everyone is 99% fiat and where it is not, we have a few circular (and some self mine too) deals right now that overstate outside demand and the pointed framing I think it's problematic when you're claiming 10k of revenue generated by your subnet when 9.8k is from the other subnet and that 9.8k is coming from inflation. One participant ultimately shifted position: I'd rather see us do whatever we can to bring the amount of TAO we're printing in line with the amount of TAO demanded and Interesting ideas have not thus far figured out how to return value to TAO and it's been 3 years since Revolution. The only thing that has is boring old compute. (1) This goes further than Nov 9's #General thread, which only traced the registration-fee decline to a sudo-set parameter without learning who held authority over it -- today's #Rao conversation debates the policy rationale that parameter serves. (2) A rare instance in these channels of a community member revising a stated view mid-thread rather than restating it. (3) The thread ends without consensus and no official voice weighed in.

  2. The registration cooldown and reopen multiplier actually change mid-conversation in #Rao, partially matching a fix a member had proposed earlier the same thread.

    A member had proposed earlier that theres an open fix that drops the halving scaling so it reopens at 1.5x again, id rather see that ship than just raising the floor, and hours later another reported they just change the cooldown 20min ago from 2 to 1 and half day, with the full mechanism later explained as same decay rate, it just reopens at 1.25x the last lock now instead of 1x so it can actually climb again. (1) This is the first sign in either channel that the registration-price ratchet, described as one-way since Nov 9, can move back up at all. (2) It supplies the mechanism detail #General's Nov 14 thread on the shortened halving cadence never received, though again with no official confirmation. (3) A separate worry that faster cycling could crowd out slow-recovering subnets, raised with the example Almanac that really starting to shine 8-9 months after launching, went unaddressed by those describing the actual change.

  3. Release 491 ships gamma-emission controls for subnet owners and a protected-account feature with delayed-authority controls.

    (1) The release lets subnet owners set gamma emission shares and root-managed whitelisting, extending the Gamma mechanism introduced on Nov 10 one step further into protocol-level tooling. (2) Any account holder can now set up a protected account with delayed-authority controls over funds and stake, a feature relevant to the key-custody concerns raised in #Rao on Nov 14-15 following SN3 (Teutonic)'s coldkey compromise. (3) Ordinary TAO holders who do not use either feature are unaffected; a companion PR consolidating lock-aggregate bookkeeping per hotkey is purely internal.

  4. SN116 (Carbon) addresses the grading backlog flagged since Nov 14 by enabling parallel grading and buying additional hardware.

    (1) The team said the backlog stemmed from over 200 submissions being processed one at a time and responded by enabling simultaneous grades and compiled model builds. (2) It purchased two additional grading machines and moved model building to A100 GPUs to raise throughput, and traced a miner-reported grading interruption to an infrastructure fault on its own side rather than the miner's model. (3) A miner separately flagged an apparent inconsistency in the crowning margin threshold; the subnet's token fell 1.85% over 24 hours.

  5. SN80 (OpenRoboto) bridges its token to Solana while a miner's stuck submission remains unresolved, and routine #General traffic closes out a quiet day.

    (1) OpenRoboto's $ROBOTO token was bridged to Solana via machine.fun, building on its earlier collaboration with peaq, even as a miner's task submission stayed stuck in "submitted" status for over 9 hours with no resolution confirmed during the exchange. (2) On SN44 (Score), the team blacklisted hotkeys found forwarding football-track challenges to external inference APIs, and a separate task failed for all miners, scoring zero across the board. (3) #General logged hello, is https://pypi.org/project/bittensor/ 11.3.2 latest version right? with no firm answer, alongside the standing scam warning and routine redirects, continuing the week's pattern of low ecosystem substance in that channel.

Discord conversations worth following

#General

1 REPORT

Other

Other: a lost join-date/badges request, a hacked-looking X account question, a spam-report mix-up, an unanswered PyPI version question, and the standing scam warning filled a quiet #General day

Opening ·

The day opened with the recurring scam-imposter reminder, warning members about people claiming to be mods, admins, Opentensor Foundation members, subnet support, or any kind of "support" sending you a direct message or friend request and fake wallet sites.

Support ·

A member who had left and rejoined the server asked is there any way to get them back? about lost badges and join date, and was told I don't think we can give you back your join date (that's not something we can modify), with subnet roles reassignable via a separate channel.

Redirect ·

A member asked is this account hacked? about an X account, and was told No idea. You might want to ask in the appropriate subnet channel after noting they suddenly launched a token; another member's confused How can I spam report? was clarified into a genuine impersonation-DM report, resolved with instructions to use Developer Mode to find a reporting User ID.

Unanswered ·

A member asked is https://pypi.org/project/bittensor/ 11.3.2 latest version right? is it possible to announce new versions along with wheel hashes at the 'announce' channel? and received only Hi, I'm not sure, and am asking. : ), leaving the question open; the day closed with a greeting, Happy Tuesday everyone 🫡.

3 things to read from it

  1. Volume held at 16 messages, close to Nov 14-15's lows, with nothing in the channel touching tokenomics, governance or protocol direction.
  2. The unanswered PyPI/wheel-hash announcement request is notable given #Announcements logged zero messages today, meaning a concrete, answerable packaging-transparency question had no official channel to land in.
  3. The hacked-account and spam-report exchanges both resolved through redirect rather than substantive answer, continuing the channel's pattern this week of routine triage rather than ecosystem debate.

Where it stands

All items closed through standard redirect, moderator clarification, or an open-ended 'asking' response; the PyPI version/announcement question remained unresolved at day's end.

#Rao

2 REPORTS

Key

A #Rao thread traces the subnet registration scramble to lock math rather than price, then widens into a dispute over whether most of what gets called 'TAO demand' is real capital or inflation moving in circles

Opening ·

The thread opens with a member asking whether the sub-100 TAO registration cost is simply fueling a registration race where the real subnet builders can not even get a chance to register because they just can not out script the exploiters right now.

Mechanism ·

A member replies that the race is more a lock math thing than the low price itself, explaining that the reopening cost only ratchets down and that whoever lands the first block after cooldown wins, pure script race, proposing a fix: theres an open fix that drops the halving scaling so it reopens at 1.5x again, id rather see that ship than just raising the floor.

Debate ·

A third member argues high registration costs shut out unproven projects, asking if I want to do a research subnet, I need to put up 300k of capital? and suggesting I do think $50,000 USD would be fair though, saying they had complained about this to const even. Another member counters that cheap slots prove the price is too low rather than too high, and that we’re not really in a financial position to be funding research efforts for months or years with nothing returned to TAO holders, adding We can’t just keep slashing every source of TAO demand to bare minimum on the belief that doing anything else will hurt innovation.

Reframe ·

The first member reframes the whole premise of 'demand,' arguing most of it is circular: I'd maybe agree if the commodity subnets drove actual demand for people to buy and use tao or alpha, but basically everyone is 99% fiat and where it is not, we have a few circular (and some self mine too) deals right now that overstate outside demand, and flags the framing problem directly: I think it’s problematic when you’re claiming 10k of revenue generated by your subnet when 9.8k is from the other subnet and that 9.8k is coming from inflation. Then you make a twitter post saying you burned 10k or something (isn’t that just a wash?). The other member ultimately shifts position, saying I’d rather see us do whatever we can to bring the amount of TAO we’re printing in line with the amount of TAO demanded, noting Subnet registration covers like 3% of the total issuance, which is 1/4 of the total non-speculative demand for TAO, and concluding Interesting ideas have not thus far figured out how to return value to TAO and it’s been 3 years since Revolution. The only thing that has is boring old compute. So let’s lean into what’s working as hard as we possibly can.

4 things to read from it

  1. This is the first time #Rao has argued the registration-cost question as a demand-quality problem rather than a price-level one, distinguishing capital actually entering TAO from emission simply being recycled between subnets.
  2. The exchange reaches further than Nov 9's #General thread, which only identified the sudo-set parameter behind the fee decline without ever learning who held authority over it; today's #Rao conversation debates the policy rationale that parameter serves.
  3. One participant's reported shift in position -- from favoring cheap registration for research subnets to prioritizing inflation control -- is a rare instance of a community member revising a stated view mid-thread rather than restating it.
  4. The claim that much reported 'demand' is circular is made about the ecosystem in general, without naming a subnet, and stays unresolved: no team or moderator voice entered the debate.

In the week’s context

  • Nov 9's #General thread first identified the registration-fee decline and the sudo-set parameter behind it, but could not learn who held authority over the change.
  • Nov 14's #General thread confirmed the registration-cost halving cadence had been shortened from 4 days to 2 but got no official explanation of intent -- the same unanswered-rationale pattern resurfaces here, now debated among community members themselves rather than left open.

Where it stands

Unresolved; the thread ends with one participant's changed view toward prioritizing inflation control, but no agreement on registration pricing policy and no official voice weighed in at any point.

Notable

A #Rao thread reports the registration cooldown and reopen multiplier actually changing mid-conversation, partially matching a fix proposed earlier the same day

Proposal ·

A member proposes removing the two-day cooldown entirely, writing I think the 2 days cooldown should be removed like 3 hotkey register per epoch was removed. We can control the price and duration with other parameters., and another adds it's time to remove subnets cap or increase it, at least.

Pushback ·

A member worries a faster cycle could let someone register 10 subnets in 2 weeks for 30-50 tao each, could quickly kill good subnets that are trying hard to come back, citing a subnet that really starting to shine 8-9 months after launching.

Clarification ·

A member replies that the cooldown, not the reopen cost, is the actual driver, writing cant really, every reg relocks at 2x the last lock so its only cheap for whoever goes first after a quiet stretch and the cooldown is the real issue imo, drop it and let the doubling do the work. Another member then reports a live change: they just change the cooldown 20min ago from 2 to 1 and half day.

Outcome ·

Asked whether the change affects the price-drop rate or just lets prices climb back, the explanation is given: same decay rate, it just reopens at 1.25x the last lock now instead of 1x so it can actually climb again.

3 things to read from it

  1. The shipped reopen multiplier of 1.25x is a partial version of the 1.5x fix a member had proposed earlier in the same thread, suggesting the channel's own discussion tracked a real change rather than just reacting to it after the fact.
  2. This gives #Rao, for the first time, the concrete mechanism detail that #General's Nov 14 thread on the halving-cadence change never got -- today's exchange explains exactly what moved and how, even though no official voice confirmed it.
  3. The worry that faster cycling could crowd out slow-recovering subnets, raised with a specific example, was not addressed by anyone describing the actual parameter change, leaving that concern standing separately from the mechanism clarification.

In the week’s context

  • Nov 9's #General thread first surfaced the registration-price ratchet as a one-way decline; today's #Rao exchange shows the reopen multiplier being adjusted upward, the first sign in either channel that the ratchet can move back up at all.
  • Nov 14's #General thread confirmed the halving cadence had shortened from 4 days to 2 but got no explanation of rationale; today's live parameter change in #Rao supplies the missing mechanism detail without any official acknowledgment either.

Where it stands

Partially resolved as a mechanism description -- the cooldown and reopen multiplier changes were both confirmed by members in real time -- but no official source named or explained the change, and the dereg-risk concern raised earlier in the thread went unaddressed.

Subnets

5 REPORTS

SN80 24h +9.91%

OpenRoboto

What happened
SN80 (OpenRoboto)'s existing token, $ROBOTO, was bridged to Solana via forevermoney_ai and made tradeable on machine.fun, building on a prior collaboration with peaq on onchain identities for robotics data collectors. The Solana official account also posted about the move. Separately, a miner's task submission remained stuck in 'submitted' status for over 9 hours while later submissions were admitted within minutes, and a team member escalated the issue to developers without resolution during the exchange.
Community
A miner flagged the stuck submission repeatedly over several days, expressing frustration at being the only one affected while others proceeded normally. Another community member noted the Solana shoutout, and machine.fun's account clarified the bridged token is not new but SN80's existing Bittensor token.
Official
The team acknowledged the stuck submission, said devs were still looking into the logs, and asked the miner to hold off resubmitting to avoid an unnecessary fee. OpenRoboto's official account announced the Solana bridge via machine.fun and reiterated it builds on the peaq collaboration.

SN124 24h −3.96%

Swarm

What happened
SN124 (Swarm) is preparing to deploy Update V6.0.1 on November 17. The release adds a new task called 'Swarm Test Field,' where drones take off from a dock on a real test field in Andorra, locate people in the field, report their positions, and land within set altitude and boundary limits. The team stated this task is public, not tied to a specific client, and is intended for the subnet's own field tests. The update also includes associated security fixes.
Community
Several miners described the current top score as extremely difficult to beat, with some saying they were giving up on improving further. One miner asked whether the subnet owners could make the next epoch more challenging to give others a better chance against the leading score. A miner also asked whether the update affects the Sentinel task family.
Official
The team confirmed the new Swarm Test Field task is public and used for the subnet's own field tests, and clarified that the update does not change anything in the Sentinel task family.

SN44 24h +0.83%

Score

What happened
The team blacklisted several hotkeys on the football private track after finding they forwarded challenges to external inference APIs. A separate football private task at block 9,520,800 failed for all miners, resulting in zero scores across the board. The team also removed disputed ground-truth-labeled blocks on the Petrol Station and Vehicle challenges after miners flagged annotation errors.
Community
Miners repeatedly flagged what they considered incorrect ground-truth annotations across petrol, vehicle, football, and PPE tasks, asking the team to review specific blocks. One community member asked about the reason for a miner being blocked on the football track and raised concerns about fairness in top-performing models.
Official
The team confirmed it blacklisted three hotkeys on the football private track for forwarding requests to external APIs, stated the all-zero scoring task was "no problem for everybody," and listed the specific Petrol Station and Vehicle blocks it removed after review.

In debate · Disputes over ground-truth annotation accuracy on football and other challenges · 7 participants

  • A miner argued the ground truth on a petrol station block was missing an object.

    “plz check this block 9518400 in petrol track. GT missed this roof canopy. plz check it. thx for reviewing it.”
  • A validator defended the annotation on a football aerial-duel scenario as correct.

    “Two player contesting a ball in the air while one is clearly jumping i dont understand your confusion”
  • Another miner disagreed with the validator's reading of the same football scenario.

    “I can't see that anyone is jumping. And those two players are not contesting a ball. Only one player is receiving a ball, I think.”

SN116 24h −1.85%

Carbon

What happened
SN116 (Carbon) rolled out Motor 1.8 grading improvements, enabling two simultaneous grades and compiled model builds. The team said the change addressed a backlog of over 200 submissions, and it purchased two additional grading machines plus moved model building to A100 GPUs to increase throughput. A miner reported a grading interruption labeled 'FAILED_INFRA', which the team traced to an infrastructure fault on its own side rather than the miner's model. Separately, a miner questioned an apparent inconsistency in the crowning margin threshold used to determine new kings.
Community
Miners repeatedly flagged slow evaluation speed and long queue times, with one saying the delays had a bad effect on their operations, while others thanked the team once speed later improved. A miner also raised suspicion about inconsistent crowning thresholds and asked the team for an explanation.
Official
The team said the queue stemmed from over 200 submissions being processed one at a time, announced parallel grading and builds, and detailed the purchase of two additional grading machines and a move of model building to A100 GPUs. The team also confirmed a grading interruption was caused by an infrastructure fault on its side, not a problem with the miner's model.

In debate · SN116 evaluation queue delays · 6 participants

  • Miners said evaluation speed was too slow and it was hurting their operations.

    “Hey, Eval speed is too slow. Plz speed up to encourage our miners”
  • The team explained the delay was due to queue volume and described the capacity upgrades underway.

    “The wait is the queue, we've had over 200 submissions, and until tonight one grader handled them one at a time.”
  • Some miners acknowledged a later improvement in evaluation speed.

    “Evaluation speed has improved dramatically!”

SN54 24h +1.07%

Yanez

What happened
The team is alpha-testing BioVault, a password manager that replaces a master password with face-scan approval through YID; the team says early testers asked for a browser plugin and the ability to import passwords from other managers. Separately, a co-founder presented at the MOVE Conference 2026 in Arlington, Virginia on November 17 on AI agent identity verification. In the subnet channel, a miner questioned the scoring methodology, specifically why UID 186 showed low "final rewards" while holding a high predicted score.
Community
Several miners said their voice and screen-replay submissions had gone ungraded for extended periods and asked for faster grading. One miner repeatedly pressed the team on the scoring discrepancy for UID 186, while others raised general concerns about transparency and the use of AI-generated replies from the team.
Official
A team member explained that the final-reward figure reflects ranking among miners on a single randomly graded sample, while the prediction score is driven by consistency in face-variation scores across multiple samples over time, and said this is not abnormal behavior; the team also linked its scoring documentation on GitHub.

More subnet community coverage every day in the Telegram channel →

Ecosystem debate

5 participants

Subnet registration cost and the 'registration race'

Participants discussed whether the current subnet registration price and lock mechanism create a speed-based race that favors well-resourced or script-using registrants over new or research-oriented teams. The debate expanded into whether registration fees serve as meaningful non-speculative demand for TAO, and whether raising costs would exclude fresh or research subnets in favor of funded or speculative ones.

Positions

  • Some participants describe the registration dynamic as a pure speed contest, since the lock price reopens near the last paid price after cooldown, meaning whoever registers first after a quiet period secures the slot.

    “so whoever lands the first block after cooldown wins, pure script race”
  • Others argue the price itself is not the core issue and that a high capital requirement would shut out research or early-stage projects that cannot compete with funded or speculative entrants.

    “if I want to do a research subnet, I need to put up 300k of capital?”
  • A third view holds that subnet registration already represents a meaningful share of non-speculative TAO demand, and that this source of demand should not be reduced further given limited other demand drivers.

    “Subnet registration covers like 3% of the total issuance, which is 1/4 of the total non-speculative demand for TAO”
Official response
There is no explicit official statement from Opentensor addressing the registration-race concerns directly. Protocol-level parameter changes were made during the discussion period -- the registration cooldown was reduced from 2 days to 1.5 days, and the price reopen multiplier was changed from roughly 1x to 1.25x the last lock -- which some participants linked to the issues raised, though no formal explanation accompanied these changes.
What to watch
Whether further adjustments to the cooldown period or reopen multiplier follow, and whether they visibly change the pace or cost pattern of new subnet registrations.

Voices on X

  • @taodaily_io

    @taodaily_io reports SN53 (Engy) processed over 3 trillion AI inference tokens in 30 days, more than Venice's 2.56 trillion, while valued 28x lower and ranking second on KDP's leaderboard.

  • @taodaily_io

    @taodaily_io reports a compromised SN3 owner key was used on Nov 11 to transfer 101,141 alpha, sold roughly eight minutes later for about 2,068 TAO.

  • @lium_io

    @lium_io says its hardware supply has nearly doubled in a month, now hosting about $50m in hardware including 481 B300s across more than 25 GPU types.

  • @openroboto

    @openroboto announces $ROBOTO will list on @machinedotfun with a Solana trading pair, building on its collaboration with @peaq for onchain identities for its robotics data collectors.

  • @DFVTAO

    @DFVTAO highlights SN53 (Engy) at 3.05T tokens and 14.7% of reported volume, second on the list, versus Venice's $981M market cap against Engy's $34.7M.

  • @xeveryframe

    @xeveryframe's first SN117 (EveryFrame) monthly report cites 46,733 jobs run since its Sept 28 launch, 43,399 verified, 671 creator accounts, and 94.3% delivery reliability over 30 days.

  • @ChipForge

    @ChipForge says a miner's 28nm pre-silicon design for SN108 (ChipForge) reached 16,912 images per second on EfficientNet-Lite0, versus 1,953 for the Hailo-8 chip, about two weeks into Quest 2.

  • @TaoOutsider

    @TaoOutsider notes TAO closed the Nov 9-15 week at 195 USDT on Binance Spot, down 10.06%, with TAO trading volume up 87.87% from the prior week.

◆2026-11-17 · End of this edition

Protocol

2 PRs

#3327 Breaking change Affects emission

Release 491

What
This release advances the Subtensor runtime version from 490 to 491 and bundles in the gamma-emission SDK/btcli support and a new protected-account feature for the proxy system.
Why
The upgrade packages a set of previously built features into a coordinated runtime and tooling release so they ship together and stay in sync with the node version.
Who it affects
Subnet owners gain the ability to set gamma emission shares and root-managed whitelisting, and any account holder can now set up a protected account with delayed-authority controls for funds and stake; ordinary TAO holders who do not use these features are not affected.

#3331 Breaking change

Keep one lock aggregate per hotkey

What
This PR consolidates each hotkey's stake-lock bookkeeping into a single aggregate, removing the previous separate owner-lock tracking and adding an on-chain migration to rebuild the combined figure.
Why
Maintaining two separate lock totals per hotkey was redundant and the team simplified it to one source of truth.
Who it affects
This is an internal accounting change with no visible effect for validators, miners or TAO holders; it mainly concerns how the chain stores lock data internally.

Recent editions

Past editions +