Daily digest of the Bittensor ecosystem

中文/EN

Daily edition
Nov 15, 2026

Bittensor Daily A 1T1B product
1T1B ↗
BITTENSOR DAILY

24H Market

01/10

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

THE DAILY BRIEFING

Bittensor Daily — November 15, 2026

Today's substance ran almost entirely through #Rao, where a wallet-security conversation moved from spend-limit workarounds to a broader argument over whether airgapping alone can prevent the kind of compromise the ecosystem saw this week. #General stayed thin, repeating an unresolved registration-cost question from earlier in the week, while the subnet cards carried the day's other concrete news: Carbon's mainnet launch backlog and a conviction-accounting bug that briefly flipped SN24 (ms.computer)'s ownership. Neither channel returned to the root-basket freeze or the owner-accountability debate that dominated prior days.

Today's highlights

  1. A #Rao thread debates whether airgapped signing devices actually stop key theft, running directly off SN3 (Teutonic)'s coldkey compromise from Nov 11.

    people should once for all understand that they sohuld invest in a separate device, totally airgapped just for singning stuff, including their development commits, everything that requires signing shuold be done on a separate device drew the challenge airgap only saves you if the device actually shows what youre signing tho and kinda worried thats where the ledger flow for hashed accounts is going, device only displays a digest. The exchange ended with so id want both, airgapped and the device decoding what it signs. (1) This is the first extended #Rao discussion of personal key-custody practice rather than protocol mechanics, and it runs entirely among community members with no team voice weighing in. (2) It follows directly from Nov 11's SN3 compromise, where the owner later acknowledged a proper multisig or vault setup could have prevented the loss, and extends Nov 12's quantum-resistant wallet rollout beyond key type to how keys are actually signed day to day. (3) No official voice addressed the risk, and the thread closed on a personal preference rather than a concrete recommendation.

  2. SN116 (Carbon) launches Motor 1.8 on mainnet and immediately develops a grading backlog, cancelling roughly 75 queued submissions after two validator restarts.

    (1) The team confirmed cancelled submissions were not the miners' fault and were not charged, committing to restore them to their original queue position on resubmission. (2) The backlog formed because the validator processes one submission at a time while crown-check rebuilds took up to two hours each, prompting the team to pause king rechecks and raise the immunity_period hyperparameter from 5000 to 14400 blocks to protect queued miners from deregistration. (3) UID 90 was crowned Carbon's first king on the new mainnet, and the team said a live queue list would be added to the dashboard by the end of the weekend; the subnet's token fell 19.89% over 24 hours.

  3. A #General member questions why subnet registration costs now halve every 2 days instead of 4, confirming a deliberate change but getting no insight into the team's reasoning.

    Is there an error with the Subnet Registration Cost? I saw that it halves every 2 days (should be be 4 days), so the cost is only going down, right? drew confirmation: There was a recent change to the Subnet Registration Cost mechanism, and it now halves every 2 days instead of every 4 days. Pressed on intent with Is that the purpose the team wants to achieve?, the reply was That's not something I can confidently speak to, as I don't have insight into the team's specific intentions behind the change. (1) This is a second, independent run at the registration-price ratchet mechanism the channel worked through on Nov 9, when a member traced a fee drop to a specific sudo-set parameter without ever learning who held authority over it. (2) The unanswered question of intent today mirrors that unanswered question of authority five days earlier -- across two threads, no official source has explained the rationale behind changes to this pricing mechanism. (3) The exchange stayed civil, with the responding member explicitly declining to speculate rather than guess.

  4. SN24 (ms.computer)'s ownership briefly flips on-chain via an unintended conviction takeover, which the owner attributes to a coldkey-swap bug.

    (1) Ownership shifted at block 9503363 from one coldkey to another; the team later said a bug in conviction accounting following a coldkey swap doubled locked conviction on the owner's own validator key, triggering the unintended takeover of the subnet's own hotkey. (2) The owner confirmed he remained the subnet owner and referenced an upcoming reveal involving decentralized compute infrastructure with up to 2,500 miners. (3) The subnet's token fell 13.5% over 24 hours; the episode surfaces conviction-based ownership transfer as a live mechanism risk rather than only a theoretical one, following #Rao's Nov 11 clarification that subnet ownership is contestable through conviction.

  5. Other: SN120 (Affine) launches a full validator rewrite, SN44 (Score) tightens its recovery mechanism, and SN105 (Beam) faces a Prism-bypass allegation.

    (1) Affine v2 moved weight-setting to a new writer and cycled through six new kings as round time limits were extended from 100 to 180 minutes. (2) SN44 (Score) restricted its commit_recover mechanism to hotkeys that have missed no more than 6 consecutive triggers, while miners separately flagged a display delay between weight changes and shown incentives. (3) On SN105 (Beam), a miner alleged a paid transfer was used to bypass the Prism mechanism, concentrating work among five orchestrators; the team said it would look into it.

Discord conversations worth following

#General

2 REPORTS

Notable

A member questions whether a faster subnet registration-cost decay is intentional, and gets confirmation of the mechanism but no insight into the team's reasoning

Question ·

A member asked Is there an error with the Subnet Registration Cost? I saw that it halves every 2 days (should be be 4 days), so the cost is only going down, right?

Clarification ·

Another member replied that this reflects a deliberate change rather than a bug: There was a recent change to the Subnet Registration Cost mechanism, and it now halves every 2 days instead of every 4 days. So yes, the cost decreases at a faster rate under the updated schedule.

Pushback ·

The original asker pressed on the design implication: But the unlock time for registration remains the same (2 days). So the cost keeps decreasing over time. Is that the purpose the team wants to achieve?

Non-answer ·

The responding member declined to speculate on intent: That's not something I can confidently speak to, as I don't have insight into the team's specific intentions behind the change. I believe the team has its reasons for adjusting the mechanism, but I'd rather not speculate on the intended outcome.

3 things to read from it

  1. This is a second, independent run at the registration-price ratchet mechanism the channel worked through on Nov 9, when a member traced a fee drop to a specific sudo-set parameter and described the price as 'a slide — easy to go down, almost impossible to climb back up'; today's thread confirms the decay cadence itself was changed again rather than buggy, but again without any team voice confirming why.
  2. The unanswered question of intent mirrors Nov 9's unanswered question of who holds sudo authority over the parameter -- across two separate threads five days apart, no official source has explained the rationale behind changes to this pricing mechanism.
  3. The exchange stayed civil and mechanism-level, with the responding community member explicitly declining to guess at the team's purpose rather than offering an unfounded explanation.

In the week’s context

  • On Nov 9, a #General member traced a subnet registration-fee drop to the sudo_set_lock_reduction_interval parameter and described the pricing as a one-way ratchet, with the question of who executed the change left unanswered.
  • On Nov 10, a member credited a named figure with following through on a prior promise to lower the registration fee, a claim that went uncorroborated in the channel.

Where it stands

Open: the faster halving schedule is confirmed as an intentional protocol change, but no official voice has addressed why the schedule was altered or who authorized it.

Other

Other: a new member's DM solicitation drew a scam warning, a sturdy-related rug-pull question went unanswered, routine scam alerts repeated, and weekend greetings filled out the rest of the day

Mixed ·

A new member introducing themselves as an AI task designer was warned off soliciting DMs -- Asking people to DM you is probably not the best move, as that’s what scammers do, and that’s how you can get scammed the easiest -- and edited their post in response. Separately, a member asked why did sturdy rug pull? with no reply in the material. The channel's standing scam warnings recurred twice, Don't talk with him, wallet drainer inside. and scam, alongside a weekend greeting exchange, Happy weekend bittensor, and an unanswered question about a dedicated subtensor EVM channel, I think there was a dedicated channel for subtensor evm, no?.

What to read from it

  1. Volume fell to 17 messages, the channel's lowest point since Nov 8's 15, with nothing in this remainder touching tokenomics, governance or protocol direction.

Where it stands

Closed through standard warning, edit, or silence; the sturdy and EVM-channel questions remain unanswered in the material.

#Rao

1 REPORT

Notable

A day-long #Rao conversation shifts from on-chain spending limits to personal key-security practices, debating whether airgapped signing devices actually stop theft

Opening ·

Continuing an earlier thread about securing a wallet with spend limits, a member weighed the native option's weakness: only catch with the native one is smalltransfer caps each call under 0.5 tao but theres no per day cap, warning so a leaked key can just loop it till its empty and concluding id rather do the contract route you said if the limit actually has to hold.

Expansion ·

Hours later another member reframed the risk around everyday device hygiene: the biggest problem is that people use the same laptops/desktops to sign things and also to work, browse, emails, "job interviews" and so on, so of course sooner or later you will compromise your machine, maybe from your own mistake or just becase there is a bug in some part of the toolchain, arguing people should once for all understand that they sohuld invest in a separate device, totally airgapped just for singning stuff, including their development commits, everything that requires signing shuold be done on a separate device, and cautioning that discipline alone isn't enough since some dev responsible of some component of your machine may get compromosied and distribute malware wihtout knowing months before hackers make them effective.

Pushback ·

A member challenged the airgap solution directly: airgap only saves you if the device actually shows what youre signing tho and if it just shows a hash youre still trusting the infected laptop to build the tx, adding concern that kinda worried thats where the ledger flow for hashed accounts is going, device only displays a digest.

Reply ·

The first member defended airgapping's value regardless -- It helps so not to expose potential vulnerabile services to the internet, -- and pointed to precedent: Besides many of the latest high profile heists were possible exactly for the reasons i described, naming Including binance one. The other member partly conceded, fair, keeping the signer off the network kills a whole class of attacks, but raised a scarier case -- but bybit was the scary one for me, signers had hardware wallets and still approved stuff they couldnt actually read -- landing on so id want both, airgapped and the device decoding what it signs.

4 things to read from it

  1. This is the first extended #Rao discussion of personal key-custody practice rather than protocol mechanics, and it runs entirely among community members with no team voice weighing in.
  2. The worry about ledger flows that 'device only displays a digest' speaks to the same hashed-account signing concept Subtensor's own SDK change addressed on Nov 9 by erasing unused hashed-account signing keys from memory -- today's thread questions whether that design is secure enough in practice.
  3. The discussion follows directly from Nov 11's SN3 (Teutonic) owner-coldkey compromise, giving the channel's abstract security debate a concrete recent precedent inside the ecosystem itself, alongside the Bybit and Binance examples cited from outside it.
  4. It also extends Nov 12's quantum-resistant wallet rollout thread beyond key type to the harder question of how keys are actually held and signed day to day -- a gap neither day's wallet discussion addressed.

In the week’s context

  • Nov 11: SN3 (Teutonic)'s owner coldkey was compromised, with 101,141 alpha moved and sold for roughly 2,068 TAO, the owner later acknowledging a proper multisig, proxy or vault setup could have prevented the loss.
  • Nov 9: Subtensor merged an SDK change erasing hashed-account signing keys from memory once unused, among other non-breaking hardening fixes.
  • Nov 12: #Rao moved the quantum-resistant wallet thread from promise to product as the TAOApp extension went live, with a member walking through key-type tradeoffs and confirming migration works via a coldkey swap.

Where it stands

Unresolved; the exchange ends with one member's personal preference for combining airgapping with a transaction-readable signing device, but no team or official voice addressed the underlying risk, and no concrete fix or recommendation was adopted.

Subnets

5 REPORTS

SN116 24h −19.89%

Carbon

What happened
SN116 (Carbon) launched Motor 1.8 on mainnet, with the team publishing registration and submission instructions for miners. Validator restarts to deploy updates cancelled roughly 75 queued submissions without charge, and a grading backlog formed because the validator processes one submission at a time while crown-check rebuilds took up to two hours each. The team paused king rechecks to clear the queue and raised the immunity_period hyperparameter from 5000 to 14400 blocks to protect queued miners from deregistration. UID 90 was crowned Carbon's first king on the new mainnet.
Community
Miners asked repeatedly when mainnet and the Cooling module would go live, and some questioned slow validation speeds and specific grading outcomes. A community member asked the team to add a queue list to the dashboard so miners could see when submissions would be scored.
Official
The team confirmed the cancelled submissions were not the miners' fault and were not charged, and committed to restoring affected submissions to their original queue position upon resubmission of the same recipe. The team also said a live queue list would be added to the dashboard by the end of the weekend.

SN105 24h −3.58%

Beam

What happened
A miner alleged that an orchestrator used a paid transfer of $50 inside 'rooms' to bypass the Prism mechanism, with the work ending up concentrated among only five orchestrators. The team said it would look into the issue. Separately, the team confirmed it had optimized 1TB transfer speeds, cutting transfer time from over 50 minutes to under 4 minutes, which temporarily disrupted some miners' tier rankings.
Community
Several miners argued paid transfers let money buy Prism scores and higher tiers regardless of actual performance, while others maintained the ranking system reflects genuine performance and that paid transfers are not inherently unfair. Some miners reported being stuck in lower tiers for an extended period and said they were considering stopping.
Official
The team said it will look into the reported room exploit and acknowledged the recent transfer-speed optimization temporarily affected miners' tier positions.

In debate · Do paid transfers in SN105 (Beam) 'rooms' let money buy Prism rankings? · 6 participants

  • Paid transfers let whoever has the bigger budget capture emissions regardless of real performance.

    “allowing the paid transfers making the subnet more of a casino. with the right budget you get to harvest all the emissions.”
  • Paid transfers simply reflect genuine transfer performance and are not unfair.

    “Paid transfers aren't a problem. If your system performs well on the :43 transfers, you'll rank higher and receive more chunks across ALL transfers.”
  • A miner running the network suggested the ranking discrepancy may stem from BeamCore's own mechanism rather than deliberate bypassing.

    “I believe this is an issue with BeamCore itself, rather than something the miner is bypassing.”

SN24 24h −13.50%

ms.computer

What happened
Ownership of SN24 (ms.computer) briefly shifted on-chain at block 9503363 from one coldkey to another via a conviction takeover. The team later said the shift was caused by a bug in conviction accounting following a coldkey swap, which doubled locked conviction on the owner's own validator key, "Missy", triggering an unintended takeover of the subnet's own hotkey. The team also referenced an upcoming reveal involving decentralized compute infrastructure with up to 2,500 miners.
Community
Community members asked what had happened and whether the subnet had been sold, with some posting sarcastic remarks about the ownership change. Others pressed for confirmation that the owner still controlled the subnet and asked for a fuller explanation of the mechanism involved.
Official
The owner confirmed he remained the subnet owner and later explained that a coldkey-swap bug doubled locked conviction on the "Missy" validator key, causing the unintended takeover.

In debate · SN24 conviction takeover incident · 25 participants

  • Some community members suspected the ownership shift meant the subnet had been sold off.

    “thanks for selling lol”
  • The owner insisted he retained control and that the change resulted from an unexplained conviction lock issue.

    “I'm still the owner. The old conviction lock went through somehow”
  • A community member sought clarification on whether a prior takeover attempt had simply completed through the bug rather than being a new action.

    “you tried to takeover the subnet through conviction(months ago). It wasnt needed because Const handed over the keys.”

SN120 24h −1.88%

Affine

What happened
SN120 (Affine) launched Affine v2, a full rewrite of the validator with new rules, batch limits and a round time cap, moving weight-setting from the v1 to the v2 writer. Successive v2 rounds produced a series of new kings -- UID 929, UID 876, UID 983, UID 865, UID 957 and UID 961 -- as the round-based training and scoring system iterated from r000005 through r000011. The round time limit was later extended from 100 to 180 minutes, while batch_cap and the z_bar threshold stayed unchanged.
Community
Miners discussed training setups and hardware, with one miner describing high ongoing compute costs that later became cheaper and more stable. Several community members raised concerns about registration fees and asked the team about them, and one miner suggested adding a period between rounds so miners could train on the latest head rather than the old one; another miner responded that this could slow down evaluation.
Official
The SN120 (Affine) team posted regular round-by-round updates on affine.io covering new kings, pass rates, z-scores and parameter changes, including the extension of round time limits from 100 to 180 minutes.

SN44 24h −2.02%

Score

What happened
SN44 (Score) miners continued to flag disputed ground-truth annotations across several tracks, including room dirtiness, vehicle, beverage, PPE and petrol detection, with repeated requests to review and remove specific blocks. The team introduced a new restriction on the commit_recover mechanism, allowing recovery only if a hotkey has missed no more than 6 consecutive triggers for an element, after which the commitment is permanently lost. Separately, miners reported that after winners changed in the PPE, beverage and vehicle elements, non-winning miners kept receiving emission; the team said this reflected a delay between setting weights and the incentives displayed, not a backend issue. A miner also asked why an unregistered, non-top miner was receiving incentive on the football task.
Community
Miners submitted numerous block-check requests disputing false positives and missed ground-truth annotations in several tracks. A community member questioned the football scoring behavior, and others pressed for clarification on the emission discrepancy after winners changed.
Official
The team announced the new commit_recover restriction, removed several disputed blocks after review, and said the emission issue reflected a display delay rather than a backend bug, while flagging the unregistered miner's incentive on the football task as something to check.

More subnet community coverage every day in the Telegram channel →

Ecosystem debate

16 participants

SN116 (Motor) mainnet launch: queue backlog and cancelled submissions

SN116 (Motor), run by Carbon, launched on mainnet and quickly developed a severe evaluation backlog, with grading taking 7-11 minutes per submission and crown checks adding further delay. The team confirmed that two validator restarts cancelled roughly 75 queued submissions across many miners, none of which were charged or counted against a slot.

Positions

  • Some community members questioned whether the first king competition was fair given the server failures and lost submissions.

    “Is the result fair? because as your server failure, some submissions are lost, so I think the competition should start again.”
  • Miners reported long waits with no visibility into when their own submissions would be evaluated.

    “How many recipes are currently in the queue? I’ve been waiting for over 8 hours, but my recipe still hasn’t been evaluated.”
  • At least one miner characterized the disruption as a minor, manageable cost compared to issues seen on other subnets.

    “No worries, I'm used to it, it's really not a big deal.”
Official response
Carbon acknowledged the cancellations were caused by its own validator restarts, said nothing was charged and no slots were used, and committed to restoring cancelled submissions to their original queue position if resubmitted with the same recipe from the same hotkey. The team also said king regrades were paused to let the backlog clear, a queue-position dashboard feature was coming, and said a miner who kept the crown would be regraded like any other contender to confirm the result holds.
What to watch
Whether the throughput fixes and restored queue positions clear the backlog without further cancellations, and whether the first crowned submission holds up after regrading.

Voices on X

  • @taodaily_io

    Yanez (SN54) reported doubled sales in Q3 and October, estimated ARR above $500K, and a new client described as one of the world's largest social networks.

  • @robotensor_ai

    Attune (SN58) recapped becoming the first subnet live on Gamma, four new Vector Level 4 champions, and a discussion on which robot policies will succeed.

  • @cathedralhq

    Cathedral (SN94) described accidentally deleting three of 4,000 sealed agent computers on October 20, saying signed records proved the deletion was their own and illustrated the subnet's proof-of-execution offering.

  • @robotensor_ai

    Attune (SN58) said it is the first subnet live with Bittensor's new Gamma onchain usage credits, letting users buy Horizon robotics training data, with model and playground access coming next.

  • @CryptoZPunisher

    A community member noted that since dTAO's February 2025 launch, the share of TAO allocated to subnets has risen to 17.28%, though the metric has recently stalled amid weak market sentiment.

  • @carbonphysicsai

    Carbon announced that its mainnet is now live.

  • @VictorVL_EN

    A community member released a 22-minute video explaining Bittensor's concept, funding model, real products and revenue, and its fit for AI agents.

  • @JesusMartinez

    A community member argued TAO emissions flow into subnets tied to TAO, calling root staking the real inflation at about 5%, and cited subnet revenue and buybacks rising with month-over-month revenue up 70%.

◆2026-11-15 · End of this edition

Recent editions

Past editions +