Bittensor Daily News
Latest editionCorrection · · Gamma now leads; all source follow-ups and finalized coverage are retained.
Gamma reaches SN58: Attune offers robotics data for credits
At 13:57 UTC November 10, Robotensor announced that Attune’s first Gamma service is available: customers can arrange direct team access to Horizon robotics data using Attune Gamma. Horizon collects phone-controlled simulated robot demonstrations and first-person recordings of everyday tasks. Finalized block 9475959, extrinsic 15, event 208 at 12:54:12 UTC independently confirms successful SN58 opt-in. A full event scan from global activation block 9471027 through finalized 9476439 found the global activation and this subnet opt-in, with no GammaMinted, GammaSpent or GammaSpendSkipped events. Thus configuration and the team’s availability announcement are established; paid customer use or revenue was not demonstrated through that cutoff. “First subnet” is the team’s claim, consistent with the sole enabled entry in the reviewed state. Model access, bring-your-own-model playground usage and broader robotics services remain future plans.
Around Bittensor
26 more storiesBittensor enables the network-wide Gamma switch; subnet opt-in remains separate
On November 9 at 20:27:48 UTC, successful finalized extrinsic 9471027-5 enabled Gamma globally. The flag changed from false to true, with successful sudo and multisig outcomes. This was already quoted by Shizzy and is carried into the next morning edition without another quote. SN58 opted in separately the following morning, as covered in the lead story. Official documentation distinguishes network enablement, subnet enablement and oracle requirements; availability is separate from credit minting, spending and customer revenue.
SN124 Swarm announces a four-part operating plan and executes its allocation change
On November 10 at 14:09 UTC, Swarm announced four changes: a 20 TAO buyback-and-burn, an ongoing buyback programme, intelligence supplied exclusively to paying clients, and an increased allocation for stakers. The written original establishes an operating-policy announcement, not independently verified customer contracts, revenue or future purchases. At 14:33:12 UTC November 10, finalized block 9476454, extrinsic 7, event 186 successfully changed Swarm’s five-slot allocation from 0% burn / 0% recycle / about 18% owner / 41% validators / 41% miners to 0% / 0% / about 18% / 49% / 33%. Exact raw vectors, unchanged identity/registration and successful proxy execution are preserved. This redirects future emissions; it is not an 8% transaction burn. Root’s baseline is paid from and capped by the validator bucket, and historical yield estimates can lag. Separately, successful block 9476305 at 14:03:24 UTC used 19.9 TAO in add_stake_burn and actually burned 2,366.605401253 Swarm alpha. Its marker records one raw alpha unit more than AlphaBurned; the archive retains the actual burned amount. TAO is the purchase input, not destroyed TAO; the initiating wallet is not the registered owner and no team funding or customer-revenue source is inferred.
SN24 catch-up: Quasar became ms.computer; launch timing remains unconfirmed
Dated background, not a new November 10 rename: Successful SubnetIdentitySet at finalized block 9378507 changes Quasar to ms.computer and updates the website field. Full-window owner and registration comparisons found no executed ownership/registration change for this SN. Discord channel-claim commentary does not establish a subnet sale. The large dated alpha/TAO move does not independently establish its cause or a product launch. On November 1, block 9409822 changed the owner hotkey while the owner coldkey and registration 2538424 stayed unchanged. On November 4, block 9431616 announced a coldkey swap, without proving a completed transfer or sale. This morning’s full finalized lifecycle review found no newer SN24 identity or ownership execution in its covered interval. The official @ms_computer profile links ms.computer, whose page still presents “coming soon.” Community reports about testing, miner onboarding and a launch this week could not be confirmed from the primary Discord channel because it requires login. No sale, buyer, launch date or misconduct allegation is established.
SN78 Umi deregisters; a new unnamed subnet takes the slot
Finalized November 9 events removed Umi at block 9471098 (20:42 UTC), cleared its identity at 9471122 and registered a replacement at 9471124 (20:47:12 UTC). The replacement has no on-chain identity at reviewed block 9471431. Historical storage identifies the outgoing owner as 5GvYz1…cxVQ3 and the new owner as 5GQWJ3…EJuE. The successful registration queued 85.943556872 TAO as a lock, not a verified sale price or transaction burn. The SN78 number persists, but Umi history and the replacement are separate registrations. The allocation vector reset to the default during cleanup; no owner allocation-change event occurred.
Sami Kassab publishes internal ecosystem revenue and demand estimates
Sami Kassab, founder and managing partner of Unsupervised Capital, published November 9 research estimating ecosystem annualized revenue near $54 million, compared with roughly $20 million at the end of 2025. This is an annualized estimate from an internal dashboard, not audited realized revenue or a public reproducible dataset. His accompanying 90-day trailing chart estimates demand equal to 10.9% of TAO issuance: team buybacks 4.7%, miner registration 3.2%, subnet registration 2.1% and transaction fees 0.9%. The chart runs from June 2025 into November 2026. It is useful because it separates sources of demand instead of treating token issuance alone as the network economy, but the components are not all permanent burns, external customer revenue or independently verified cash flows. His goal of reaching 100% offset is an aspiration, not an achieved break-even result. Rounded $20 million and $54 million imply about 2.7 times, while the article describes roughly 2.5 times; retain that approximation difference rather than invent precision.
Stillcore publishes its Commanding Heights thesis
Stillcore argues that an open market commissioning, grading and paying for intelligence can capture value as individual models become commodities. Its November9 paper connects compute and storage, distributed pretraining, post-training and paid inference, and uses Reliquary as an example of a model pipeline drawing on separate subnets. It cites a SubConnect estimate of roughly $70 million annualized subnet revenue at November4, then projects $100 million by end2026 and $1 billion by end2027. Figure6 compares those run rates with about $300 million annual issuance valued at the fixed November6 TAO price; the resulting approximate23%,33% and3.3-times ratios are scenario comparisons, not measured cash coverage or current network break-even. Hatched bars identify forecasts, and projected revenue excludes Gamma burns. The paper also discloses that Stillcore and its principals own TAO and named subnet tokens. These are attributed investment arguments and unaudited third-party/firm estimates, not verified future returns or endorsements by the public figures quoted. Its future-tense Gamma discussion predates the separately verified global activation; no subnet adoption is established by this paper. Figures 2 and 3 map the subnet supply chain and Reliquary’s use of Teutonic, Targon and Lium; the cited 43.6% versus 35.1% same-size SWE-bench comparison is attributed to the report and team, not a reproduced benchmark. Figures 4 and 5 compare model-control and funding structures and group public AI commentary; the named commentators have not endorsed TAO. All six numbered figures were inspected across the original review and this morning’s follow-up. The report’s future-tense Gamma discussion is covered by this edition’s Gamma lead and network-activation report.
Yuma reports October Large Cap performance in USD and TAO
Yuma reported an October decline of16.9% in USD for its Large Cap Fund versus20.9% for TAO, a4.0percentage-point smaller loss. The same comparison corresponds to approximately5.0% relative appreciation measured in TAO, so the positive TAO figure does not mean a positive dollar month. Yuma also reports a5.5% decline since launch and says this was the strongest October TAO result among its three funds. Its chart shows daily USD changes as bars and cumulative USD performance as a line for October1-31. These are manager-reported results; this review did not reproduce fund NAV, fee treatment, custody or subscription/redemption-adjusted returns. The claim is historical performance, not external customer revenue or a return forecast.
Root dividend price-gate fix merges into release-489
PR3319 changes the all-holdings gate to weighted basket price impact and separates problematic deferred origins from healthy settlement. The primary PR describes a0.1% holding-NAV budget and separate2% incoming-value and mint-ratio limits, with recovery regressions. Merge is into release-489. Our fresh finalized block9473251 still runs488, so this is preparation for a fix, without a claim that live baskets have been repaired.
SDK hashed-account key erasure change merges into release-489
PR3317 changes core-owned derived Ed25519 signing keys to a wiping owner and removes an intermediate master-seed copy. Published regression vectors check unchanged key and signature outputs. Its scope is the core-owned derived key, not every temporary or caller buffer. This is a source-branch merge; packaged SDK distribution, installed-client uptake and exploit activity were not verified.
SN112 THOR details its demand-funded robotics labelling pilot
On November10 at14:01UTC, THOR said its first jobs come from signed robotics labelling contracts and linked its demand-first economics documentation. The docs describe a private beta with qualified, whitelisted firms, a full worker-reward reserve funded from alpha already held, staged verification/acceptance payments, and customer receipts allocated to separate SN112 alpha purchases. Signed contracts, collected receipts and executed buybacks are explicitly different milestones. The documented plan also proposes changing miner-burn/emissions routing into a disclosed multisig. The public revenue dashboard displayed0USDC deposits and customer receipts with an empty All transaction ledger. Its displayed contract is labelledEthereum chain1, while the docs describe a BittensorEVM deposit contract, leaving deployment/network reconciliation unresolved. This review did not establish executed customer payments, deployed policy changes or buybacks from this announcement.
SN103 Terza lists Fable5.2 with miner-specific token prices
Terza’s November10 12:29UTC original advertises Claude Fable5.2 at$1.69 per million input tokens and$8.47 per million output tokens, with1M context and Messages/Chat Completions compatibility. The current public pricing page lists claude-fable-5-2 with three quotes from one miner and those lowest input/output rates. These are provider listings and observed quotes, not an independently authenticated upstream model, a vendor endorsement or an executed inference/billing test. The advertised comparison with official-provider prices was not independently reproduced and is not used as a verified discount claim.
SN75 Hippius follows its mobile launch with a 10 GB free Drive tier
On November 10 at10:28UTC, Hippius highlighted the first10GB of storage free for its iPhone and Android app users. The current official pricing page independently lists a Free Drive plan with10GB storage. This follows the mobile availability already covered in the November9 edition, rather than another newly dated app launch. The team also says paid upgrades currently use its website rather than an in-app upgrade flow. On-device encryption remains the developer’s description; no signup, upgrade, key-recovery or encryption audit was performed.
SN117 EveryFrame adds model offerings for short video and animated images
EveryFrame’s November9 21:16UTC original advertises Gemini Omni1.2 with clips up to15seconds, generated audio and speech,1080p, vertical or widescreen formats, and reference/first/last-frame controls. Its published example is$0.82 for10seconds at720p, compared with$1.00 for its earlier Omni1.1 offering. A separately dated16:46UTC original, included as a missed earlier update, advertises GPT Image2.6,12-frame GIFs, photo-to-GIF and circle-to-edit, with stills from$0.04. These are the provider’s model labels, availability claims and example prices. The public product shell was accessible, but no generation, quality benchmark, vendor-licensing relationship or paid billing test was independently established.
SN81 Reliquary codes an opt-in price auction and tightens completion audits
A November 10 implementation and design document add an order-job auction that offers work to cheaper active miners first, limits work in flight according to the audit backlog, and settles verified output tokens at a common period price under the job’s cap. The feature is off unless an administrator explicitly enables auction pricing; the design still lists rollout and tuning questions. The inspected settlement and lease code supports the described mechanism, but no live order, realized price reduction, payment or associated burn was verified. A separate guard forces very short drawn completions through audit instead of allowing an unaudited pass when they are too short to be charged for refusal. Another patch expands accepted chat-template parsing. These are coded changes, not evidence that every production job now uses the auction; upstream tests were not executed by this review.
SN116 Carbon reports a testnet grading pause and fixes queue starvation
Carbon’s November 10 source notice says BurstThermal and Cooling testnet grading is paused while a fix awaits release. Uploads remain accepted and queued; the notice explicitly asks users not to resubmit those jobs. A different, small subset of requeued RibbedBracket and Motor uploads cannot be graded after a provider failure and should be submitted again using the same model, counting as a normal submission. Other requeued uploads are described as grading normally. The accompanying scheduler patch tracks already-served queue places and falls back to round-robin to prevent a frozen host entry from starving other tasks. These are a published operational notice and source fix, not independently verified restoration of grading or a mainnet rollout.
SN91 Cascade fixes evaluation history and adds checkpoint-integrity guards
November 10 source changes keep enough history for the full 4,096-point context at a 720-step evaluation horizon; shorter series retain their available history without changing eligibility. Separate changes cache up to four private-evaluation checkpoints on the GPU host and verify SHA-256 before reuse. The documented host and validator release must match: a capability preflight holds a round when required fields are missing rather than silently falling back or penalizing a miner. Dashboard lineage also restarts at the v2 activation instead of carrying the old model’s generation history into a fresh initialization. These are source changes and operator requirements, without independently verified fleet deployment or improved forecasting accuracy. Relevant implementation and documentation were inspected; upstream tests were not run in this review.
SN70 Fugal prepares a pinned v2 release and repairs monitoring tools
November 10 deployment configurations explicitly identify Fugal as Finney subnet70 and pin a v2 miner image by digest, with a planned activation at block9484920. That block is later than this report’s finalized9476685 cutoff, so this is release preparation, not verified activation. Testnet uses subnet552 and its own activation block. November 10 source fixes make the watcher read the score field from the reveal object rather than treating the entire object as a number, resolving a TypeError. The separate offline consensus-audit utility now compares the complete set of weighted UIDs, including the burn UID, so an all-burn result does not produce a false 100% divergence warning. This utility is not called by the validator’s consensus path, so the patch does not establish a change in live consensus or emissions. The nightly rehearsal also targets the main branch instead of a deleted release branch. Implementation and changelog were inspected; deployment and an independent upstream test run were not verified.
SN50 Synth confirms Forge partnership ahead of mainnet
Forge announced its partnership with Synth on November 9 and said Synth had committed capital before mainnet. Synth later quoted the announcement and replied “We’re in,” directly corroborating participation. Neither original specifies the amount or terms, and this review did not verify a transfer, live lending market or mainnet activation. The development is a confirmed partnership announcement with a stated capital commitment, rather than an executed funding total.
SN41 Almanac reports October fees and a verified alpha buyback
On November 9, Almanac reported October fees of $109,440, its first six-figure month, with 32% month-over-month volume/revenue growth and 111% growth in average trade size. Its chart labels $109,440 October buyback/burn value. These are team-reported business metrics, not independently audited customer revenue or a demonstrated one-to-one funding reconciliation. A successful add_stake_burn call at finalized block 9471196 on November 9 used 459 TAO to buy and burn 44,220.614304990 Almanac alpha. The matching AlphaBurned and AddStakeBurn describe one operation and are counted once. The TAO is buyback input, not TAO destroyed. Historical identity and registration were unchanged across execution. The initiating wallet differs from the currently verified owner; this is a subnet burn, without an unsupported claim that the team funded it.
SN1 Apex and SN25 URnetwork report latency gains in round-one tests
On November9, both Apex and URnetwork reported round-one latency-test results: p95 request time fell from19.6to7.1seconds, described as roughly63%, and larger downloads were about twice as fast in their tests. Apex says14other solutions exceeded the20% target. URnetwork says the winning code is the starting point for round two, not that every customer’s production traffic already achieves this improvement. The linked public result API identifies an elected provisional reward winner, approved integrity/source review and an active reviewed-source promotion datedNovember10 07:56UTC; settlement remains pending with no settlement receipts. The review read the result data but did not independently rerun the benchmark or verify a reward payout.
SN25 URnetwork carries Apex’s latency-test winner into round two
On November9, both Apex and URnetwork reported round-one latency-test results: p95 request time fell from19.6to7.1seconds, described as roughly63%, and larger downloads were about twice as fast in their tests. Apex says14other solutions exceeded the20% target. URnetwork says the winning code is the starting point for round two, not that every customer’s production traffic already achieves this improvement. The linked public result API identifies an elected provisional reward winner, approved integrity/source review and an active reviewed-source promotion datedNovember10 07:56UTC; settlement remains pending with no settlement receipts. The review read the result data but did not independently rerun the benchmark or verify a reward payout.
SN89 InfiniteQuant schedules a lower LF bond threshold and stricter drawdown rules
The code changes the elimination rule from8% to3% at day ends from November16 00:00 UTC. Equity strictly below97% of the unchanged all-time end-of-day high eliminates a book; the high is not reset at cutover, and exactly3% survives the strict comparison. Tests cover replays across the boundary. Live deployment and any individual account outcome remain unverified. A separate November 10 commit reduces the LF eligibility threshold from 1,000 to 500 alpha per UID starting with the November 11 00:00 UTC snapshot, which is November 10 at 7 PM Eastern. November 9 and 10 snapshots retain the 1,000-alpha threshold when replayed. The calculation divides a coldkey’s alpha across its hotkeys; configuration can override the defaults. This is a coded eligibility rule, without a verified payment, escrow transfer or burn.
SN8 Vanta removes Classic-account cooldowns and account-count limits
Vanta announced November 9 at 19:10 UTC that customers can buy multiple Classic accounts without the previous cooldown or account-count limit. A separate 19:52 UTC original says self-authored strategies can use its API under the same rules, split and weekly-reward structure. These are team-published participation terms; no account was purchased and no API trading or payout test was performed.
SN127 Zils documents admission tracks and worker safeguards
November 9 documentation describes SN127 mainnet admission and an 80% Soup / 20% customer-job adapter launch structure. It says two larger Soup pools are not yet open and their shares are burned. These are documented launch terms, without independently verified live weights or completed paid jobs. Qualification jobs are queued only while a worker is online and sending heartbeats; only runs trained by that hotkey count toward its three qualification runs per 30 days. Customer jobs require three passing qualification reports in the relevant size band. Merged queue-heartbeat code tolerates transient server errors until three consecutive failed renewals, resets the counter on success and stops immediately on refusal. These source changes do not verify production deployment, customer payments or payouts.
TaoStats announces PIN protection as an alternative to biometrics
On November 10 at 09:00 UTC, TaoStats announced that its app now offers PIN or biometric protection under Settings > Security > Protection. This is the provider’s release announcement, not an independently installed version or a security audit. No account security setting was changed during this review.
Shizzy publishes Can TAO Hold? and the site adds the public replay
The official channel’s November 9 public video, “Can TAO Hold? Bittensor Price Action vs. Real Progress,” is 18:35. Its watch page, public availability and initial playback were checked, and it is added as the latest item and default player in the video library. No complete audiovisual review or transcript verification is claimed. Carbon remains canceled, and no replacement date is promoted.
Quick briefs & watchlist
7 updatesSmaller developments and early plans. Verification labels distinguish confirmed information from developing claims.
SN8 Vanta records two qualifying direct alpha burns
Two successful direct burn_alpha calls on November 10 removed 145.543240996 alpha at 09:14:24 UTC, block9474860 extrinsic28 event216, and 131.741583863 alpha at 14:52:12 UTC, block9476549. Together they total 277.284824859 alpha. Independent event, call and historical identity checks confirm successful execution without a Vanta registration change. Both wallets differ from the registered owner, so owner funding or customer-revenue provenance is not inferred. Smaller events remain only in the new underlying scan evidence. These are explicit transaction burns, separate from emissions withholding, recycling and fees.
SN21 AdTAO clarifies its baseline gate, earning seats and receipt-history limits
November 10 documentation requires a score strictly above 1.05 times baseline and predictions for at least 90% of the outcomes scored for that baseline, preventing a submission from simply skipping difficult cases. One hotkey per coldkey earns for the day: highest standing wins, then earlier commitment and then the hotkey tie-breaker; other hotkeys may still run and score. The documentation distinguishes 28-day standing from the separate 35-day absence/charge eligibility window. It also discloses that the daily basis map contains 6,116 settled episodes only through August 23 and has not been rebuilt; later receipts expose per-outcome efficiency basis and account goal basis separately. These are documentation clarifications and data-coverage disclosures, not new customer results or verified payouts.
SN22 Desearch shares a reproducible research-report example
The November10 13:01UTC team post links a guide publishedNovember9. Its Python example searches five sub-questions, deduplicates URLs and near-identical titles, selects dated quoted findings, and exports Markdown, a sources CSV and rawJSON. It describes rebuilding a stored sample offline without making API calls. Returned publication dates can reflect page updates, and the guide explicitly limits coverage to returned sources. This is a documented example workflow, not a new search-engine release or an independently executed live API benchmark.
SN28 SayGM updates workers and reports an interrupted-image request fix
SayGM merged its v0.5.24 stable-version promotion on November 9 after dev preparation. An earlier merge makes gmcli worker list show each worker’s release tag and image standing, including supported, deprecated with a grace-end epoch, revoked, unknown or not reported. A failed image lookup still prints the list with a warning. These are reviewed repository changes; the version bump alone does not prove every production worker upgraded or that a particular deployment is supported. Separately, a November 9 at23:17UTC team reply reports a bug that kept an interrupted image request paused while charge status was checked. The team says the pause now clears within15minutes or through Continue. This is a provider-reported fix, without an independent paid generation or billing test.
SN36 Weir adds a GPU UUID format check to validator inventory
Commit3171d70 rejects GPU UUIDs outside the supported NVIDIA format with BAD_GPU_ID, assigns no weight and omits those GPUs from sellable inventory. Simulation IDs remain allowed when timing is off. This is validation code with regression coverage, not hardware attestation or proof that live validators deployed it.
SN51 Lium binds remembered filler load to the original container ID
The validator now remembers container IDs rather than names alone. A replacement container reusing a filler name does not count as the validator own load. The patch includes replacement-container regression coverage. Merge-only evidence, with fleet deployment unverified.
SN66 Conjectures retires source27 from active admissions
The validator merge pins a tasks revision where Math30Catalog.source27 is retired, and reduces the default tier from307 to306 targets. Tests distinguish readable historical tasks from admissible active tasks. This confirms a source-selection change in code, not a newly solved mathematical result or production rollout.
Stars indicate editorial priority, not investment quality. Routine competition updates receive at most three stars. Shizzy videos receive three; substantive non-video Shizzy coverage receives at least four. Scheduled events remain plans.
Sources & coverage notes
Audit trail
Coverage notes
SN58 adds a Gamma robotics-data offering. Swarm announces four changes, including a verified allocation shift. Plus the SN24 transition catch-up, Apex/URnetwork testing, product updates, fund research and finalized burns.
Coverage: Nov 9, 2026, 1:22 PM – Nov 10, 2026, 10:19 AM ET
Searched
- 128 active subnet rows reconciled with finalized owner/identity and current market data; 106 official sites loaded, 19 missing configured websites and 3 site errors retained individually.
- 18 required ecosystem accounts, active Injective partner and verified fund/researcher accounts reviewed through native X Posts/Replies/Reposts; actual bounded windows retained in the audit.
- Real Grok Auto discovery, attention and omission queries submitted; primary source and finalized chain verification override inconsistent generated claims.
- Complete finalized chain and lifecycle intervals, prior pending leads, public channel Videos and Live tabs, source repositories and sector classification review.
- Direct native source checks expanded to66 exact official-site-linked accounts and primary team profiles; unsuccessful loads, bounded histories and unreviewed media remain individually recorded.
- Existing authenticated GitHub access recovered 57 public repositories after initial unauthenticated errors. Selected primary code/documentation follow-ups were read and reconciled; scope and activation limits are explicit.
Public-data gaps
- Primary SN24 Discord requires login; testing, launch timing and sale claims remain unconfirmed.
- Some original subnet social posts and attached media, including selected older carried leads, remain unreviewed or inaccessible. Secondary digests and repository errors are discovery gaps, not evidence of no news.
- The mapped fund/people review includes verified native accounts and official public pages, but unresolved identities, unreviewed PDFs and portions of seven-day timeline history remain pending.
- Authenticated TaoStats v1 credit/entitlement reconciliation was not completed; no paid provider usage or direct account access is claimed.
- The site audit includes actual desktop rendering and automated desktop/phone tests; a native production phone viewport and a fresh synthetic wallet lookup remain unverified.
Browse all 94 editions
Conference news · September 30, 2026
Exploit Conference Catch Up · Day 1
The teams, the ideas and the practical takeaways from Exploit Summit’s first day in Montreal. From Gamma and phone-based AI to robotics, science and the business of running a subnet.
Conference news · October 1, 2026
Exploit Conference Catch Up · Day 2
Catch up on Exploit Summit’s second day: Carbon’s Pitchtensor win, SN107 Minos’ research app, agent controls, robotics data, inference partnerships and the decisions behind building a subnet business.