Dates and verification labels belong to the original edition. A past verified announcement can now be outdated. Historical project names are retained; an SN filter searches the slot across registrations. All keywords must match the story text or subnet label. Ratings describe editorial priority, not investment quality.
SN55 NIOME announces its pharmacogenomic inference network
SN55 NIOME announced a live pharmacogenomic inference network at 13:03 UTC today. Miners analyze synthetic patient cases, infer genetic variants, derive metabolizer phenotypes and match prescribing actions to CPIC reference tables; validators compare them with known answers and score accuracy, consistency and confidence. This makes the target a reproducible research benchmark rather than collecting private patient genomes. The announcement names seven genes, but the linked repository’s current scope lists five: CYP2C19, TPMT, CYP2C9, SLCO1B1 and CYP3A5. DPYD, CYP2D6 and more difficult read-based cases remain on its documented development path. The launch is a direct team announcement backed by public implementation; this scan did not independently demonstrate production rewards, clinical validation or use in patient treatment.
SN55 NIOME publishes a substantive new inference task and inspectable scoring design. The actual implemented scope and unverified clinical and production outcomes are explicit.
SN65 True Performance Network opens TPN-005 with an 11 GiB model ceiling
SN65 True Performance Network has opened TPN-005, its first competition since the Fluid Bench upgrade. The target is Ministral-3-8B-Uncensored: fit within an 11 GiB measured memory ceiling while retaining a 32,768-token context and capability on Fluid Knowledge. The live validator configuration lists three winners with a 60/25/15 split of this competition’s pool. Miners now pay for and run their own benchmarks before committing; runs must finish before scoring starts and must match the submitted model’s exact revision and file. The team says the private question set is intended to resist benchmark memorization. Published competition rules are not a winning result: weights begin during the later distribution phase, and no TPN-005 winner or payment was established in this scan.
SN65 True Performance Network has an active, directly checked competition with a new evaluation process and concrete memory, context and reward rules.
SN74 Gittensor assigns 30% of its miner pool to compute
SN74 Gittensor’s new code allocates 70% of miner emissions to open-source work and 30% to compute, replacing the previous 100% open-source allocation. Compute weights come from the controller’s signed scoring record; an invalid or stale record recycles that share rather than paying unsupported work. Related changes add GPU rates and restore multi-GPU handling. This gives supplied compute an explicit place beside software contributions in the reward model. The split is verified in the merged constants and validator path, but this review did not reconstruct the first applied on-chain distribution or miner receipts. Published rate targets are not guaranteed dollar earnings.
SN74 Gittensor makes a material incentive-allocation change with clear accounting rules, while actual distributions and returns are not inferred from code.
SN74 Gittensor’s SparkInfer v0.6.32 improves Muse Glimmer serving
SN74 Gittensor released SparkInfer v0.6.32 with a targeted memory change for Muse Glimmer 30B. On an RTX 5090 with sixteen simultaneous 8,000-token prompts, its published test reports 351 output tokens per second and a 2.5-second first token, versus 281 and 9.7 seconds before the change. The same-GGUF llama.cpp comparison reports 76 tokens per second and 7.9 seconds. The gain comes from reserving less spare GPU memory and making more available for active request context; other models keep their existing headroom. These are workload-specific release measurements, not a universal model or hardware ranking, and the tests were not independently repeated.
SN74 Gittensor publishes a fresh serving release with explicit hardware, prompt, concurrency and competitor conditions.
SN78 UMI merges C5 delivery repairs; rewards remain inactive
SN78 UMI has merged a substantial C5 repair covering request delivery, retained-work recovery and bounded retries through runtime changes. The aim is to preserve acknowledged work and distinguish missing evidence from a miner’s missed opportunity. The code still requires qualification of the actual evaluator service environment before broader release. Its checked public API at 14:29 UTC continued to show open admission and 203 accepted submissions, but rewards_active and chain_submission_authorized were both false. That prevents the repair from being mistaken for settled rewards. An October sign-language app beta remains a reported target, not a confirmed public launch.
SN78 UMI ships meaningful recovery code while a directly checked status endpoint keeps intake, reward activation and the future app launch separate.
SN81 Reliquary adds faster reward-backlog recovery and fixed task contracts
SN81 Reliquary’s new accounting code allows up to four archived periods of rewards to enter together, rather than making a settled backlog wait one period at a time behind ongoing work. Reward totals stay unchanged, and replacement records prevent the same archive being counted twice. Operators must update weight setters before moving the backlog, or older software can clamp the catch-up and burn it. A separate merge fixes task identities and result delivery to their signed contracts, while sandbox infrastructure failures are kept distinct from an agent failing its task. These changes target delayed settlement and reliable training data; merged repairs are not independent proof that affected miners have received their outstanding rewards.
SN81 Reliquary fixes a specific persistent reward delay and strengthens task consistency, with upgrade order and payment verification limits visible.
SN87 Provenonce opens its code and testnet attestation history
SN87 Provenonce now publishes MIT-licensed code and a checker for its repeated weight-setting on Bittensor testnet subnet 582. The public page lists 21 attested runs beginning October 4, with the first block at 8152574; this builds on the September 28 first graded cycle already covered. The proposal evaluates whether autonomous work followed its declared requirements, rather than merely finishing. Its own limits are explicit: miners are run by the team, tasks are synthetic fixtures, outside miners are not enrolled, and public readers cannot reproduce scoring because the reference executor is private. The verifier’s overall result therefore stays UNVERIFIED even when chain checks pass. This is new public evidence and access, not a mainnet commercial launch.
SN87 Provenonce makes its implementation and repeatable testnet evidence inspectable, including what outsiders still cannot verify.
SN91 Cascade’s Ephemeris ensemble leads its current live forecasting benchmark
SN91 Cascade has published an Ephemeris update combining six forecasting models instead of betting on one. Its Weir page showed the ensemble first in the current round, with CRPS 0.581 and MASE 0.767; lower is better. That round covers 1,280 forecast windows across 289 live series and 19 entries, including statistical baselines and other ensembles. The October 6 update weights models by measured accuracy and replaces PatchTST-FM r1 with r2. This is a useful product and evaluation advance: forecasts can combine strengths across changing real-world series. Weir is separate from the static TIME comparison mentioned in the announcement, and these displayed results were not independently rerun. SN91 Cascade’s own Bittensor-trained model is still described as coming soon, not already included.
SN91 Cascade supplies a fresh ensemble release with a visible, scoped benchmark lead and a clear distinction between current inference and its future subnet-trained model.
SN97 Albedo moves evaluation toward SN53 engy and quicker fallback
SN97 Albedo’s new code routes GLM-5.3-Flash judging and question generation through SN53 engy, with an OpenRouter fallback using GLM-5.2. It also uses Jev for question and milestone deduplication, and changes generation so timeouts can fall back promptly rather than holding up the whole queue. The practical aim is faster, cheaper evaluation using another Bittensor service, while retaining a route when the preferred provider fails. The implementation is verified, but the stated activation after pending evaluations finish is only a channel report. This scan did not confirm all validators had changed providers or measure production savings.
SN97 Albedo has concrete evaluation and cross-subnet inference changes, with fallback model differences and rollout limits stated.
SN105 Beam announces a separate professional-services partnership
SN105 Beam says Beam Professional Services has partnered with a consulting firm to support current client engagements and integrate Beam into existing workflows. The announcement explicitly describes that organization as a separate services entity supporting the ecosystem. Its scope includes data movement, streaming, Rooms and programmable actions, but the consulting firm is unnamed and no contract value or independently confirmed customer deployment is supplied. This is a reported integration effort, not proof of subnet revenue. The separate QoS-whitelist issue remains under repair; the October 1 chunk-size fix is not a new fix today.
SN105 Beam has a fresh business-development announcement, with the separate entity, unnamed counterparty and unverified commercial outcomes retained.
SN107 Minos details its second mechanism for genomic AI research
Following its September 29 conference preview, SN107 Minos detailed Incentive Mechanism 2 in an article published at 22:33 Eastern on October 5. Its existing synthetic genomic data engine supplies examples with known answers; the new mechanism is designed to have miner-run AI agents find research questions in the scientific literature and turn them into verifiable evaluation briefs. The team describes a native Mac app, an iPhone companion and inference through SN28 SayGM, SN53 engy or SN64 Chutes, with evidence stored on SN75 Hippius. The angle is broader participation and demand between subnets: agentic research can use an AI subscription or API instead of rented GPUs. The article describes internal testing and the intended workflow, but this review did not verify an open public app download, mechanism-wide activation or rewards to outside participants.
SN107 Minos details a substantial second incentive design linking research questions, synthetic data and other subnet services. Public availability and paid participation remain developing.
SN108 ChipForge’s NPU challenge is open, with real submissions under evaluation
SN108 ChipForge has crossed the launch line. Its public leaderboard at 14:34 UTC showed an active challenge to build an NPU and optimized kernels for four AI models, six submissions from five miners, and no winner yet. Designs run MLPerf Tiny workloads for voice, vision and sensors; outputs must match reference answers, with chip area and performance evaluated separately. The challenge ends October 20. This advances yesterday’s no-active-challenge observation, rather than repeating a promised start date. All three validators still showed their stake burning emissions until a qualifying design wins, so an open contest is not a completed payout. At 14:28 UTC, alpha/TAO was up 293.01% over 24 hours; that is market attention, not a dollar return or evidence of chip sales.
SN108 ChipForge now has a directly observed active hardware-design competition and submitted work. Its rating reflects activation and evaluation, not the price spike.
SN116 Carbon’s new validator code retains payout activation gates
SN116 Carbon’s latest validator change permits its service key to sign a registered winner-weight intent while retaining integrity checks and refusing other weight shapes. The commit also says broader mainnet authorization and switching the testnet path on remain held. This is progress in the reward infrastructure for its industrial physics AI work, not proof of completed rewarded mining or customer delivery. At 14:28 UTC the market snapshot showed +110.94% over 24 hours in alpha/TAO; its +2,552.02% seven-day figure spans the slot’s pre-registration history and is not a Carbon investment-performance record. Yesterday’s eight proposed challenges are not eight completed customers, and a newly claimed Discord channel is not evidence of a subnet ownership transfer.
SN116 Carbon’s market outlier warrants a fresh primary-source check; its actual authorization change is covered with retained gates and slot-history limits.
SN120 Affine records a completed optimizer step and remaining training limits
SN120 Affine’s operator log now records E15 finishing with durable optimizer step six, then E16 continuing under a bounded training selection after a 256-document cap caused a retry loop. It reports 309 eligible contributions, with 256 selected for that training job and 53 unselected; all remain eligible for continuous audit and reward estimation. The distinction is progress in the training pipeline, not demonstrated learning improvement. Its separate 128-task comparison scored 83 before and 82 after, which the log explicitly says is not evidence of convergence. Step seven and a faster completed epoch remain unclaimed. These are public operator records, not independent validation of every run or affected miners’ wallet receipts.
SN120 Affine has a concrete completed training checkpoint and documented recovery, with negative evaluation evidence and unfinished steps preserved.
SN124 Swarm reportedly re-evaluates champions after five environment fixes
Today’s team-attributed brief says SN124 Swarm is rolling out V5.1.6.5 to correct five miner-reported environment problems: landing-score penalties, disappearing village pads, rescue victims placed below slopes, coupled forest difficulty and search area, and terrain loading order. Champions in affected modes are being re-evaluated; Office Interceptor is unchanged. These repairs matter because an environment defect can distort which drone controller earns rewards. The original release and full validator rollout were not independently confirmed, so this remains a reported fix rather than proof that every affected ranking has settled. Its October 7–8 World Summit AI appearance is still upcoming.
SN124 Swarm reports a substantive scoring-environment repair and targeted re-evaluation, with source and rollout limits visible.
Talisman adds ForeverMoney’s TAO bridge inside wallet swaps
Talisman and ForeverMoney both announced on October 5 that TAO bridging is now available inside Talisman Swaps. The route connects Bittensor Finney and EVM with Base and Robinhood Chain using ForeverMoney’s Chainlink CCIP infrastructure. The practical advance is distribution and access: holders can begin the cross-chain route in a wallet they already use for Bittensor staking, while people acquiring bridged TAO can move back toward native subnet participation. This is a new wallet integration, not the first launch of TAO on those networks. TAO on Base and Robinhood is a bridged representation. Both announcements and the integration article were checked; this review did not submit a bridge transaction or independently audit reserves and security.
A confirmed wallet and bridge integration expands practical TAO access, supported by both parties and direct documentation rather than unrelated platform activity.
Grayscale adds Coinbase custody for its Bittensor Trust
Grayscale’s October 5 Form 8-K confirms Coinbase was added to the Bittensor Trust’s custody arrangements through a September 29 agreement amendment. Coinbase is intended to safeguard a portion of the trust’s TAO; BitGo remains primary, and the filing says Grayscale had not yet decided how much TAO to move. It also updates the BitGo agreement through a September 30 amendment. Adding a second custodian is a concrete institutional infrastructure development and is described as risk management as the trust grows. The document does not announce an ETF approval, exchange listing, new Coinbase retail TAO listing or completed transfer amount. Those are separate claims from the custody change verified here.
The primary SEC filing establishes a material TAO trust custody change, with exact agreement dates, retained primary custodian and no invented ETF approval.
Today on Shizzy: validator choice at 11 AM Eastern
Today’s scheduled public stream is “Your Stake. Your Decision. Inside Bittensor’s First Democratised Validator,” with guests from Subnet Summer and Medulla Labs. YouTube lists the start as October 6 at 11 AM Eastern / 8 AM Pacific. The discussion is about validator choice and stake participation; the title’s “first” wording is the episode’s framing, not an independently established global-first claim. The site also now lists yesterday’s public “Bittensor TAO at $300: Is This the Launchpad for Liftoff?” video, 22:09. The October 4 members video remains correctly gated, and the SN49 Nepher Robotics replay stays in the library.
A verified public schedule and refreshed video library give readers a concrete way to follow today’s Bittensor discussion.
SN116 Carbon announces it is live and publishes eight industrial challenges
SN116 Carbon announced it was live at 18:26 UTC on October 4, just after yesterday’s news cutoff. Today it published an initial eight-challenge physics AI portfolio: server cooling, chip thermal management, optical links, battery charging, precision motors, low-vibration robotics, acoustic equipment and passive mixing. The idea is to help engineers screen designs and make decisions with evidence, rather than treating every AI prediction as trustworthy. The portfolio is proposed research, with proof requirements attached to each use case. Its dollar-saving examples are hypothetical customer benefits, not achieved results, eight customers or Carbon revenue. The launch announcement advances yesterday’s registration story; completed rewarded mining and customer deployments were not independently demonstrated.
SN116 Carbon moves beyond registration with a dated team launch announcement and a concrete industrial research portfolio. The rating rests on that progress, not price or hypothetical savings.
SN9 iota reportedly deploys a peer-discovery backend fix
Today’s team-attributed brief says SN9 iota upgraded its backend to redis-py 8.0.1 to resolve a peer-discovery problem that prevented miners from finding adjacent-layer nodes. This gives yesterday’s transient Redis recovery and staged-fix story a more specific reported implementation. The live backend and network-wide recovery were not independently checked. Macrocosmos’s distributed-training SDK remains a preview; this fix is not a full SDK launch.
A reported SN9 iota backend change addresses a defined miner-connectivity issue, while rollout and SDK readiness remain separate.
Searches published headlines and summaries, not private bookmarks, unpublished leads or source-page contents. Duplicate headlines within an edition appear once; separate editions remain separate so follow-ups are preserved. Share the page URL to retain your filters.
