August 22, 2026 · Published daily
Subnet News — August 22, 2026
Vocence led a 45%-positive market tape while BASE fell hardest and iota reached 4.03% emission. Subtensor release 448 merged, BASE promoted B200 provisioning, and production or scoring changes landed across Ditto, AdTAO, MVTRX, Verathos, Phylax, MicroTensor, Gittensor, engy, OpenRoboto, Albedo, Pareton, Vanta, and KubeTEE. Score, Beam, Cascade, Sundae Bar, and Leadpoet carried active operational or incentive incidents.
- Coverage window
- Aug 21, 2026, 10:00 AM – 10:00 AM ET
- Subnets checked
- 128
- Subnets with updates
- 19
What matters first
Top signals
Subtensor release 448 merges with bittensor-core 0.1.4 versions
The release-448 branch merged after the August 21 cutoff, followed by a version sync for bittensor-core 0.1.4. This verifies the public release merge; it does not by itself prove the exact Finney activation block.
Vocence leads the tape as 42 subnets burn at least 90% of miner emissions
The August 22 cutoff tape showed 45% positive breadth, Vocence up about 118.9%, BASE down about 10.9%, and iota’s emission share at 4.03%. It also counted 42 of 128 subnets at or above 90% miner burn. Vocence’s move came from a thin market and should be read with its liquidity context.
Organized by SN#
Subnet desk
Vanta
Vanta corrects funding rates and snapshots unrealized PnL for payouts
Vanta fixed Hyperliquid funding-rate handling, made entity collateral-margin work non-blocking, and changed payouts to use unrealized-PnL snapshots. The code changes target payout correctness under concurrent market updates.
Pareton
Pareton isolates compile caches and keeps one bad image from aborting a round
Pareton pinned compile-cache directories per engine and changed GPU handling so one unpullable challenger image no longer aborts the entire round, while failed registry authentication remains fatal.
AdTAO
AdTAO publishes live score verification and miner status tools
AdTAO pointed its documentation to the live daily-verification mirror, documented the active scoring formula, and surfaced miner-specific status data. The public implementation and live-link changes are verified.
Score
Score compliance filter zeros multiple challenge elements
Miners reported zero scores at block 8,896,500 after a compliance_failed_tuple registry filter affected multiple elements. The incident is verified from the team channel; root cause, affected scope, and full remediation were still developing at cutoff.
engy
engy releases 0.4.5 to stop a chain connection leak
engy changed its validator to reuse one persistent chain connection instead of leaking one socket per tick, then published release 0.4.5. This is a verified validator-stability fix.
Leadpoet
Leadpoet burns all emissions while Lab 2.0 remains unfinished
The team said subnet emissions were being fully burned pending Lab 2.0. The burn state is observable, but no firm launch date or completed Lab 2.0 rollout was available at cutoff.
Gittensor
Gittensor makes native NVFP4 decode the default and detects poisoned CUDA cache
SparkInfer switched Qwen3.8 checkpoint-native NVFP4 decoding from opt-in to default and added detection and cleanup for poisoned CUDA compiler caches before evaluation builds.
Phylax
Phylax moves champion execution into a required jailed sandbox
Phylax changed champion runs to execute in a jailed sandbox and made the sandbox token mandatory instead of failing open. This is a verified security boundary improvement.
MVTRX
MVTRX launches SWAP on mainnet and patches scoring continuity
MVTRX announced its non-custodial SWAP interface on mainnet and shipped a hotfix intended to preserve scoring across simulation restarts. The launch and hotfix announcement are verified; downstream trading volume was not independently measured.
OpenRoboto
OpenRoboto alpha2 restores logs and refuses bad weight tables
OpenRoboto released 0.1.0a2 after fixing Bittensor-driven log suppression and adding fail-closed handling for invalid weight tables. It also added a v1 weights endpoint with fallback to the prior address.
KubeTEE
KubeTEE stops TaoStats rate limits from skipping validator cycles
KubeTEE cached TAO/USD prices, provided a writable cache home, and added CoinGecko fallback so a TaoStats HTTP 429 no longer skips the validation cycle.
cascade
Cascade finds a duel configuration that undertrained models
The team identified max_total_points set too low in chain.toml, causing heat and duel models to train for roughly 7,000 steps. The dashboard later self-corrected, but the exact number of affected contests and scoring consequences remained under review.
MicroTensor
MicroTensor delays Round 1 while hardening signed releases
MicroTensor kept Round 1 closed pending validator testing. In parallel it pinned and verified an Ed25519 release-signing key, read the configuration anchor from chain, set measured worker ceilings, and temporarily disabled provenance until the run store can accept miner runs.
Verathos
Verathos activates Sleipnir Mesh on testnet
Verathos activated Sleipnir Mesh on testnet netuid 405 for distributed serving of large models across multiple machines. The testnet launch is verified; mainnet availability and production reliability were not established.
Albedo
Albedo replaces scoring with a weighted reference-anchored rubric
Albedo shipped weighted reference-anchored scoring, a split rubric, simulation memoization, strict token checks, and provider token streaming. This is a material verified scoring-path revision.
BASE
BASE promotes one-B200 provisioning on split Lium hosts
BASE changed PRISM provisioning to rent one B200 on Lium’s eight-way split hosts, then promoted the image pins through staging and production. Deployment is verified by the production pin commit.
Beam
Beam dashboard outage blocks worker registration
Beam’s dashboard hit a database transaction error and returned HTTP 500 during worker registration. The team acknowledged remediation, but a full root-cause report and independent recovery check were not available at cutoff.
Ditto
Ditto ships a rapid benchmark-concurrency and copy-review repair cycle
Ditto released versions 0.99.9 through 0.105.0 while raising serial scoring timeouts, overlapping /run execution, isolating tool provenance under concurrency, tightening resubmission containment, reclaiming idle retest leases, and ranking lineages by their best score. The release stream is verified; the team’s ATH review remained operationally active.
sundae_bar
Sundae Bar ends a challenge after a hidden lookup-table exploit
The team terminated the challenge early after a leading submission embedded hidden lookup tables in dataset responses to bypass evaluation safeguards. The incident and early termination are verified from the team channel; final penalties and replacement challenge design were still developing.
No material update found: SN1, SN2, SN3, SN4, SN5, SN6, SN7, SN9, SN11, SN12, SN13, SN14, SN15, SN16, SN17, SN18, SN19, SN20, SN22, SN23, SN24, SN25, SN26, SN27, SN28, SN29, SN30, SN31, SN32, SN33, SN34, SN35, SN36, SN37, SN38, SN39, SN40, SN41, SN42, SN43, SN45, SN46, SN47, SN48, SN49, SN50, SN51, SN52, SN54, SN55, SN56, SN57, SN58, SN59, SN60, SN61, SN62, SN63, SN64, SN65, SN66, SN67, SN68, SN69, SN70, SN72, SN73, SN75, SN77, SN78, SN81, SN82, SN83, SN84, SN85, SN86, SN87, SN88, SN89, SN93, SN94, SN95, SN98, SN99, SN101, SN102, SN103, SN104, SN106, SN107, SN108, SN109, SN110, SN111, SN112, SN113, SN114, SN115, SN116, SN117, SN119, SN120, SN122, SN123, SN124, SN125, SN126, SN127, SN128.
Beyond one subnet
Ecosystem desk
Six high-performing wallets accumulate 383 TAO of Lium
SubnetRadar’s cutoff tape identified six top wallets accumulating a combined 383 TAO of SN51 Lium. This is a verified cluster signal under the tracker’s wallet methodology, not proof of coordination or inside information.
Bittensor-to-Base CCIP lane is operational
Chainlink’s live CCIP status page lists the Bittensor Mainnet to Base Mainnet lane as operational. This verifies the transport lane, not liquidity depth or application adoption on Base.
No material completed registration or deregistration is verified in the window
The moments, slot, and deregistration review found no additional completed slot takeover, major scheduled alpha unlock, exchange listing, alpha halving, or chain-level security exploit before cutoff.
Audit trail
Coverage notes
Searched
- All 128 active subnet slots and every August 22 public daily brief
- Exact public GitHub commits across 68 Bittensor-linked repositories from August 21 14:00 UTC through August 22 14:00 UTC
- SubnetRadar cutoff tape, moments, conviction, registration and burn views, plus TaoStats and fixed chain snapshots
- Official repositories, release records, Chainlink lane status, and public web and X indexing
- Earlier August reports for duplicate exclusion
Public-data gaps
- Public X and Discord indexing is incomplete; deleted, private, login-gated and rate-limited posts cannot be exhaustively reviewed.
- The August 22 dTAO snapshot repository stopped updating near 00:40 ET, so final market breadth and movers use SubnetRadar’s cutoff tape rather than that stale snapshot.
- Incident scope and remediation are Developing where only team-channel evidence or incomplete recovery information was available.
- No verified scheduled alpha unlock of at least 50,000 alpha was found.
- Empty update arrays mean no new material public signal survived fixed-window and duplicate checks; this is informational, not financial advice.