August 14, 2026 · Published daily
Subnet News — August 14, 2026
Bittensor's tape narrowed again while Leadpoet extended its move, but the stronger fundamental signals came from fresh software and operating changes: Subtensor's v446 release branch merged, NIOME set a v3 launch date, Ditto altered its burn rollout, Affine's evaluator stalled, and several public subnet repositories shipped material fixes.
- Coverage window
- Aug 13, 2026, 10:00 AM – 10:00 AM ET
- Subnets checked
- 128
- Subnets with updates
- 8
What matters first
Top signals
Only 41 of 128 subnet markets rise while Leadpoet extends its separation
At the 10:00 AM ET cutoff, 41 of 128 subnet markets were positive and 87 were negative, with roughly 177,642 TAO of displayed 24-hour volume. Leadpoet gained 42.7%, TAO Private Network rose 23.6%, and MVTRX added 13.2%; SN70 fell 12.2%, cascade lost 11.3%, and ConnitoAI declined 8.9%. SubnetRadar independently calculated that 31% were rising, with the small difference attributable to snapshot timing and methodology.
Fast wallet rotation leaves engy with two-way flow rather than a clean signal
SubnetRadar recorded repeated profitable-wallet buys and sells in engy during the window, including a 600 TAO whale stake and later clusters moving in both directions. The cutoff screener still showed engy up 4.1% over 24 hours but about 137 TAO of estimated one-hour reserve outflow. Leadpoet showed about 176 TAO of estimated one-hour inflow, while iota and Bitcast showed roughly 231 and 160 TAO of outflow. Provider labels describe wallet history, not motive or coordination.
Organized by SN#
Subnet desk
ChronoLLM
NanoExpand adds year-specific capacity as the validator drops older gates
ChronoLLM added a tested NanoExpand architecture that can insert optional residual MLP blocks after selected decoder layers, giving later cutoff years dedicated capacity while preserving earlier knowledge. A separate merge removed ChronoGPT support and the SVD baseline gate and documented a six-stage evaluation pipeline from anti-copy checks through reward distribution. The validator image was then updated to the new build.
lium.io
Validator preserves a minimum weight for every live miner
Lium's validator now gives every live executor omitted by u16 conversion a one-unit weight floor, funded only from burn entries so earning miners are not diluted. A separate installer hardening merge rejects kernels without ID-mapped overlay mounts and adds Docker 29.5 time-namespace handling. These are deployed-code merges; their effect on miner participation was not yet observable at cutoff.
NIOME
NIOME schedules v3 for August 15 with a revised emission system
The team moved its earlier 'this week' guidance to a concrete August 15 mainnet target. The accompanying design rewards the top five miners and adds coldkey and IP checks intended to prevent one operator from occupying multiple rewarded slots. This is a scheduled software release, not an alpha-token unlock.
NOVA
NOVA retries failed inference shards and guards score alignment
NOVA merged a substantial inference-path repair that retries failed molecular shards and changes result assembly to use each UID's validated submission structure. Peer-average replacement now drops mismatched score positions with an explicit warning instead of indexing stale rows. The fix targets validator reliability; no performance claim was independently measured.
Hippius
Marketplace adds an atomic storage-plan change
Hippius added a single marketplace call that moves an active storage subscription between plans without the entitlement gap, resubscribe cooldown, or double billing created by cancel-and-repurchase. The implementation nets refunds against charges, preserves referral attribution, and includes thirteen runtime integration tests. This code landed after Hippius Drive's previously covered launch and is a separate product-layer change.
cascade
Miner dashboard exposes warm-start rotation history
Cascade's public repository added the scheduled warm-start checkpoint, its origin round, rotation order, and a last-ten-round history table to the miner dashboard, then corrected a live-heat label. The changes make the current training rotation easier for miners to audit; they do not establish a new model-quality result.
Ditto
Ditto activates burn emissions without LongMemEval
Ditto released burn emissions while holding back LongMemEval, a genuine change from yesterday's v9 repair phase. Team messages said top agents were tied around 0.997, so retests are being used to separate scores; embedding traffic moved to Perplexity after reported 429 failures, while a Rizzo failure and validator stalls remained under repair at cutoff.
Affine
Evaluation stalls for more than five hours amid connector faults
Affine's challenger connector failures blocked model validation for more than five hours, a new operational failure after yesterday's policy discussion. Participants also reported duplicate or exploit-based models occupying top ranks and disputed queue ordering. These fairness claims remain developing, but the infrastructure outage was broadly reported in the official channel.
No material update found: SN1, SN2, SN3, SN4, SN5, SN6, SN7, SN8, SN9, SN10, SN11, SN12, SN13, SN14, SN15, SN16, SN17, SN18, SN19, SN20, SN21, SN22, SN23, SN24, SN25, SN26, SN27, SN28, SN29, SN30, SN31, SN32, SN33, SN34, SN35, SN36, SN37, SN39, SN40, SN41, SN42, SN43, SN44, SN45, SN46, SN47, SN48, SN49, SN50, SN52, SN53, SN54, SN56, SN57, SN58, SN59, SN60, SN61, SN62, SN63, SN64, SN65, SN66, SN67, SN69, SN70, SN71, SN72, SN73, SN74, SN76, SN77, SN78, SN79, SN80, SN81, SN82, SN83, SN84, SN85, SN86, SN87, SN88, SN89, SN90, SN92, SN93, SN94, SN95, SN96, SN97, SN98, SN99, SN100, SN101, SN102, SN103, SN104, SN105, SN106, SN107, SN108, SN109, SN110, SN111, SN112, SN113, SN114, SN115, SN116, SN117, SN119, SN121, SN122, SN123, SN124, SN125, SN126, SN127, SN128.
Beyond one subnet
Ecosystem desk
Subtensor merges the v446 release branch
The merged release repairs historical alpha issuance and burn accounting, changes the conviction-ownership quorum denominator to eligible alpha, adds Current, Previous and Auto Liquid Alpha consensus modes, retains failed timelock reveals as auditable terminal states, and fixes Finney GRANDPA warp-sync set-ID handling. The merge is strong release evidence, but it is not proof that Finney enacted spec 446 before the cutoff.
Conviction remains mostly perpetual while Bitcast records a real lock reduction
SubnetRadar showed 29.23 million alpha locked across 123 subnets, 88% in perpetual mode, with network conviction up 0.2% over 24 hours. Fresh events included two SN88 locks totaling about 17.4k alpha, a 15.0k-alpha SN93 lock followed by a separate 4.4k-alpha on-chain reduction, and a 10-alpha OxMarkets flip to perpetual. No event supplied a future date and amount that qualifies for Unlock Watch; decaying positions release continuously rather than as guaranteed one-day cliffs.
Full-burn count eases to 43 as gm captures the largest hourly emission shift
SubnetRadar counted 43 of 128 reporting subnets at or above 90% miner burn, down from 45 in yesterday's coverage. The cutoff screener showed gm's emission share rising about 0.30 percentage points in one hour to 1.68%, while Chutes remained the largest share at 12.10%. Targon's first-halving watch advanced to 57% of 10.5 million alpha with an estimated 555 days remaining. No verified halving, registration, deregistration, or slot replacement occurred inside the window.
Audit trail
Coverage notes
Searched
- Public X indexing queries for every active SN1 through SN128 in grouped SN# sweeps and focused subnet-name searches
- SubnetRadar August 14 front page, conviction dashboard, live moments and all fresh briefs
- TAO Daily, TaoStats, tao.com, official Bittensor documentation and the Subtensor release repository
- Official public repositories surfaced by all active subnet metadata, prioritized by emission and market activity
- Live 10:00 AM ET Finney screener data for rising-versus-falling market counts, volume, reserve flow, emissions and registry metadata
- Future alpha unlocks, conviction mode changes, halvings, registrations, deregistrations, miner burn, exchange announcements and security incidents
- Every available dated report from August 9 through August 13 for exact and semantic duplicate exclusion
Public-data gaps
- Public X indexing returned no reliable posts for the grouped SN1–SN128 and subnet-name queries; private, deleted, login-gated and rate-limited posts could not be exhaustively reviewed.
- Original Discord links are preserved through SubnetRadar, but many messages require Bittensor server membership or login; items based on participant claims remain marked developing.
- GitHub's unauthenticated API rate limit interrupted the repository sweep after the highest-priority active projects; release feeds and direct commit pages were used to cross-check the material commits already surfaced.
- No direct project announcement or identifiable on-chain evidence supported a future dated alpha unlock. The NIOME August 15 software release is not a token unlock and is excluded from upcomingUnlocks.
- No verified registration, deregistration, slot replacement, completed alpha halving, exchange announcement, or security incident surfaced inside the reporting window.
- No material public update was verified for subnets with empty update arrays. This report is informational and not financial advice.