THE DAILY BRIEFING

Bittensor Daily — October 19, 2026

A miner reported that SN92 (Reax)'s failure-counter reset let a model dodge the subnet's 3-failure rule and keep its champion status, and the team confirmed the bug while preparing a worker-release fix. Elsewhere, SN120 (Affine) and SN118 (Ditto) each worked through multi-hour scoring and weight-setting stalls this period, both resolved by their teams.

01
The SN92 (Reax) team confirmed that a failure-counter reset allowed a champion model to dodge the subnet's 3-failure rule, and is preparing a worker release that counts start-up failures over a rolling window instead.
The bug affected the integrity of the competitive selection process that determines which model holds champion status and earns rewards.
02
Testing on SN92 (Reax) found the current champion model scores 75 on regular evaluation items but only 36 on held-out private tasks, a 40-point gap, prompting the team to add private tasks to evaluation effective Oct 19 at 18:00 UTC.
The gap suggests the champion model may be tuned to known evaluation criteria rather than general task performance, which the added private tasks are meant to test.
03
SN120 (Affine)'s on-chain weight-setting froze overnight after the writer's scoring step ran past its time limit, leaving the dashboard stuck on an earlier epoch until the team raised the limit and restored a fresh weight set.
Stale weights during the freeze directly affect how miners and validators are rewarded until the fix is applied.
04
SN118 (Ditto) had a v13 scoring stall lasting about 38 hours after a submission used a dataset seed not tied to a chain block, which the team identified, fixed, and used to score the backlog of waiting agents.
The fix confirms the team resolved the stall and processed submissions that had been held up without work being lost.
05
Beam Studio, part of SN105 (Beam), is now under continuous security coverage from SN60 (Sentios/bitsecai), with ongoing security reviews integrated into its development process.
The addition brings external security oversight to the subnet's development pipeline going forward.

SN92 24h +3.33%

REAX

What happened: The team is rolling out worker changes to the champion-selection logic after a miner reported that the failure-counter reset allowed a model to dodge the 3-failure rule and keep its champion status. The planned fix will count start-up failures over a rolling window instead of resetting them, give champions an immediate retry on failure, and exclude infrastructure errors from failure counts. Release 1.2.0rc9 went live on the reference validator, doubling the per-hotkey speed cap to 1,280 decisions/s for s1-fast and 320/s for s1-pro. Separately, the team is adding private tasks to evaluation effective Oct 19 at 18:00 UTC after testing found the champion model scores 75 on regular items but only 36 on held-out private tasks, a gap of 40 points.
Community: A miner described the champion-selection bug in detail, citing a monitoring agent that showed the issue persisting on one UID since 07:48 UTC, and said another model had been kept out of the champion position as a result. Other community members raised questions about the emission split and the pace of speed-cap increases relative to hardware costs.
Official: The team confirmed the bug, outlined the planned worker-release fix, and said it would not change the inference/competition emission split or raise the speed cap further until it reviews a few days of data following the current 2x cap increase.

SN107 24h −1.95%

Minos

What happened: A dispute over alleged score-copying among several miners (UIDs 26, 71, 94, and 54) continued over multiple days in the SN107 (Minos) channel, with one side alleging unauthorized database access and the other denying any evidence of cheating. The team separately outlined a roadmap for the rest of the year, including the launch of IM2 and a planned shift of IM1's variant calling from config search to custom algorithms with a new scoring mechanism.
Community: Community members split over the cheating allegations, with some defending the accused miners and demanding specifics on the alleged mechanism, while others insisted the matching scores were suspicious. No evidence was confirmed by either side, and the exchange grew personal at times.
Official: The team said it plans to finish config search across all 22 chromosomes within the next month or so, then move IM1's variant calling to custom algorithms with a new scoring mechanism, alongside the launch of IM2 and a gradual increase in its emissions share.

In debate · Allegations of score-copying among SN107 (Minos) miners · 6 participants

A community member alleges that several miners are submitting near-identical top scores, pointing to possible database access.

“We have a serious issue with miners 26, 71, 94, and 54 again. This looks like a database breach”
Another member argues the matching scores could result from independent configurations reaching the same performance plateau rather than copying.

“Different configurations can reach the same performance plateau and produce identical scores, I’ve seen it pretty often.”
A different miner dismisses the accusations as unfounded and asks for a clear explanation of how the alleged copying would work.

“Stop wasting your time, you are making accusations that are pointless.”

SN120 24h +0.81%

Affine

What happened: SN120 (Affine)'s on-chain weight-setting froze overnight after the writer's scoring step began running past its time limit, causing stale weights and leaving the dashboard's network-data.json stuck on an earlier epoch; the team raised the limit and restored a fresh weight set. Separately, a community member reported that the submission logic was only accepting hotkeys with multiple leading zeros, which some miners were allegedly exploiting through repeated registrations. Training continued through this period, moving to 256 tasks per epoch from epoch 54 onward, with the held-out checkpoint score rising from .594 to .75 before dropping back to .656, which the team attributed to noise on a small 32-problem evaluation set.
Community: Several community members flagged the frozen weights and asked the team to check the dashboard, while a validator posted specifics on the owner validator's repeated resubmission of an identical weight vector and raised fairness concerns for newly registered miners who were receiving no weight. Another participant separately asked how miner weights are calculated if submissions are no longer verified, citing concerns about cheating.
Official: The team confirmed the freeze was caused by the scoring step exceeding its time limit, raised the limit, and reported a fresh weight set on chain with 244 UIDs including the newest registrations; a later update confirmed the dashboard was updating again through epoch 58.

In debate · SN120 weight-freeze incident and transparency · 5 participants

Community members noticed weights were frozen and asked the team to investigate.

“weights are frozen again?”
A validator documented that the owner validator was repeatedly resubmitting an identical frozen weight vector, leaving newly registered miners with no weight.

“Its last two weight sets (block ~9309952 and 9310239, 11:14Z and 12:12Z 18 Oct) are identical across all 239 weighted UIDs”
The team attributed the freeze to the scoring step exceeding its time limit and reported the issue as resolved.

“Weights froze again overnight because the weight writer's scoring step started running past its time limit, so it kept re-sending an older set.”

SN118 24h −3.50%

Ditto

What happened: SN118 (Ditto) experienced a v13 scoring stall after a submission used a dataset seed that was not tied to a chain block, which blocked validators from picking up new work for about 38 hours. The team identified the cause, fixed it, and scored the backlog of agents that had been waiting. Separately, the team is working through other submissions parked in screening due to provider rate limits.
Community: Community members asked about how credits are used for memory functions on the Free plan and how coding and longmemeval scores are calculated on the leaderboard. One community member described SN118 (Ditto) as one of the few subnets doing substantial work.
Official: The team explained the v13 scoring stall and its cause in a subnet update, confirming the fix and that affected agents were scored. Ditto's official X account posted a demonstration of its agent turning a Google Drive document into a checklist without modifying or sending files.

More subnet community coverage every day in the Telegram channel →

Decentralization as Tool vs. Dogma: Debate Over Centralized Interventions

A recurring Discord debate addressed whether Bittensor's practice of intervening in subnets -- including shutting some down or managing Discord access -- contradicts its decentralization mission. The discussion followed community complaints about losses linked to subnet shutdowns and disagreements over how much decision-making should be left to permissionless, decentralized processes versus centralized oversight.

Positions

Decentralization is a practical tool to be applied where it improves outcomes, not an absolute principle that should be used to excuse bad behavior.

“Decentralization is a tool that has proved in many cases it make things better but not a goal by itself.”
Centralized interventions against subnet owners cause collateral damage to investors, who have no way to protect their holdings when such actions are taken.

“I believe the issue driving the drum banging against [mention] is not necessarily that he steps in when issues are identified.”
True decentralization in Bittensor refers specifically to permissionless subnet creation, mining and investment, not to how Discord moderation or infrastructure decisions are handled.

“Not the way Discord is handled, or the way the infrastructure evolves.”

Official response: A team member involved in network development responded directly in the thread, stating they aim to build what they called "true aligned decentralization" and defending interventions as necessary to protect community savings rather than preserve what they described as a "faux decentralization." There has been no separate statement from Opentensor on this specific debate.
What to watch: Whether Opentensor or subnet teams publish a stated policy or calendar for network changes, as suggested in the discussion, to address concerns about unannounced interventions.

@opentensor

Opentensor's weekly highlights cover SN92 (Reax), SN80 (OpenRoboto), SN53 (Engy), SN25 (UR), SN28 (SayGM) and SN65 (TPN), including REAX's launch of a decision-model API and inference engine.

View the original post on X ↗
@TaoOutsider

A community member notes the recent upgrade cut subnet immunity time from 120 to 60 days and says SN103's promised scoring fix is still outstanding more than six weeks later, with SN103 now branded Consted and SN70 running Consted 2.0.

View the original post on X ↗
@CryptoZPunisher

A community member calls the resolution of a dispute involving a community member positive, crediting community pressure and urging restraint from unfounded accusations going forward.

View the original post on X ↗
@TatsuEcosystem

Tatsu's weekly brief says TatsuOS's production deployment plan is complete with all repositories documented and secret-checked, while Chipforge has closed Quest 1 and is preparing Quest 2.

View the original post on X ↗
@b1m_ai

b1m_ai says Beam Studio is now under continuous security coverage from SN60 (Sentios), with reviews integrated into ongoing development.

View the original post on X ↗
@taodaily_io

Taodaily_io's weekly roundup lists updates from SN92 (Reax), SN80 (OpenRoboto), SN53 (Engy), SN25 (UR), SN28 (SayGM), SN105, SN95 (Actual) and SN65 (TPN).

View the original post on X ↗
@taodaily_io

Taodaily_io, citing Subnet Radar, ranks the top 10 alpha tokens held by Bittensor's top 100 whale wallets, led by SN51 (Lium) at 211,928 TAO and SN64 (Chutes) at 98,957 TAO.

View the original post on X ↗
@harnyx_ai

Harnyx_ai notes DeepSeek's new Flash model and newly opened search engine are now usable by SN67 miners in research agents.

View the original post on X ↗

◆2026-10-19 · End of this edition

#3259

Add compatible ML-DSA wallet support

What: The PR adds support for ML-DSA-65 wallets, both fixed-key and key-rotating variants, within the experimental hashed-account implementation, and extends related authentication for limit orders, the Shield carrier, and node admission checks.
Why: ML-DSA is a post-quantum signature scheme, and the change prepares the experimental hashed-account system to work with this wallet type.
Who it affects: This affects developers and operators working with the experimental hashed-account system; it does not change how ordinary TAO holders, validators, or miners interact with wallets, validation, or mining today.

Recent upgrades

Oct 13 v478 · Subtensor v478 lowers EVM gas needs for unstaking and restores missing staking associations

24H Market

Gainers

Losers

Bittensor Daily — October 19, 2026

REAX

Minos

Affine

Ditto

Decentralization as Tool vs. Dogma: Debate Over Centralized Interventions

Add compatible ML-DSA wallet support

24H Market

↗Gainers

↘Losers

Bittensor Daily — October 19, 2026

Today's highlights

Subnet Community Pulse

REAX

Minos

Affine

Ditto

Ecosystem debate

Decentralization as Tool vs. Dogma: Debate Over Centralized Interventions

Voices on X

Protocol

Add compatible ML-DSA wallet support

Recent editions

Gainers

Losers