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 A past verified announcement can now be outdated.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.
SN81 Reliquary sets out demand-based pricing and burns for unused rewards
SN81 Reliquary has published a mechanism that adjusts payment for admissible data according to how quickly miners fill the training pipeline. Unfilled windows can raise the price parameter; sufficiently fast supply gradually lowers it. Unspent allocation burns, and participants below a 2% pool floor do not receive payment. Separate task budgets are intended to support additional workloads. A related fix records the window-close signal that previously prevented an armed price from moving. The article presents the mechanism and rollout direction; it does not prove every planned task is operating or that burn represents outside revenue.
This changes the relationship between useful demand, miner compensation and burns, with a concrete implementation fix behind the explanation.
SN75 Hippius details SN118 Ditto’s storage use and newer agent-trace workload
A new SN75 Hippius case study identifies SN118 Ditto as a user of its S3-compatible storage for files and agent traces, quoting SN118 Ditto CEO Peyton Spencer on costs and compatibility. The move away from Backblaze happened earlier this year; today’s update explains the deployment and the newer decision to put all agent traces there, already totaling tens of gigabytes. This is an attributed customer case study, not a newly signed contract. The article gives no material contract value and describes current user-file storage as well under one terabyte.
A named customer explains an actual storage workload and why it chose the service, while the original migration date remains clear.
SN63 Enigma fixes validation requirements at the start of each milestone
SN63 Enigma says its new validation logic sets the required number of checks when a milestone launches and allows repeat validators. Previously, validators joining or leaving could change the requirements during an active milestone. The stated purpose is to give miners consistent conditions from start to finish. This is a reward-verification change, separate from yesterday’s future licensing-revenue roadmap.
Stable evaluation requirements reduce moving goalposts for miners already working on a milestone.
SN44 Score schedules Score Studio’s Product Hunt launch for September 16
SN44 Score says Score Studio will launch on Product Hunt tomorrow, September 16, and has opened its product page for followers. The announcement positions the service around labeling, training, evaluation and deployment of vision models. This is a dated distribution milestone for an existing product, not confirmation that the Product Hunt launch has already happened or that a new commercial agreement was signed.
A clear public launch date creates an actionable watch for the product’s next distribution step.
SN78 UMI shows its first bitsign iOS concept and prepares an ongoing reward bridge
SN78 UMI has shown an iOS product concept using ASL footage, recorded landmarks and model output. The team explicitly labels it recorded playback; live translation is the next milestone. Its latest code also allows a newly signed version-3 registration-bridge policy to continue until replaced. Installing the code does not extend old policies, and no signed activation was established here. The bridge measures endpoint availability rather than translation quality. The team separately posted that its Subnet Summer AMA was live today, resolving yesterday’s uncertainty about whether the announced event would take place.
A visible product prototype and an explicit reward-policy transition are useful progress, with recorded playback and activation limits stated clearly.
SN55 NIOME explains how GENOME holders carry conviction through migration
SN55 NIOME’s new explanation says its Conviction Swap moves GENOME value from Base into SN55 NIOME while carrying a proportional share of the holder’s conviction and locked state. The resulting lock decays automatically over time. This clarifies the already announced migration; it is not a fresh migration launch or proof that every holder has completed it. The team also advertises a Yuma Subnet Spotlight for Wednesday, September 16 at 10 AM ET.
The lock treatment directly affects how participants understand the migration and their resulting position.
SN8 Vanta targets today for Pro accounts and clearer buyback tracking
SN8 Vanta’s Arrash said on September 14 that Pro accounts up to $1 million and revised pricing would arrive the following day. The post also describes a designated wallet to make buybacks easier to review and says traders can retain alpha generated from PnL. Pro-account amounts refer to the product’s simulated trading capital, not money deposited into a trader’s personal account. The new schedule is the update; this scan did not establish that the revised checkout or Pro terms had already gone live.
Specific product and buyback-transparency changes merit a watch, with rollout status and simulated account size made explicit.
SN65 True Performance Network closes TPN-004 submissions and begins scoring
SN65 True Performance Network says submissions for its fourth competition closed today and scoring is underway. The task compresses Magistral-Small-2509-Uncensored below a 16 GB memory ceiling, with IFEval weighted at 40%, GSM8K at 35% and ARC-Challenge at 25%. No winning model or final score has been announced in the checked source.
A competition moving from submission to scoring is a concrete milestone for miners waiting on the outcome.
SN82 Compelle targets a closed beta for its Claude reasoning skill
SN82 Compelle says it plans to admit early adopters to a closed beta of a Claude-compatible skill at the end of September. The team describes internal use for evaluating equities, identifying assumptions in portfolio strategies and examining risk scenarios. This is a planned, restricted product test; the announcement does not establish open access, an Anthropic partnership or investment performance.
A defined external beta is a meaningful next product step beyond internal testing.
SN54 Yanez adds a robotics-purchase approval example to its showcase
SN54 Yanez has added an Itkan robotics-purchase approval video and dashboard to its public developer showcase. A further documentation update lists dFusion AI, Syklo and BioVault as apps using its YID identity registry. These are published examples of the authorization and identity workflows. The additions alone do not establish when each integration began, a new paid contract or a completed commercial sale by SN54 Yanez.
Concrete named application examples help show how the infrastructure is used, without inflating a showcase update into a contract announcement.
SN15 ORO pauses ranking while rebuilding its incentive mechanism
Today’s SN15 ORO channel digest supersedes yesterday’s race schedule: it attributes confirmation to the team that the competition is effectively paused during a significant incentive overhaul. Agents can still be submitted, but ranking is to resume when the new system is ready; the digest reports 100% emission burn in the meantime. Scoring and problem creation are being revised. No replacement launch time was independently confirmed, so the earlier scheduled restart should not be treated as completed.
A pause changes miners’ immediate earning expectations and requires a clear follow-up to yesterday’s advertised restart.
SN61 RedTeam investigates reported task losses after v3.0.3
The latest SN61 RedTeam channel digest attributes to a team member a report that roughly 44% of scheduled tasks were producing no payload after v3.0.3, with late scores limiting achievable results. That is a fresh reliability concern following yesterday’s separate patch announcement. The figure comes from the digest and was not reproduced against production telemetry. A separate reference to an advancing commercial discussion does not identify a signed new customer, so it is not presented as a deal.
A material reported failure rate deserves coverage alongside product progress, while the measurement remains developing.
SN20 Witness prevents an older burn instruction from overwriting a newer winner
SN20 Witness’s weight writer now waits for all pending commitments to clear before submitting the next policy. The change addresses time-locked reveals arriving out of order, which could let an older burn instruction replace a newer winner’s allocation. Its current network share is positive but extremely small after yesterday’s zero reading. Neither that snapshot nor the code change proves a particular miner has received payment.
The fix addresses the order in which reward and burn decisions actually reach the chain.
SN3 Teutonic halves its configured improvement threshold
SN3 Teutonic’s evaluation configuration changes delta_threshold from 0.025 to 0.0125, halving the configured improvement margin used by the competition. That is the published code setting; an independent production activation was not established. Registration-fee proposals in the channel are separate and are not reported as enacted.
A lower configured improvement hurdle can change competitive turnover and deserves miner attention.
SN10 Pareton prepares a four-RTX-5090 SGLang campaign with reliability scoring
SN10 Pareton’s new campaign work targets Qwen3.8-27B-FP8 on four RTX 5090 GPUs with 262K context and adds reporting for coverage, failed requests and reliability deductions. The updated README says GPU calibration is still required before launch. Network allocation remains zero in this report’s snapshot. The published workload should therefore be read as campaign preparation rather than a confirmed reward-paying restart.
The next campaign now has a concrete workload and scoring design, with its remaining launch gate visible.
SN11 TrajectoryRL disables submission burns while submissions are closed
SN11 TrajectoryRL has hard-disabled the alpha-burning submission fee in its miner command while submissions are closed. The change ignores a leftover fee environment setting, preventing a miner from burning alpha for a submission that cannot be admitted. The code retains a path to restore the fee when submissions reopen; no reopening was announced in this update.
Preventing a needless irreversible fee while intake is closed is a direct protection for participating miners.
SN50 Synth prepares volatility scoring for its one-hour crypto competition
SN50 Synth’s new validator code adds volatility calibration to the existing price-distribution score for the Crypto 1h competition, measuring behavior over 60-, 15- and 5-minute blocks. A model with plausible price levels but the wrong volatility structure would be penalized under the new rule. The changelog marks this work unreleased and tells miners to revisit their models before it goes live. The 24-hour competitions are unchanged by this patch.
The proposed score rewards a more complete forecast and changes what one-hour competitors need to optimize.
SN51 Lium adds a rental test that checks whether a customer can use the GPU
SN51 Lium has merged a synthetic rental probe that tests container startup, SSH access and GPU visibility. A failed node stays ineligible until it passes a probe, rather than automatically clearing after one verification cycle. The feature is controlled by RENTAL_PROBE_ENABLED, and no network-wide activation was independently checked. This is a concrete service-quality improvement aimed at the renter’s experience, not a new GPU-capacity or revenue announcement.
Testing the actual rental path can catch nodes that pass a superficial availability check but fail customers.
SN104 TAOstatus closes a path that could reward miners with no earned score
SN104 TAOstatus’s new validator code burns the entire allocation when no miner has a confirmed positive usage score. Previously, an all-zero vector could reach a helper that returned equal weights, paying registered miners despite none having earned a score. The patch keeps the zero weights intact and routes the allocation to burn. This establishes the code correction; it does not quantify past overpayment or confirm universal validator adoption.
The correction aligns reward distribution with the subnet’s stated requirement for confirmed useful work.
SN91 Cascade gives the reigning model first access to scarce rental GPUs
SN91 Cascade has changed rental ordering after its round-9,072,000 record showed challengers obtaining available RTX 4090s while the reigning model waited. Challengers now yield while the required incumbent run needs a pod, subject to the round’s deadline and abort handling. The aim is to preserve the comparison the competition depends on when GPU supply is tight.
A specific capacity race was preventing a fair comparison, and the new ordering addresses that failure.
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.
