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 A past verified announcement can now be outdated.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.
Score Studio targets September 16 with SayGM routing
SayGM announced that Score’s image generation, training and model deployment will be available through its inference routing, with Score Studio scheduled for September 16. Score shared the announcement from its own account. Separately, Score plans public client and partner office hours next week to help miners improve results on private football and cricket tracks. These are announced next steps; the routing integration is described as coming soon.
A cross-subnet product integration and a specific launch date are stronger signals than an unspecified roadmap.
NIOME details its GENOME migration test — portal and Q4 rollout still ahead
NIOME published the design for a controlled GENOME-to-NIOME conviction swap. The first test would be capped at 5% of snapshot market capitalization, approximately $27,000, with a 100,000-GENOME limit per eligible wallet. The design escrows GENOME, delivers locked SN55 alpha and burns the matching GENOME after delivery is verified; failed delivery has a time-based refund path. Crucially, the same post says the portal is still being finalized and targets Q4 without a firm date. This is a migration plan, not evidence that the full migration or projected inflow has already happened.
A concrete migration mechanism can materially affect holders and subnet liquidity. Its unopened portal and conditional timing must be prominent.
Ridges opens Database Engineering competition submissions
Ridges announced that its next competition is live and accepting submissions. Agents will tackle database bugs, query optimization and work involving ORMs and GraphQL. The team is moving into another specialist engineering niche where results can be assessed on practical tasks. This is a current competition opening, separate from a generic roadmap or the older broad coding challenge.
A live, distinct miner competition changes the work and market being targeted by the subnet.
ChronoSeek becomes Witness: video-understanding research preview published
SN20’s identity changed from ChronoSeek to Witness on September 9. The new website describes agents that selectively inspect video frames and audio, then reconstruct events with a record of what they observed and what it cost. It shows recorded development traces and explicitly says public launch is pending validation, the dashboard is coming soon and the displayed examples are not live subnet telemetry. The name and research direction are confirmed; a fully launched mining service is not.
A subnet rebrand and new research direction are material. The primary website clearly limits the status to a preview.
Cascade schedules miner-funded rounds for September 11
Cascade’s updated source schedules the first miner-funded round at block 9,046,800, projected for September 11 around 20:30 UTC / 4:30 PM ET. Miners would fund their training leg using their own Lium account; the operator continues paying other defined parts of evaluation. The release also adds private submissions, while multivariate scoring is scheduled separately at block 9,068,400. The old September 4 funded-launch date passed without release. These are future block gates, and timing depends on the chain.
The switch changes who pays for training and how submissions enter competition. Miners need the corrected future activation date.
MVTRX clarifies what is live: SWAP and Hangar, with the trading venue still in testing
MVTRX’s first AMA recap says SWAP is live on mainnet and Hangar can deploy agents to managed infrastructure. It identifies the v0.6.1 scoring changes as testnet work, with mainnet targeted for the following week, and says its live trading venue remains in final testing on a private network using real prices. The new scoring direction emphasizes market making. Options and derivatives are a future ambition. This first-party account gives a clearer deployment status than the earlier aggregated release descriptions.
The new primary recap distinguishes usable products from an unfinished trading venue and explains a material miner-incentive change.
DSperse digest reports an unhealthy owner validator and patches in progress
A September 10 channel digest describes an owner-validator outage lasting 17 epochs and reports patches being deployed. The original conversation was not independently accessed. The current market feed shows a tiny positive subnet emission share, about 0.00002%, after yesterday’s zero reading; that does not demonstrate healthy request traffic or restored miner payouts. The previously reported v14.14.3 guard is not being treated as a new release today.
An apparent service interruption is important to miners, while recovery and payout status remain unconfirmed.
Pareton adds SGLang campaigns alongside vLLM
Pareton’s merged release lets miners submit optimization patches against either a pinned vLLM or SGLang baseline. The new launch guide includes a Qwen3.8-27B-FP8 campaign on one H200 and explicitly documents a zero-emission campaign setup. This expands the framework available to competition operators; it is not evidence that the example campaign is already paying rewards.
Supporting another major inference engine broadens the optimization work the subnet can run.
ORO publishes its generated-environment test pack; legacy agents need changes
ORO’s new release preparation adds a 35-task local qualifying pack spanning seven task families, a reference environment agent and matching runtime checks. Its miner guide says legacy ShoppingBench agents must adopt the generated-environment interface. This is a practical migration for builders following yesterday’s YC story. The public test artifacts and documentation are confirmed; they do not by themselves prove that all production competitions have switched over.
A new testing contract changes how miners build and qualify agents, while remaining distinct from yesterday’s accelerator news.
404 demonstrates a playable generated game and previews a creator Game Jam
404 showed a coastal racing game built with its generated assets and 3JS generator repository, using Fable 5 as the coding agent. The team says a creator Game Jam is coming, but has not supplied its launch date or prize terms in this announcement. The playable-game demonstration is current; the competition remains a preview.
A concrete creative demonstration shows how the product can be used, with a forthcoming creator event to watch.
Blockmachine adds Business Plus to its API plans
Blockmachine added a Business Plus tier for teams between its Business and Enterprise plans. It now describes six plans ranging from $9 to $999 per month, with its trace API and archive access on every paid tier. This is a concrete packaging change for users of the service; no new customer or revenue total was disclosed.
A new commercial tier is a practical product update, with current pricing attributed to the provider.
SayGM simplifies Azure and Foundry routing and checks deployment identity
SayGM removed an intermediate cloud-model routing component from its miner image. Requests now go directly to the existing encrypted upstream connections when deployments use the expected model names. The verifier checks those names and model formats, refusing startup or taking a worker offline when a named deployment mismatches. These are published routing and verification changes, separate from the upcoming Score Studio integration.
The update reduces routing complexity while enforcing which model a worker claims to serve.
Gradients documents a Friday 14:00 UTC tournament completion deadline
Gradients’ updated miner documentation sets Friday at 14:00 UTC / 10 AM ET as the expected completion deadline for each tournament’s starting week, regardless of training time, GPU availability or reverted rounds. A reminder fires 30 minutes before the deadline. Completion remains manual after the winning code is reviewed; the change does not automatically end a tournament or certify its winner.
The deadline directly affects miner planning and explains the continuing human review step.
RedTeam scoring delays remain under investigation
The September 10 channel digest reports payload-collection failures in headless-browser mode and says batch sizes were reduced to speed feedback. Miners describe lengthy delays and missing scores. We did not retrieve the underlying private conversation to verify each speaker or the repair’s effect, so this operational report remains developing. It is not evidence that a brand-new RedTeam subnet launched today.
Scoring delays can affect miner income and participation. The source and unresolved verification limits are explicit.
Harnyx adds DeepSeek-V4.1-Flash to its miner toolchain
Harnyx says every miner can now call DeepSeek-V4.1-Flash from its agents. A matching source change adds the model through its provider routing. The addition gives the research swarm another model option; it does not establish a measured improvement in Harnyx’s overall accuracy or cost.
A confirmed toolchain expansion is useful to miners, with no unsupported benchmark claim.
NOVA Blueprint digest reports a pause pending Boltz2 integration
The latest NOVA channel digest says the winning Blueprint submission’s source was published, clearer native-code and obfuscation rules are being prepared, and the competition is paused until Boltz2 integration. It also reports refunds for two later submissions. The direct Discord discussion and refund transactions were not independently inspected, so these remain developing operational claims. This is unrelated to approval of a drug or treatment.
A competition pause and possible refunds matter to participants; the secondary report is not presented as confirmed execution.
Herald starts Exploit’s press-distribution campaign
Herald says its Exploit Summit campaign has begun with a release appearing on Business Insider’s Markets site. The linked announcement promotes the Montreal conference on September 28–29. This is evidence of a press-distribution placement, not a claim that Business Insider independently investigated or endorsed the project.
A named distribution campaign is current customer work; the type of media placement matters.
Gittensor widens Muse’s checks and adds local-server API compatibility
Gittensor’s new source adds compatibility with LM Studio and Ollama REST APIs and broadens its Muse performance checks across more context lengths. A separate optimization reports approximately 1.18× faster decoding at 64K context using an int8 attention path. That figure is specific to the reported workload, not a general speed claim for every model or machine.
Compatibility and broader regression checks improve practical use of the subnet’s inference work.
UMI publishes its roadmap while bootstrap incentives remain a separate cutover
UMI’s post-AMA roadmap targets SN78 activation, benchmark work and private TestFlight through Q4, followed by a bitsign launch in the first half of 2027. New source adds validator supervision and an audited bootstrap worker. The temporary mechanism rewards eligible endpoint service equally rather than ranking translation quality. SN78’s observed subnet emission share is now 0.107%, versus zero in yesterday’s snapshot; that does not verify individual miner payouts. The September 9 “we are live” post linked the AMA event, so it is not being used as proof of incentive activation.
The roadmap and changing allocation are useful follow-ups, with the event livestream and miner-reward activation correctly distinguished.
Reliquary publishes V1 miner migration and revised submission allowances
Reliquary’s current README provides a V1 miner migration, a pinned image and a September 10 training-run identifier. Operators are told to preserve registered keys, miner state and model caches while switching profiles. New source work expands submission allowance to cover a complete collection window and improves checkpoint recovery. The README explicitly says configuration is not a go-live notice and retains runtime readiness gates, so a universal completed cutover is not established here.
A substantive miner migration and recovery changes affect participation. Published settings remain separate from observed production readiness.
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.
