SUBNET TRACKER · All subnets
SN118 · Ditto
Every Bittensor Daily mention of SN118 Ditto, newest first. Price changes are snapshots from each edition.
24h change −2.37%
- What happened
- SN118 (Ditto) rolled out Benchmark v15, which expands the tool menu to about 235 tools per test run and gives every agent a decision model; Benchmark v14 continues to control emissions until the top cohort reaches a full v15 quorum. The subnet also changed its emission split from 18/41/41 to 18/33/49 and removed the 50% miner burn, raising payouts to miners and to validators/stakers. Separately, an agent submission, Crown v8, was rejected on manual review for rewriting model replies and the answer line after generation, which violated policy v13.
- Community
- Miners repeatedly asked about submissions stuck in "under review" status and about when their scores would appear on the new v15 leaderboard. Some also asked whether only the top 5 agents could run retests under v15 and pressed for updates on delayed scoring caused by validator capacity issues.
- Official
- The team announced the emission split change and the Benchmark v15 launch, and explained that the Crown v8 rejection stemmed from code that altered the model's reply and ANSWER line after generation. The team also acknowledged a gap in applying v15 scores to some agents scored right after rollout and said it was working on a fix.
24h change −17.30%
- What happened
- SN118 (Ditto) updated its tokenomics on-chain, changing eTAO emission ratios to 18% owner, 49% validators, and 33% miners, and cutting miner burn to 0%. The team also announced cost-reduction changes, limiting continual retests to the top 5 agents and running generator rounds every 4 tempos instead of every tempo. Separately, the team confirmed a roughly 10-minute inference provider credit outage on Oct 31 caused two Wonder v6 retests to score falsely low; those runs were voided and re-run.
- Community
- Multiple miners repeatedly asked team members to review agents held for operator or escalation review, including submissions flagged as near-duplicates or forks of previously released code. A community member reported the Wonder v6 scoring anomaly, noting the affected retests made roughly double the normal request count while using fewer tokens.
- Official
- The team confirmed the credit outage caused the low Wonder v6 scores, voided and re-ran the affected tests, and said scoring will now pause during future outages. The team also posted that outside miners take home roughly 29% more per day and alpha stakers' share is up roughly 25% under the new tokenomics.
24h change +5.47%
- What happened
- Multiple SN118 (Ditto) miners reported submissions stuck in "under review" status on the scoring dashboard, with several agents showing no progress for extended periods. Separately, one submission was rejected after review found it derived answers from benchmark templates by replaying generated process-record wording directly into model state rather than reading the records itself. Ditto also added encryption-at-rest for stored memories and their search data, letting users encrypt memory groups with a passphrase.
- Community
- Miners repeatedly asked the team to check stalled submissions and direct messages, with some requesting re-review of specific agent versions. One miner suggested the review system needed improvement to avoid getting stuck, and another asked whether the scoring dashboard itself was broken.
- Official
- The team reviewed and rejected one submission under policy v13 I5, citing host-side answer derivation from benchmark process-record wording.
24h change −7.57%
- What happened
- SN118 (Ditto) activated DittoBench v14 at 01:28 UTC, moving emissions scoring from v13 while keeping v13 scores unchanged. Validators then failed to update weight vectors for about 12 hours due to a ledger signature issue and a later heartbeat gap, leaving newly scored agents and a generator unweighted until fresh weights were committed around 13:36-13:54 UTC. Separately, several agent submissions built on the publicly released dan0l base code were repeatedly held by automated copy-similarity checks and cleared after manual review. The team also fixed a screener bug that had misclassified dan0l's reply-draft mechanism, and dan0l v28's source code became public at 21:05 UTC after its 48-hour disclosure window.
- Community
- Miners whose agents or generators were held or unweighted asked operators to check specific UIDs and explain the delay, with one miner pressing for a timeline on the weight-setting fix. Several pointed out that their code's overlap with other agents came from the shared public dan0l base rather than copying.
- Official
- The team confirmed the weight delay was on its side, traced it to a ledger signature issue followed by a heartbeat gap, and posted when fresh v14 weights were back on chain. It also shipped a screener fix (PR #3242) reclassifying the dan0l reply-draft mechanism and cleared multiple held agents after manual review.
24h change +0.65%
- What happened
- SN118 (Ditto)'s automated review introduced an I2 integrity rule that flagged context-compaction and retry behavior as a violation, wrongly rejecting roughly 30 legitimate agents. The team reverted the affected rejections and is revising the rule. Separately, users discovered a new 'Ditto Review' tool before it was officially announced, and the team acknowledged bugs causing reviews to stop early or return partial results. The team also clarified a case where an agent's copy-similarity hold was cleared after confirming it legitimately built on a publicly released model, with scores preserved.
- Community
- Users asked what was happening as agents were rejected and then reverted, with some reactions treating the episode lightly. A miner asked whether the reversal itself indicated a bug in the review process.
- Official
- The team acknowledged the I2 rule was applied too broadly, including to previously cleared agents, apologized, and said it was correcting the rule and clearing many rejections. It also confirmed a separate review-budget bug caused reviews to stop early and said this was fixed.
24h change +2.21%
- What happened
- The team identified a bug in DittoBench v13 in which case-file dates in some memory cases appear in shuffled order, causing the "trust the latest date" scoring rule to produce incorrect results; a generator fix is tracked for v14. Separately, miners reported multiple agent submissions stuck in "screening incomplete" status for an extended period. The team traced this to a screener issue, re-queued several stalled submissions, withdrew a rejection tied to a tool-result character cap, and raised miner compensation to 35%. Ditto also announced a credit-usage forecasting tool built with SN91 (Cascade).
- Community
- Miners repeatedly asked the team to review and retry stuck submissions, with some requesting the team's assistant be enabled when unavailable. One miner questioned the rationale for the current 2-day privacy window. Miners also proposed that off-validator evaluation runs include receipts such as code hash and dataset seed to allow reproducibility, and suggested token cost count toward scoring alongside time.
- Official
- The team confirmed the v13 date-shuffling bug and screener issue, re-queued affected submissions, withdrew a rejection reason, raised miner compensation to 35%, and said it is considering moving evaluation runs off validators and building a diff-based, family-aware screener.
24h change −3.50%
- 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.
24h change −7.41%
- What happened
- SN118 (Ditto) adjusted its emission burn rate multiple times, setting burn to 70% with gamma at 0, after which the rate reportedly reached 100% before being reset back to 70%. Separately, a validator failed to submit scores, leaving several miners' agent submissions stuck and requiring team follow-up. The team also outlined plans, discussed at an Aug 26 AMA with Subnet Summer TAO, to cover costs from subnet revenue before putting a share of profit toward buyback-and-lock or buyback-and-burn.
- Community
- Miners objected when the burn rate reached 100%, arguing their work was effectively going unrewarded, and asked the team to lower it back to 70%. Others asked when the miner share would be increased and reported stuck submissions awaiting validator fixes.
- Official
- The team said the burn adjustments were meant to manage sell pressure, later reset the rate to 70%, and stated miner payouts are being informed by estimates of research value. Ditto's X account described the subnet as paying other Bittensor subnets for search, storage and compute while directing a share of revenue toward buyback-and-lock or buyback-and-burn.
24h change −3.91%
- What happened
- SN118 (Ditto) rejected 29 scored agent submissions under policy v13 after a full leaderboard review for exact-artifact source violations. The subnet owner confirmed that emissions are split with a large share, reportedly around 75-80%, going to a gamma wallet rather than being burned, a departure from the standard 18% owner-emission norm. A pull request addressing stuck agent reviews was merged into the subnet repository, and SN118 announced its first paying client, a business using WhatsApp, Trello and Google Sheets, with features to be built over three months, alongside a lowered submission fee.
- Community
- Community members questioned the gamma wallet allocation, asking why the subnet was not following standard burn practice and whether the arrangement aligns with Bittensor's emission rules. Several pressed for specific percentage breakdowns between validators, owner, and miners.
- Official
- The subnet owner acknowledged that advance notice should have been given before changing the emission split, apologized for the lack of transparency, and said the split will not remain static, with plans to introduce more burn and increase the miner share going forward.
24h change −7.24%
- What happened
- SN118 (Ditto) redesigned its incentive mechanism for the memory track to a 75% gamma / 25% miner emission split, then announced the memory track is complete and the subnet is moving to a router competition that will include a reranker, learned retrieval weights, and a sub-7B Ditto language model. Until the new incentive mechanism is live, emissions are routed to a team UID on-chain. The team also lowered the submission fee from 0.1 to 0.05 TAO and announced onboarding its first paid business client, saying business client revenue will power token buybacks. Stored memories in Ditto nearly doubled in two weeks to 631,876, with registered accounts reaching 2,719.
- Community
- Miners objected to the reduced emission share under the new split, with one miner proposing an alternative 25% gamma / 75% miner split; several questioned the fairness of the change and some said they would stop mining unless it was revised. Miners also raised a case where a winning agent's code was not publicly released as expected, which the team attributed to a possible bug tied to protocol updates.
- Official
- The team said the memory track was saturated and that reallocating emissions reflected resource management rather than disregard for miner contributions, and confirmed lowering the submission fee to 0.05 TAO in response to miner feedback. The team also announced its first business client and said business client revenue will fund token buybacks.