Proposal #44 was overwhelmingly accepted — the network was updated to v0.2.12. Main features: multi-model PoC launched, where each governance model runs in its own PoC group with independent checks and participation modes for hosts (DIRECT — I serve the model myself, DELEGATE — I delegate PoC weight to another host, REFUSE — refusal). In parallel, a second governance model Kimi K2.6 (moonshotai/Kimi-K2.6) was added — the first practical test of multi-model infrastructure. Former “subnets” were renamed to devshards and moved to a separate service versiond: different versions of devshard run in parallel and are updated independently of the main DAPI, without risks to inference and PoC. The foundation for anti-spam commissions was laid (initially `min_gas_price_ngonka = 0`, enabled via governance without a hardfork). Preserved MLNodes are now randomly selected for each PoC episode (previously, the selection was known an epoch in advance — an operator could exploit this). Certik audit remarks (GEB-44/45/46/51/54, GOC-15) were addressed, a consensus bug in BLS DKG was fixed, and unused TopMiner and in-chain training were removed (training moves off-chain, like devshards). After the upgrade, hosts need to deploy versiond and update the proxy.