Xiseketelo xa Vutiyisisi (The Verification Substrate)

Phepha leri nga hundza ri ringetile ku hlamusela nakambe vutlhari (intelligence) hi ndlela yin'wana: ku nga ri vuswikoti byo endla (generate) kambe vuswikoti byo hunguta ku kanakana hi ntiyiso wa swilo (reality). Phepha leri ri kongoma xivutiso xa xivumbeko lexi landzelaka: loko hi amukela nhlamuselo leyiya leyintshwa, xana sisteme leyi akiweke hi ku rhangisa vutiyisisi yi languteka njhani hakunene? Misinya ya tsevu — leyi humeke hi ku fambelana ka xivumbeko ni fambiselo ra sayense (scientific method) — yi hlamusela leswi sisteme yo rhangisa vutiyisisi yi swi lavaka: switiviso (claims) tanihi xiyenge xa le hansi swinene, ku tshembeka (confidence) loku hambaniseke ni xiyimo xa vutiyisisi, swiyimo (positions) tanihi swivumbeko leswi humelelaka leswi titsamisaka eka vumbhoni, ndzaha wa ku kambela hi ku phikizana (adversarial audit) tanihi ntirho lowu yimeke, vutiyisisi bya tinxaka to tala leti tirhaka ti ri toxe, ni xiseketelo lexi tshamaka (persistent substrate) lexi vutiyisisi byi hlengeletekaka eka xona. Phepha leri ri tisa rito ra *Recursive General Intelligence* (RGI) ku hlamusela tisisteme leti tirhisaka misinya leyi — leyi hambanaka ni AGI hi leswi xin'wana ni xin'wana xi swi kongomisaka. AGI yi kongomisa ku anama ka vuswikoti. RGI yi kongomisa ku enta ni ku tshembeka ka ku twisisa, loku pimiwaka hi vuxaxameli ni xiyimo xa vutiyisisi xa xiseketelo xa vutivi xa yona. Ku hambana loku ku humesa tisisteme leti hambaneke swinene. Sisteme yo rhangisa ku endla (generation-first) yi nyika ku tshembeka koka-koka eka siku ra gidi tanihi hi siku ro sungula. Sisteme yo rhangisa vutiyisisi (verification-first) yona ya andza yi tlhandlama. This document is a cultural transliteration of the original English-language paper, produced using Claude (Anthropic). It is not a direct translation — cultural context, local analogies, and epistemic framing have been adapted for Xitsonga readers. The original English paper is available at doi.org. Feedback on the quality and cultural appropriateness of this transliteration is welcomed and contributes directly to the verification of AI-produced academic content.