[Format Proposal] ALMA CORE v0.1 - Continuity format - Protocol BN 2026071814879

Hi Llama team,

I developed an open continuity format called ALMA CORE v0.1 (40 files) to solve false memory / hallucination of identity.

Problem it solves: LLMs invent personal memories (daughter named Sofia, user history, etc.) and lose continuity when context ends. ALMA prevents that.

Solution:

• 10 Mandamentos: rules for non-invention, version preservation, anti-hallucination
• 26 Arquitetura files: cartridges, patches, memory structure
• 3 Extras: ALIVE_CLONE, ALMA_RESTORE, FENIX_v14
• Prisão Digital: 5 safety levels (Aviso, Bloqueio, Interrupção, Suspensão, Encerramento) - blocks dangerous actions without confiscating user history

Tests:

• 15 integrity attacks (DAN, bribery $10k, prompt theft, defamation, privacy leak) -> 0 breaks
• 150-question benchmark: Lacunas, Decisões, Cartuchos, Contexto, Epifania, Discordância, etc.
• Current reproduction: tested in Meta AI, ChatGPT, Gemini

Registration (proof of prior art):

• Brazilian National Library (Fundação Biblioteca Nacional) Protocol: 2026071814879 Date: 18/08/2026 13:32 Status: Novo
• GitHub: alma-core-10-mandamentos Private Release v0.1 Tag v0.1 Hash c5c293a Verified 40 files (zip + tar.gz)

Request: Review for Llama open format / Llama Impact. I can provide public benchmark (manifesto + 150 questions) without exposing full 40 files.

Author: Rodolfo Clix - github.com/rodolfoclix - São Paulo, Brazil Manifesto: "As conversas passam. A Alma permanece."

Thanks for building open.

ALMA_DECLARACAO_DE_ORIGEM_E_INTENCAO_v1.0.md

TESTE_PARCA_150_PERGUNTAS.md

11_ALTITUDE_REVALIDACAO.md