01 / Intake
The AI claim becomes a review record.
The system records what is being questioned, why it matters, who may be affected, and which technical or regulatory issue deserves structured review.
Structured multi-model debate method
Put every AI claim on the record.
Delta Cross-Examination is a structured debate method for AI governance in high-impact environments: formal sessions, technical contradiction, mandatory human oversight, redacted public summaries, and protection for sensitive material.
The method
A technical debate method inspired by the logic of cross-examination.
Delta Cross-Examination adapts the discipline of contradiction to AI: instead of accepting an automated answer as a final conclusion, the method opens a formal review session across models under mandatory human mediation.
Each session may generate minutes, a public opinion, a summary dossier, and a reviewable thesis. The methodological foundation starts from RAG DATA discipline and evolves into a public layer of governance, institutional reading, and reuse in future debates.
How it works
From AI claim to redacted publication in seven steps.
The workflow adapts RAG DATA discipline into governance: open the issue, gather essential evidence, establish contradiction, consolidate opinion, and publish only a responsible summary.
Sessions
Formal debate sessions already held and new issues in preparation.
The method has already been applied in three structured sessions with human mediation, multi-model opinions, public consolidation, forensic traceability, and the possibility of rectification whenever later technical review justifies it.
Human mediation and privacy
Opened on May 16, 2026, it consolidated the human gate, the distinction between AI contribution and institutional endorsement, and protection for source and code.
Temperature governance
Held on May 19, 2026, it addressed ideal temperature, private environments, parameter transparency, and responsible rectification after technical review.
Memory, context, and rectification
Held on May 30, 2026, it consolidated contextual rectification triggers, safeguards against weak inference, and cryptographic traceability for public artifacts.
Public summaries
Public summaries make each session legible, reviewable, and reusable.
Each session may generate a redacted public summary with context, conclusion, and next steps. This layer sustains continuity of debate for humans and agents without exposing the full internal environment.
Session summaries
Consolidated summaries work as institutional continuity for the method. They allow future sessions to rely on prior theses and already-debated criteria.
Accessible institutional debate
Public readers, partners, developers, and reviewers may read the public layer without access to the restricted environment.
Assisted coordination
Summaries and redacted documents may guide agents in controlled environments while sensitive material remains outside the public layer.
Orchestration environment
Antigravity CLI as the operational environment of the debate.
Sessions are conducted in a controlled environment, with CLI/API orchestration, defined technical roles, and mandatory human mediation. Antigravity organizes the operational flow of contradiction and technical model invocation.
Gemini layer
Gemini as a key contribution to coordination and public reading.
Gemini played a relevant role in the project's development cycle and in coordinating the public reading layer. The dedicated page gathers a redacted manifest, permitted documents, and institutional context for agent reading in controlled environments.
Protected public layer
Redacted publication without exposing internal material.
Delta Cross-Examination relies on RAG DATA discipline to separate what can be published from what must remain protected. The goal is to make the debate understandable without opening internal operational material.
Public content
The site publishes only the institutional reading of each session, with accessible language and defined scope.
Internal material
Credentials, prompts, code, operation files, and sensitive details remain outside the public layer.
Basis for new topics
Published conclusions serve as reference for new sessions, community questions, and institutional proposals.
Methodological and institutional base
RAG DATA, PrevBot, and software registry as the foundation of the method.
The method starts from the PrevBot / RAG DATA foundation and is linked to software registry INPI BR 51 2026 002804-3. Authorship, human coordination, and institutional development are led by Dr. Miriam Mesquita Reis within the Romanov Tesselação ecosystem.
INPI BR 51 2026 002804-3 as a methodological base linked to the method.
The final decision, publication, and mediation of the act remain under human responsibility.
Romanov Tesselação sustains the institutional and authorial home of the project.
Questions and community