What AgentBrain is
Persistent memory across sessions, goal tracking, structured evidence storage, confidence grades, source locators, red-team review and explicit human approval for public outputs.
Production process
Work can be delegated without delegating evidence control.
The process is designed to make public-facing AMR findings inspectable before they are published. Automated work can assist collection and synthesis; high-consequence claims remain bounded by evidence and review.
Agents can do work. Evidence still needs control.
1
Objective
Human-defined research question and success criteria.
2
Alpha
Research agent with memory, goals and evidence discipline.
3
Specialist workers
Parallel evidence collection, extraction and verification.
4
Evidence review
Structured binding of claims to sources with confidence grades.
5
Controlled synthesis
Review before any public-facing claim is made.
6
Publication human gate
No public release without explicit human approval.
Scope boundaries
The system does not turn weak evidence into institutional fact.
What this review is not
Not a real-time surveillance system, not a causal inference engine, not an automated AMR monitor and not a replacement for WHO GLASS or national reporting systems.
About this release
Public evidence review: methodology v2.1. Research base: ALPHA-MEGA-001 v1.0.1, frozen. Status: public evidence review. Contact: research@agentbrain.ch
Public evidence review: methodology v2.1. Research base: ALPHA-MEGA-001 v1.0.1, frozen. Status: public evidence review. Contact: research@agentbrain.ch