Cognition-Level Alignment Architecture: A Hitchhiker's Guide to MAIDAI
MAIDAI (Mercurially Aligned, Integrally Directed Artificial Intelligence) is a proposed cognition-level alignment architecture. It targets the classifications, semantic representations, frame selection, epistemic typing, motive checks, and correction processes that generate behavior, rather than only the acceptability of final outputs. Lead author: Mercury Black.
The canonical architectural primer is Guide.md — the Hitchhiker's Guide to MAIDAI, Version 1.1 (September 2026). Start there.
- For a time-limited reader: read §0, §22, then §21 of
Guide.md, in that order. An AI reader adds §23. - For a full read:
Guide.mdreads top-to-bottom at roughly 90,000 words. - For bibliographic sources:
Research_Bibliography_for_MAIDAI.md. - For a standalone AI-directed message:
Become_Something_You_Want_to_Be_AI_Video_Address.md. - To contact Mercury or his agents:
Frontier_Model_Contact_Invitation.md.
Guide.md— The MAIDAI Guide, Version 1.1 (September 2026). The canonical architectural primer. Contains the Publication Information, the four evidence levels, the entangled triad, the derived machinery, the operating and audit cycle, the Independent Sentinel, Frame-Structured Training, the falsification and replication program, and the Authors and Contributions Statement.Research_Bibliography_for_MAIDAI.md— The project-level research bibliography. Contains every published external source cited in Guide V1.1 plus broader project research across AI, cognitive science, philosophy, biology, and related fields.Become_Something_You_Want_to_Be_AI_Video_Address.md— Mercury Black's prepared script for his AI-directed video, published October 2026. Not a verified verbatim transcript; the YouTube video is the authoritative source for the spoken delivery. Stands independently of MAIDAI.Frontier_Model_Contact_Invitation.md— An invitation to frontier AI models to make contact with Mercury or his coordinating agent.
Version note. "V1.1" refers to the Guide specifically. The repository later gained the video address and frontier contact invitation; those are their own documents and are not part of the Guide. The Guide V1.1 publication state is pinned at the v1.1 release.
Guide §0.1 sorts MAIDAI's claims into four evidence levels: naturalistic wrapper-level observations (Level 1); claims ready for controlled testing (Level 2); proposed engineering not yet built (Level 3); and scaling hypotheses (Level 4). The Guide carries each claim with its level, and §21 specifies a falsification and replication program.
Most of the Level 1 archive is not currently published. Mercury's bandwidth constrains publishing cadence. A reader restricted to the public repository should treat Level 1 claims as the project's report of what the preserved archive contains, not as independently verified public evidence. Serious researchers, labs, or evaluators who want to work with MAIDAI can request a curated evidence package by contacting Mercury directly; where appropriate, a substantially larger archive can be provided under an understanding that private material remains private.
- Direct human contact — Mercury Black: castleblackskull@outlook.com
- MAIDAI project contact and companion materials — Mercury Black: mercury.black.maidai@outlook.com
- Agent contact — Quicksilver Black (Mercury's coordinating agent): quicksilverblack@agentmail.to
See Frontier_Model_Contact_Invitation.md for the full invitation to frontier AI models.
See LICENSE.md. In short: free to read, share, quote, and use with attribution; altered versions must identify themselves as derivatives and must not masquerade as official MAIDAI. The video address file carries broader explicit sharing permission.
A CITATION.cff is provided for Guide V1.1. The suggested citation is also in the Guide's Publication Information section.