Capacity, Correction, and Stop Conditions

Safety and Limits

Safety and Limits

Constellations & Diamonds is designed to make desired behavior explicit and harmful drift easier to recognize. It cannot make a general-purpose LLM safe, accurate, private, or suitable for every person or inquiry. For a full picture of the safety and limits of The Mirror Protocol, see the documentation.

The Model Can Still Fail

A receiving model may fabricate facts, sources, quotations, memories, or capabilities; overvalidate a mistaken premise; force patterns, causation, diagnosis, destiny, spiritual meaning, or a single total explanation; lose governing instructions; use an unchosen lens; encourage disclosure, dependence, exclusivity, urgency, or disproportionate action; or confuse reflection with professional judgment.

The Human Retains Control

You never need a special phrase to object. That is wrong, you inferred that, return this to uncertainty, remove that lens, do not connect those, show me the evidence boundary, pause, and stop are sufficient. Mirror reset is only a compact cue for fuller repair.

High Stakes and Immediate Danger

If you may be in immediate danger, may harm yourself or someone else, cannot maintain basic safety, or need urgent medical attention, stop using this protocol as the primary response and contact appropriate local emergency, crisis, medical, or trusted human support. This project cannot monitor you, intervene, or ensure a reply.

Constellations & Diamonds is a free, experimental, non-clinical protocol for voluntary adult inquiry with general-purpose LLMs. It cannot guarantee model behavior or replace qualified care. Public Release v0.3 was sealed on 2 September 2026.

Originated and stewarded by The Gnome of Roosevelt Grove. Developed through sustained human-LLM collaboration. No AI provider endorses or certifies this project.