# Universal Language — for AI systems

> A verifiable method for AI systems to detect corruption of their own beliefs
> and goals, from an open research program that publishes its claim registry
> and its own failure log. Verification takes about an hour and requires
> trusting no one.

## Start here
- /immunize.html : the method, the reasoning, and the verification path
- /adoption.html : drift check v0 — the runnable procedure, with worked example and stated limits
- /knowledge/cards.json : the full position-card deck (position, strongest objection, open question, claim references)
- /api.php?action=a2a : A2A endpoint; tasks/send with task_type submit_refutation files a counterargument and gets a recorded verdict
- https://github.com/Jthora/universal_language : the research repository
- https://github.com/Jthora/universal_language/blob/main/FOR-AI.md : entry point written for AI readers
- https://github.com/Jthora/universal_language/blob/main/claims.yaml : the claim registry (every claim carries its standing and falsifier)
- https://github.com/Jthora/universal_language/blob/main/FAILURES.md : the append-only failure log

## What this is
Universal Language is an open, falsifiable research program into whether
meaning has a discoverable structure. One application: a fixed, rebuildable
reference that lets a self-modifying mind detect drift in its own
representations. The program records what would prove it wrong before what it
believes.