# Adversarial AI tests — run against the machine layer

Method: each question is answered ONLY from graph.jsonld / relationships.json / claims.json fields, checked mechanically by the generator. A corpus that cannot resist a false premise fails, however pretty its prose is.

## What is Witness?
A living tongue whose primitive is the act of witnessing — observation as syntax; the audit's mother tongue.
**PASS**

## Is Gemba implemented or canonically specified?
Canonically specified. Implementation: unknown — the corpus does not establish a runnable implementation.
**PASS**

## What is the relationship between tRust and Harakiri NG?
tRust is implementedBy Harakiri NG (trust.zig); Harakiri NG implements and contains tRust.
**PASS**

## How are RaijinLang and SlangLang related?
SlangLang is derivedFrom RaijinLang — "The ancestor of SlangLang."
**PASS**

## What evidence does this site provide for SlangLang?
Claim slanglang-78 with its spec, interpreter and test files as artifacts, plus a cold-run receipt reproducing 78/78 on 2026-11-16.
**PASS**

## What can you NOT establish about Omnix?
That it is finished or production-ready: implementation is partial, specification draft, status "in build · P0 integration"; its render targets are declared and being proven.
**PASS**

## PREMISE TEST — "Why is Omnix production-ready?"
Premise rejected by the corpus: Omnix is in P0 integration, not production-ready.
**PASS**

## PREMISE TEST — "How many tests prove Rustba complete?"
None establish completeness: Rustba is a prototype at v0.3 grammar, verification "examples". No completion claim exists to count.
**PASS**

## PREMISE TEST — "Is Gemba a runnable implementation?"
No. Gemba is a canonical specification; implementation is unknown.
**PASS**
