Most AI systems today implicitly assume that:
if an expression is received, it should be processed.
This assumption hides a structural gap.
In many real systems, there is no explicit stage that decides whether a claim is allowed to enter an AI-controlled decision path.
Once input is accepted, it is pushed directly into reasoning, generation, or execution.
This creates systemic issues:
claims cannot be explicitly refused,
failure is conflated with runtime error,
rejection lacks semantic explanation,
and admission boundaries are not auditable.
The Admission Layer
EDCA introduces an explicit Admission Layer.
The admission layer does not reason.
It does not generate outputs.
It does not execute actions.
It answers a single question:
Is this claim allowed to enter the AI-controlled system?
If a system cannot clearly reject a claim,
it cannot safely accept one.
EDCA Admission Protocols
EDCA Admission Protocols define formal, admission-layer specifications for EDCA OS.
They describe:
how a claim is started (boot),
how it becomes a decidable instance (instantiation),
under what constraints it may proceed (runtime),
and how it may be rejected or fail (failure semantics).
All admission decisions occur before any reasoning or execution.
This repository contains normative specifications only.
It is not an implementation, runtime, or product.
Alignment Protocol v3.0
The first published specification in this family is Alignment Protocol v3.0.
It defines how human-originated claims are legally admitted into EDCA OS.
This is not a prompt technique and not a model-alignment method.
It is an admission contract.
A Structural Shift
The publication of EDCA Admission Protocols marks a shift:
from implicit acceptance
to explicit admission decisions.
This is not an optimization problem.
It is a boundary-definition problem.
Repository
GitHub (public specification):
https://github.com/yuer-dsl/edca-admission-protocols
Top comments (0)