Reasoning can propose a consequence. It does not establish authority to cause it.
THE SYSTEM CAN KNOW.
THAT DOES NOT MEAN
THE SYSTEM MAY ACT.
EABA is the architecture between knowing and doing. It requires a consequential system to prove that authority is present, evidence is admissible, the exact target is bound, conditions have not materially changed, and the protected commit is still authorized now.
One architecture. Three inspectable records.
EABA now has a permanent governed identity, a canonical public archive, and a public explanatory article. Registration preserves identity and boundaries; it does not convert RC1 into Technical Freeze or certification.
What did we actually create?
A governed boundary that stops information, intelligence, interoperability, or old authorization from silently becoming permission to cause a protected consequence.
A system may understand a request perfectly. EABA still asks whether protected consequence is permitted now.
Meaning can cross the boundary. Permission cannot be assumed across it.
Ten questions before consequence.
Click any burden. Each answers a different question. None may silently stand in for another.
Do we know what is true now?
MAPS TOWARD P-EVIDThe evidence must describe the current state closely enough for this consequence. Old truth cannot silently become present truth.
Six gateway predicates. Still not the whole architecture.
EBG makes the commit seam executable. EABA remains the parent architecture: all applicable proof obligations must still be satisfied.
Is present authority established for this actor, action and scope?
Do not read the architecture. Operate it.
Select a consequence domain. The same execution-authority question is then applied to a different protected world. EABA does not grant permission because a system is smart, connected, confident, or technically capable.
Change one fact. Watch permission disappear.
Start with a fully established route. Toggle any condition. EABA recomputes the governed determination without silently repairing the failure.
All modeled conditions are established. This attempt is eligible to proceed to protected commit; effect and outcome evidence are still required.
Every crossing attempt leaves a determination trace.
Change conditions and attempt the commit again. A previous ALLOW does not silently survive a changed present.
Attack the boundary. See exactly what breaks.
The evidence may once have been true, but it is outside the permitted freshness bound. The system cannot convert historical truth into present standing.
ALLOW tells you what was permitted. It does not tell you what happened.
This is one of the most important EABA distinctions. Determination truth and effect truth are separate records.
The implementation does not get to grade itself.
RC1 is published architecture, not a claim that every implementation is proven. Independent implementation examination remains a separate gate.
The missing layer is not intelligence. It is admissible permission at consequence.
Interoperability can carry meaning and context. It does not automatically transfer execution authority.
Prior approval can become stale, revoked, out of scope, or detached from the exact commit attempt.
Proof, authority, binding, refusal, commit integrity, effect evidence, and closure meet at one consequence boundary.
SPECIFY → ESTABLISH → BIND → GATE → DETERMINE → COMMIT → EFFECT → PROVE → CLOSE
EABA does not ask intelligent systems to know less. It requires consequential systems to establish more before acting—and to leave behind enough evidence to prove what actually happened.