A new public layer is now part of the corpus.
Not a side note. Not a reaction. A structured package.
Its role is simple:
before a long-lived system can review a dispute, before it can rely on a signal, before it can let a path continue into action, it has to answer a more basic question:
what exactly is sufficiently real, grounded, and present enough to be trusted at all?
That is the role of AGL — Actor Grounding Layer.
Most architectures enter too late. They begin with: review, admissibility, arbitration, or evidence.
But in real systems, many failures start earlier.
Not at the level of reasoning, but at the level of grounding.
A signal may be stale. A perceptual path may be degraded. An initiating source may be delegated, simulated, or contextually detached from the live execution boundary. A system may still look coherent, while already operating one layer too far from reality.
That is why I do not treat grounding as a soft advisory function.
I treat it as a precondition.
If grounding is weak, the path should not proceed as normal and hope later review will fix it.
It should narrow. Pause. Fail closed. Or stop.
Earth paragraph:
A long-lived system does not become dangerous only when it reasons badly. It becomes dangerous when degraded grounding is still allowed to count as actionable reality. That may begin with fatigue, sensory drift, proxy-origin abstraction, or present-state decoupling. From that point on, the problem is no longer interpretation. It is that action is being built on a source that should not yet have earned runtime reliance. A serious architecture must be able to fail closed there, before coherence in language turns into incoherence in the world. That is what this layer is for.
GitHub:
DOI: